<?xml version="1.0" encoding="UTF-8"?>
<rss version="2.0" xmlns:atom="http://www.w3.org/2005/Atom" xmlns:dc="http://purl.org/dc/elements/1.1/">
  <channel>
    <title>DEV Community: Martin Pisacreta</title>
    <description>The latest articles on DEV Community by Martin Pisacreta (@martinpisacreta).</description>
    <link>https://dev.to/martinpisacreta</link>
    <image>
      <url>https://media2.dev.to/dynamic/image/width=90,height=90,fit=cover,gravity=auto,format=auto/https:%2F%2Fdev-to-uploads.s3.us-east-2.amazonaws.com%2Fuploads%2Fuser%2Fprofile_image%2F4195178%2F462e0bce-5293-481a-9f56-31fbf8c0c365.jpg</url>
      <title>DEV Community: Martin Pisacreta</title>
      <link>https://dev.to/martinpisacreta</link>
    </image>
    <atom:link rel="self" type="application/rss+xml" href="https://dev.to/feed/martinpisacreta"/>
    <language>en</language>
    <item>
      <title>Reading Workday and Rippling job boards without a scraper (and the pagination trap)</title>
      <dc:creator>Martin Pisacreta</dc:creator>
      <pubDate>Sun, 15 Nov 2026 23:57:11 +0000</pubDate>
      <link>https://dev.to/martinpisacreta/reading-workday-and-rippling-job-boards-without-a-scraper-and-the-pagination-trap-3375</link>
      <guid>https://dev.to/martinpisacreta/reading-workday-and-rippling-job-boards-without-a-scraper-and-the-pagination-trap-3375</guid>
      <description>&lt;p&gt;I added Workday and Rippling to my jobs monitor by reading their job-list responses directly. The useful discovery was not just the endpoints. It was a pagination detail that could make a working Workday integration stop early while reporting success. My saved second page contained twenty jobs and a total of zero. Treating that zero as the end of the board would have discarded valid results.&lt;/p&gt;

&lt;p&gt;Everything below is what I observed on real boards while building that integration. Neither endpoint is a documented public API, so treat this as a starting point: it can change, and another employer's board may behave differently.&lt;/p&gt;

&lt;h2&gt;
  
  
  Workday: start with the public board URL
&lt;/h2&gt;

&lt;p&gt;My Workday example starts with NVIDIA's public board address. Its hostname supplies two pieces: &lt;code&gt;nvidia&lt;/code&gt; is the tenant and &lt;code&gt;wd5&lt;/code&gt; is the cluster. The path supplies the site, &lt;code&gt;NVIDIAExternalCareerSite&lt;/code&gt;. A locale such as &lt;code&gt;en-US&lt;/code&gt; can appear before the site; I remove that segment rather than accidentally treating the locale as the board identifier.&lt;/p&gt;

&lt;p&gt;Those pieces produce a different path on the same hostname: &lt;code&gt;/wday/cxs/nvidia/NVIDIAExternalCareerSite/jobs&lt;/code&gt;. My implementation sends a POST with a JSON body containing &lt;code&gt;appliedFacets&lt;/code&gt;, &lt;code&gt;limit&lt;/code&gt;, &lt;code&gt;offset&lt;/code&gt;, and &lt;code&gt;searchText&lt;/code&gt;. In the example, the facets object and search text are empty. I request twenty records at a time, starting at offset zero.&lt;/p&gt;

&lt;p&gt;I keep the tenant, cluster, and site separate because a single company name is insufficient to construct this address. The example checks the hostname and expects a board URL, rather than accepting an arbitrary job-detail path. The production integration also accepts a compact &lt;code&gt;tenant/wdN/site&lt;/code&gt; identifier and normalizes it to the same company identity.&lt;/p&gt;

&lt;h2&gt;
  
  
  The pagination trap
&lt;/h2&gt;

&lt;p&gt;My saved first response reports a total of 2,000 and contains twenty postings. The second response, requested at offset twenty, also contains twenty postings, but its total is zero. Together, the fixtures contain forty distinct job paths.&lt;/p&gt;

&lt;p&gt;I preserve the first response's total instead of replacing it on every page. I advance the offset by the number of records returned. I also stop on a short page, a page that adds no new job paths, or the initial reported total when it is positive. A separate page cap keeps the example bounded.&lt;/p&gt;

&lt;p&gt;Twenty is the page size used by this integration and exercised by these fixtures. I keep that limit explicit rather than assuming a larger request will return more records. The two-page example deliberately stops after forty observed postings; it does not download all 2,000 reported jobs. Its &lt;code&gt;max-pages&lt;/code&gt; stop reason makes that limitation visible.&lt;/p&gt;

&lt;p&gt;Here is the standalone example I tested. It uses built-in &lt;code&gt;fetch&lt;/code&gt;, so I run it with Node.js 22 as an &lt;code&gt;.mjs&lt;/code&gt; file. Running it normally makes requests; my verification substitutes fixture responses instead.&lt;br&gt;
&lt;/p&gt;

&lt;div class="highlight js-code-highlight"&gt;
&lt;pre class="highlight javascript"&gt;&lt;code&gt;&lt;span class="kd"&gt;const&lt;/span&gt; &lt;span class="n"&gt;board&lt;/span&gt; &lt;span class="o"&gt;=&lt;/span&gt; &lt;span class="k"&gt;new&lt;/span&gt; &lt;span class="no"&gt;URL&lt;/span&gt;&lt;span class="p"&gt;(&lt;/span&gt;&lt;span class="s1"&gt;'https://nvidia.wd5.myworkdayjobs.com/en-US/NVIDIAExternalCareerSite'&lt;/span&gt;&lt;span class="p"&gt;)&lt;/span&gt;&lt;span class="p"&gt;;&lt;/span&gt;
&lt;span class="kd"&gt;const&lt;/span&gt; &lt;span class="n"&gt;match&lt;/span&gt; &lt;span class="o"&gt;=&lt;/span&gt; &lt;span class="o"&gt;/^&lt;/span&gt;&lt;span class="p"&gt;(&lt;/span&gt;&lt;span class="p"&gt;[&lt;/span&gt;&lt;span class="n"&gt;a&lt;/span&gt;&lt;span class="o"&gt;-&lt;/span&gt;&lt;span class="n"&gt;z0&lt;/span&gt;&lt;span class="o"&gt;-&lt;/span&gt;&lt;span class="mi"&gt;9&lt;/span&gt;&lt;span class="o"&gt;-&lt;/span&gt;&lt;span class="p"&gt;]&lt;/span&gt;&lt;span class="o"&gt;+&lt;/span&gt;&lt;span class="p"&gt;)&lt;/span&gt;\&lt;span class="p"&gt;.&lt;/span&gt;&lt;span class="p"&gt;(&lt;/span&gt;&lt;span class="n"&gt;wd&lt;/span&gt;&lt;span class="p"&gt;[&lt;/span&gt;&lt;span class="mi"&gt;0&lt;/span&gt;&lt;span class="o"&gt;-&lt;/span&gt;&lt;span class="mi"&gt;9&lt;/span&gt;&lt;span class="p"&gt;]&lt;/span&gt;&lt;span class="o"&gt;+&lt;/span&gt;&lt;span class="p"&gt;)&lt;/span&gt;\&lt;span class="p"&gt;.&lt;/span&gt;&lt;span class="n"&gt;myworkdayjobs&lt;/span&gt;\&lt;span class="p"&gt;.&lt;/span&gt;&lt;span class="n"&gt;com&lt;/span&gt;$&lt;span class="o"&gt;/&lt;/span&gt;&lt;span class="p"&gt;.&lt;/span&gt;&lt;span class="nf"&gt;exec&lt;/span&gt;&lt;span class="p"&gt;(&lt;/span&gt;&lt;span class="n"&gt;board&lt;/span&gt;&lt;span class="p"&gt;.&lt;/span&gt;&lt;span class="n"&gt;hostname&lt;/span&gt;&lt;span class="p"&gt;)&lt;/span&gt;&lt;span class="p"&gt;;&lt;/span&gt;
&lt;span class="k"&gt;if&lt;/span&gt; &lt;span class="p"&gt;(&lt;/span&gt;&lt;span class="o"&gt;!&lt;/span&gt;&lt;span class="n"&gt;match&lt;/span&gt;&lt;span class="p"&gt;)&lt;/span&gt; &lt;span class="k"&gt;throw&lt;/span&gt; &lt;span class="k"&gt;new&lt;/span&gt; &lt;span class="nb"&gt;Error&lt;/span&gt;&lt;span class="p"&gt;(&lt;/span&gt;&lt;span class="s1"&gt;'Unsupported Workday hostname'&lt;/span&gt;&lt;span class="p"&gt;)&lt;/span&gt;&lt;span class="p"&gt;;&lt;/span&gt;
&lt;span class="kd"&gt;const&lt;/span&gt; &lt;span class="p"&gt;[&lt;/span&gt;&lt;span class="p"&gt;,&lt;/span&gt; &lt;span class="n"&gt;tenant&lt;/span&gt;&lt;span class="p"&gt;,&lt;/span&gt; &lt;span class="n"&gt;cluster&lt;/span&gt;&lt;span class="p"&gt;]&lt;/span&gt; &lt;span class="o"&gt;=&lt;/span&gt; &lt;span class="n"&gt;match&lt;/span&gt;&lt;span class="p"&gt;;&lt;/span&gt;
&lt;span class="kd"&gt;const&lt;/span&gt; &lt;span class="n"&gt;parts&lt;/span&gt; &lt;span class="o"&gt;=&lt;/span&gt; &lt;span class="n"&gt;board&lt;/span&gt;&lt;span class="p"&gt;.&lt;/span&gt;&lt;span class="n"&gt;pathname&lt;/span&gt;&lt;span class="p"&gt;.&lt;/span&gt;&lt;span class="nf"&gt;split&lt;/span&gt;&lt;span class="p"&gt;(&lt;/span&gt;&lt;span class="s1"&gt;'/'&lt;/span&gt;&lt;span class="p"&gt;)&lt;/span&gt;&lt;span class="p"&gt;.&lt;/span&gt;&lt;span class="nf"&gt;filter&lt;/span&gt;&lt;span class="p"&gt;(&lt;/span&gt;&lt;span class="nb"&gt;Boolean&lt;/span&gt;&lt;span class="p"&gt;)&lt;/span&gt;&lt;span class="p"&gt;;&lt;/span&gt;
&lt;span class="k"&gt;if&lt;/span&gt; &lt;span class="p"&gt;(&lt;/span&gt;&lt;span class="o"&gt;/^&lt;/span&gt;&lt;span class="p"&gt;[&lt;/span&gt;&lt;span class="n"&gt;a&lt;/span&gt;&lt;span class="o"&gt;-&lt;/span&gt;&lt;span class="n"&gt;z&lt;/span&gt;&lt;span class="p"&gt;]&lt;/span&gt;&lt;span class="p"&gt;{&lt;/span&gt;&lt;span class="mi"&gt;2&lt;/span&gt;&lt;span class="p"&gt;}&lt;/span&gt;&lt;span class="o"&gt;-&lt;/span&gt;&lt;span class="p"&gt;[&lt;/span&gt;&lt;span class="n"&gt;a&lt;/span&gt;&lt;span class="o"&gt;-&lt;/span&gt;&lt;span class="n"&gt;z&lt;/span&gt;&lt;span class="p"&gt;]&lt;/span&gt;&lt;span class="p"&gt;{&lt;/span&gt;&lt;span class="mi"&gt;2&lt;/span&gt;&lt;span class="p"&gt;}&lt;/span&gt;$&lt;span class="o"&gt;/&lt;/span&gt;&lt;span class="n"&gt;i&lt;/span&gt;&lt;span class="p"&gt;.&lt;/span&gt;&lt;span class="nf"&gt;test&lt;/span&gt;&lt;span class="p"&gt;(&lt;/span&gt;&lt;span class="n"&gt;parts&lt;/span&gt;&lt;span class="p"&gt;[&lt;/span&gt;&lt;span class="mi"&gt;0&lt;/span&gt;&lt;span class="p"&gt;]&lt;/span&gt;&lt;span class="p"&gt;)&lt;/span&gt;&lt;span class="p"&gt;)&lt;/span&gt; &lt;span class="n"&gt;parts&lt;/span&gt;&lt;span class="p"&gt;.&lt;/span&gt;&lt;span class="nf"&gt;shift&lt;/span&gt;&lt;span class="p"&gt;(&lt;/span&gt;&lt;span class="p"&gt;)&lt;/span&gt;&lt;span class="p"&gt;;&lt;/span&gt;
&lt;span class="k"&gt;if&lt;/span&gt; &lt;span class="p"&gt;(&lt;/span&gt;&lt;span class="n"&gt;parts&lt;/span&gt;&lt;span class="p"&gt;.&lt;/span&gt;&lt;span class="n"&gt;length&lt;/span&gt; &lt;span class="o"&gt;!==&lt;/span&gt; &lt;span class="mi"&gt;1&lt;/span&gt;&lt;span class="p"&gt;)&lt;/span&gt; &lt;span class="k"&gt;throw&lt;/span&gt; &lt;span class="k"&gt;new&lt;/span&gt; &lt;span class="nb"&gt;Error&lt;/span&gt;&lt;span class="p"&gt;(&lt;/span&gt;&lt;span class="s1"&gt;'Expected a board URL'&lt;/span&gt;&lt;span class="p"&gt;)&lt;/span&gt;&lt;span class="p"&gt;;&lt;/span&gt;
&lt;span class="kd"&gt;const&lt;/span&gt; &lt;span class="n"&gt;site&lt;/span&gt; &lt;span class="o"&gt;=&lt;/span&gt; &lt;span class="n"&gt;parts&lt;/span&gt;&lt;span class="p"&gt;[&lt;/span&gt;&lt;span class="mi"&gt;0&lt;/span&gt;&lt;span class="p"&gt;]&lt;/span&gt;&lt;span class="p"&gt;;&lt;/span&gt;
&lt;span class="kd"&gt;const&lt;/span&gt; &lt;span class="n"&gt;endpoint&lt;/span&gt; &lt;span class="o"&gt;=&lt;/span&gt; &lt;span class="sb"&gt;`https://${tenant}.${cluster}.myworkdayjobs.com/wday/cxs/${tenant}/${encodeURIComponent(site)}/jobs`&lt;/span&gt;&lt;span class="p"&gt;;&lt;/span&gt;
&lt;span class="kd"&gt;const&lt;/span&gt; &lt;span class="n"&gt;jobs&lt;/span&gt; &lt;span class="o"&gt;=&lt;/span&gt; &lt;span class="p"&gt;[&lt;/span&gt;&lt;span class="p"&gt;]&lt;/span&gt;&lt;span class="p"&gt;;&lt;/span&gt;
&lt;span class="kd"&gt;const&lt;/span&gt; &lt;span class="n"&gt;seen&lt;/span&gt; &lt;span class="o"&gt;=&lt;/span&gt; &lt;span class="k"&gt;new&lt;/span&gt; &lt;span class="nb"&gt;Set&lt;/span&gt;&lt;span class="p"&gt;(&lt;/span&gt;&lt;span class="p"&gt;)&lt;/span&gt;&lt;span class="p"&gt;;&lt;/span&gt;
&lt;span class="kd"&gt;const&lt;/span&gt; &lt;span class="n"&gt;maxPages&lt;/span&gt; &lt;span class="o"&gt;=&lt;/span&gt; &lt;span class="mi"&gt;2&lt;/span&gt;&lt;span class="p"&gt;;&lt;/span&gt; &lt;span class="c1"&gt;// Deliberate request cap, not a complete-board guarantee.&lt;/span&gt;
&lt;span class="kd"&gt;let&lt;/span&gt; &lt;span class="n"&gt;offset&lt;/span&gt; &lt;span class="o"&gt;=&lt;/span&gt; &lt;span class="mi"&gt;0&lt;/span&gt;&lt;span class="p"&gt;,&lt;/span&gt; &lt;span class="n"&gt;totalReported&lt;/span&gt; &lt;span class="o"&gt;=&lt;/span&gt; &lt;span class="kc"&gt;null&lt;/span&gt;&lt;span class="p"&gt;,&lt;/span&gt; &lt;span class="n"&gt;pagesFetched&lt;/span&gt; &lt;span class="o"&gt;=&lt;/span&gt; &lt;span class="mi"&gt;0&lt;/span&gt;&lt;span class="p"&gt;;&lt;/span&gt;
&lt;span class="kd"&gt;let&lt;/span&gt; &lt;span class="n"&gt;stopReason&lt;/span&gt; &lt;span class="o"&gt;=&lt;/span&gt; &lt;span class="s1"&gt;'max-pages'&lt;/span&gt;&lt;span class="p"&gt;;&lt;/span&gt;
&lt;span class="k"&gt;for&lt;/span&gt; &lt;span class="p"&gt;(&lt;/span&gt;&lt;span class="kd"&gt;let&lt;/span&gt; &lt;span class="n"&gt;page&lt;/span&gt; &lt;span class="o"&gt;=&lt;/span&gt; &lt;span class="mi"&gt;0&lt;/span&gt;&lt;span class="p"&gt;;&lt;/span&gt; &lt;span class="n"&gt;page&lt;/span&gt; &lt;span class="o"&gt;&amp;lt;&lt;/span&gt; &lt;span class="n"&gt;maxPages&lt;/span&gt;&lt;span class="p"&gt;;&lt;/span&gt; &lt;span class="n"&gt;page&lt;/span&gt;&lt;span class="o"&gt;++&lt;/span&gt;&lt;span class="p"&gt;)&lt;/span&gt; &lt;span class="p"&gt;{&lt;/span&gt;
  &lt;span class="kd"&gt;const&lt;/span&gt; &lt;span class="n"&gt;response&lt;/span&gt; &lt;span class="o"&gt;=&lt;/span&gt; &lt;span class="k"&gt;await&lt;/span&gt; &lt;span class="nb"&gt;fetch&lt;/span&gt;&lt;span class="p"&gt;(&lt;/span&gt;&lt;span class="n"&gt;endpoint&lt;/span&gt;&lt;span class="p"&gt;,&lt;/span&gt; &lt;span class="p"&gt;{&lt;/span&gt;
    &lt;span class="n"&gt;method&lt;/span&gt;&lt;span class="p"&gt;:&lt;/span&gt; &lt;span class="s1"&gt;'POST'&lt;/span&gt;&lt;span class="p"&gt;,&lt;/span&gt;
    &lt;span class="n"&gt;headers&lt;/span&gt;&lt;span class="p"&gt;:&lt;/span&gt; &lt;span class="p"&gt;{&lt;/span&gt; &lt;span class="s1"&gt;'Content-Type'&lt;/span&gt;&lt;span class="p"&gt;:&lt;/span&gt; &lt;span class="s1"&gt;'application/json'&lt;/span&gt; &lt;span class="p"&gt;}&lt;/span&gt;&lt;span class="p"&gt;,&lt;/span&gt;
    &lt;span class="n"&gt;body&lt;/span&gt;&lt;span class="p"&gt;:&lt;/span&gt; &lt;span class="nb"&gt;JSON&lt;/span&gt;&lt;span class="p"&gt;.&lt;/span&gt;&lt;span class="nf"&gt;stringify&lt;/span&gt;&lt;span class="p"&gt;(&lt;/span&gt;&lt;span class="p"&gt;{&lt;/span&gt; &lt;span class="n"&gt;appliedFacets&lt;/span&gt;&lt;span class="p"&gt;:&lt;/span&gt; &lt;span class="p"&gt;{&lt;/span&gt;&lt;span class="p"&gt;}&lt;/span&gt;&lt;span class="p"&gt;,&lt;/span&gt; &lt;span class="n"&gt;limit&lt;/span&gt;&lt;span class="p"&gt;:&lt;/span&gt; &lt;span class="mi"&gt;20&lt;/span&gt;&lt;span class="p"&gt;,&lt;/span&gt; &lt;span class="n"&gt;offset&lt;/span&gt;&lt;span class="p"&gt;,&lt;/span&gt; &lt;span class="n"&gt;searchText&lt;/span&gt;&lt;span class="p"&gt;:&lt;/span&gt; &lt;span class="s1"&gt;''&lt;/span&gt; &lt;span class="p"&gt;}&lt;/span&gt;&lt;span class="p"&gt;)&lt;/span&gt;&lt;span class="p"&gt;,&lt;/span&gt;
  &lt;span class="p"&gt;}&lt;/span&gt;&lt;span class="p"&gt;)&lt;/span&gt;&lt;span class="p"&gt;;&lt;/span&gt;
  &lt;span class="k"&gt;if&lt;/span&gt; &lt;span class="p"&gt;(&lt;/span&gt;&lt;span class="o"&gt;!&lt;/span&gt;&lt;span class="n"&gt;response&lt;/span&gt;&lt;span class="p"&gt;.&lt;/span&gt;&lt;span class="n"&gt;ok&lt;/span&gt;&lt;span class="p"&gt;)&lt;/span&gt; &lt;span class="k"&gt;throw&lt;/span&gt; &lt;span class="k"&gt;new&lt;/span&gt; &lt;span class="nb"&gt;Error&lt;/span&gt;&lt;span class="p"&gt;(&lt;/span&gt;&lt;span class="sb"&gt;`Workday HTTP ${response.status}`&lt;/span&gt;&lt;span class="p"&gt;)&lt;/span&gt;&lt;span class="p"&gt;;&lt;/span&gt;
  &lt;span class="kd"&gt;const&lt;/span&gt; &lt;span class="n"&gt;data&lt;/span&gt; &lt;span class="o"&gt;=&lt;/span&gt; &lt;span class="k"&gt;await&lt;/span&gt; &lt;span class="n"&gt;response&lt;/span&gt;&lt;span class="p"&gt;.&lt;/span&gt;&lt;span class="nf"&gt;json&lt;/span&gt;&lt;span class="p"&gt;(&lt;/span&gt;&lt;span class="p"&gt;)&lt;/span&gt;&lt;span class="p"&gt;;&lt;/span&gt;
  &lt;span class="k"&gt;if&lt;/span&gt; &lt;span class="p"&gt;(&lt;/span&gt;&lt;span class="o"&gt;!&lt;/span&gt;&lt;span class="nb"&gt;Array&lt;/span&gt;&lt;span class="p"&gt;.&lt;/span&gt;&lt;span class="nf"&gt;isArray&lt;/span&gt;&lt;span class="p"&gt;(&lt;/span&gt;&lt;span class="n"&gt;data&lt;/span&gt;&lt;span class="p"&gt;.&lt;/span&gt;&lt;span class="n"&gt;jobPostings&lt;/span&gt;&lt;span class="p"&gt;)&lt;/span&gt; &lt;span class="o"&gt;||&lt;/span&gt; &lt;span class="o"&gt;!&lt;/span&gt;&lt;span class="nb"&gt;Number&lt;/span&gt;&lt;span class="p"&gt;.&lt;/span&gt;&lt;span class="nf"&gt;isInteger&lt;/span&gt;&lt;span class="p"&gt;(&lt;/span&gt;&lt;span class="n"&gt;data&lt;/span&gt;&lt;span class="p"&gt;.&lt;/span&gt;&lt;span class="n"&gt;total&lt;/span&gt;&lt;span class="p"&gt;)&lt;/span&gt; &lt;span class="o"&gt;||&lt;/span&gt; &lt;span class="n"&gt;data&lt;/span&gt;&lt;span class="p"&gt;.&lt;/span&gt;&lt;span class="n"&gt;total&lt;/span&gt; &lt;span class="o"&gt;&amp;lt;&lt;/span&gt; &lt;span class="mi"&gt;0&lt;/span&gt; &lt;span class="o"&gt;||&lt;/span&gt; &lt;span class="n"&gt;data&lt;/span&gt;&lt;span class="p"&gt;.&lt;/span&gt;&lt;span class="n"&gt;jobPostings&lt;/span&gt;&lt;span class="p"&gt;.&lt;/span&gt;&lt;span class="n"&gt;length&lt;/span&gt; &lt;span class="o"&gt;&amp;gt;&lt;/span&gt; &lt;span class="mi"&gt;20&lt;/span&gt;&lt;span class="p"&gt;)&lt;/span&gt; &lt;span class="k"&gt;throw&lt;/span&gt; &lt;span class="k"&gt;new&lt;/span&gt; &lt;span class="nb"&gt;Error&lt;/span&gt;&lt;span class="p"&gt;(&lt;/span&gt;&lt;span class="s1"&gt;'Unexpected Workday response'&lt;/span&gt;&lt;span class="p"&gt;)&lt;/span&gt;&lt;span class="p"&gt;;&lt;/span&gt;
  &lt;span class="k"&gt;if&lt;/span&gt; &lt;span class="p"&gt;(&lt;/span&gt;&lt;span class="n"&gt;totalReported&lt;/span&gt; &lt;span class="o"&gt;===&lt;/span&gt; &lt;span class="kc"&gt;null&lt;/span&gt;&lt;span class="p"&gt;)&lt;/span&gt; &lt;span class="n"&gt;totalReported&lt;/span&gt; &lt;span class="o"&gt;=&lt;/span&gt; &lt;span class="n"&gt;data&lt;/span&gt;&lt;span class="p"&gt;.&lt;/span&gt;&lt;span class="n"&gt;total&lt;/span&gt;&lt;span class="p"&gt;;&lt;/span&gt;
  &lt;span class="n"&gt;pagesFetched&lt;/span&gt;&lt;span class="o"&gt;++&lt;/span&gt;&lt;span class="p"&gt;;&lt;/span&gt;
  &lt;span class="kd"&gt;let&lt;/span&gt; &lt;span class="n"&gt;added&lt;/span&gt; &lt;span class="o"&gt;=&lt;/span&gt; &lt;span class="mi"&gt;0&lt;/span&gt;&lt;span class="p"&gt;;&lt;/span&gt;
  &lt;span class="k"&gt;for&lt;/span&gt; &lt;span class="p"&gt;(&lt;/span&gt;&lt;span class="kd"&gt;const&lt;/span&gt; &lt;span class="n"&gt;job&lt;/span&gt; &lt;span class="k"&gt;of&lt;/span&gt; &lt;span class="n"&gt;data&lt;/span&gt;&lt;span class="p"&gt;.&lt;/span&gt;&lt;span class="n"&gt;jobPostings&lt;/span&gt;&lt;span class="p"&gt;)&lt;/span&gt; &lt;span class="p"&gt;{&lt;/span&gt;
    &lt;span class="k"&gt;if&lt;/span&gt; &lt;span class="p"&gt;(&lt;/span&gt;&lt;span class="o"&gt;!&lt;/span&gt;&lt;span class="n"&gt;seen&lt;/span&gt;&lt;span class="p"&gt;.&lt;/span&gt;&lt;span class="nf"&gt;has&lt;/span&gt;&lt;span class="p"&gt;(&lt;/span&gt;&lt;span class="n"&gt;job&lt;/span&gt;&lt;span class="p"&gt;.&lt;/span&gt;&lt;span class="n"&gt;externalPath&lt;/span&gt;&lt;span class="p"&gt;)&lt;/span&gt;&lt;span class="p"&gt;)&lt;/span&gt; &lt;span class="p"&gt;{&lt;/span&gt; &lt;span class="n"&gt;seen&lt;/span&gt;&lt;span class="p"&gt;.&lt;/span&gt;&lt;span class="nf"&gt;add&lt;/span&gt;&lt;span class="p"&gt;(&lt;/span&gt;&lt;span class="n"&gt;job&lt;/span&gt;&lt;span class="p"&gt;.&lt;/span&gt;&lt;span class="n"&gt;externalPath&lt;/span&gt;&lt;span class="p"&gt;)&lt;/span&gt;&lt;span class="p"&gt;;&lt;/span&gt; &lt;span class="n"&gt;jobs&lt;/span&gt;&lt;span class="p"&gt;.&lt;/span&gt;&lt;span class="nf"&gt;push&lt;/span&gt;&lt;span class="p"&gt;(&lt;/span&gt;&lt;span class="n"&gt;job&lt;/span&gt;&lt;span class="p"&gt;)&lt;/span&gt;&lt;span class="p"&gt;;&lt;/span&gt; &lt;span class="n"&gt;added&lt;/span&gt;&lt;span class="o"&gt;++&lt;/span&gt;&lt;span class="p"&gt;;&lt;/span&gt; &lt;span class="p"&gt;}&lt;/span&gt;
  &lt;span class="p"&gt;}&lt;/span&gt;
  &lt;span class="n"&gt;offset&lt;/span&gt; &lt;span class="o"&gt;+=&lt;/span&gt; &lt;span class="n"&gt;data&lt;/span&gt;&lt;span class="p"&gt;.&lt;/span&gt;&lt;span class="n"&gt;jobPostings&lt;/span&gt;&lt;span class="p"&gt;.&lt;/span&gt;&lt;span class="n"&gt;length&lt;/span&gt;&lt;span class="p"&gt;;&lt;/span&gt;
  &lt;span class="k"&gt;if&lt;/span&gt; &lt;span class="p"&gt;(&lt;/span&gt;&lt;span class="n"&gt;data&lt;/span&gt;&lt;span class="p"&gt;.&lt;/span&gt;&lt;span class="n"&gt;jobPostings&lt;/span&gt;&lt;span class="p"&gt;.&lt;/span&gt;&lt;span class="n"&gt;length&lt;/span&gt; &lt;span class="o"&gt;&amp;lt;&lt;/span&gt; &lt;span class="mi"&gt;20&lt;/span&gt;&lt;span class="p"&gt;)&lt;/span&gt; &lt;span class="p"&gt;{&lt;/span&gt; &lt;span class="n"&gt;stopReason&lt;/span&gt; &lt;span class="o"&gt;=&lt;/span&gt; &lt;span class="s1"&gt;'short-page'&lt;/span&gt;&lt;span class="p"&gt;;&lt;/span&gt; &lt;span class="k"&gt;break&lt;/span&gt;&lt;span class="p"&gt;;&lt;/span&gt; &lt;span class="p"&gt;}&lt;/span&gt;
  &lt;span class="k"&gt;if&lt;/span&gt; &lt;span class="p"&gt;(&lt;/span&gt;&lt;span class="o"&gt;!&lt;/span&gt;&lt;span class="n"&gt;added&lt;/span&gt;&lt;span class="p"&gt;)&lt;/span&gt; &lt;span class="p"&gt;{&lt;/span&gt; &lt;span class="n"&gt;stopReason&lt;/span&gt; &lt;span class="o"&gt;=&lt;/span&gt; &lt;span class="s1"&gt;'repeated-page'&lt;/span&gt;&lt;span class="p"&gt;;&lt;/span&gt; &lt;span class="k"&gt;break&lt;/span&gt;&lt;span class="p"&gt;;&lt;/span&gt; &lt;span class="p"&gt;}&lt;/span&gt;
  &lt;span class="k"&gt;if&lt;/span&gt; &lt;span class="p"&gt;(&lt;/span&gt;&lt;span class="n"&gt;totalReported&lt;/span&gt; &lt;span class="o"&gt;&amp;gt;&lt;/span&gt; &lt;span class="mi"&gt;0&lt;/span&gt; &lt;span class="o"&gt;&amp;amp;&amp;amp;&lt;/span&gt; &lt;span class="n"&gt;jobs&lt;/span&gt;&lt;span class="p"&gt;.&lt;/span&gt;&lt;span class="n"&gt;length&lt;/span&gt; &lt;span class="o"&gt;&amp;gt;=&lt;/span&gt; &lt;span class="n"&gt;totalReported&lt;/span&gt;&lt;span class="p"&gt;)&lt;/span&gt; &lt;span class="p"&gt;{&lt;/span&gt; &lt;span class="n"&gt;stopReason&lt;/span&gt; &lt;span class="o"&gt;=&lt;/span&gt; &lt;span class="s1"&gt;'reported-total'&lt;/span&gt;&lt;span class="p"&gt;;&lt;/span&gt; &lt;span class="k"&gt;break&lt;/span&gt;&lt;span class="p"&gt;;&lt;/span&gt; &lt;span class="p"&gt;}&lt;/span&gt;
&lt;span class="p"&gt;}&lt;/span&gt;
&lt;span class="nb"&gt;console&lt;/span&gt;&lt;span class="p"&gt;.&lt;/span&gt;&lt;span class="nf"&gt;log&lt;/span&gt;&lt;span class="p"&gt;(&lt;/span&gt;&lt;span class="nb"&gt;JSON&lt;/span&gt;&lt;span class="p"&gt;.&lt;/span&gt;&lt;span class="nf"&gt;stringify&lt;/span&gt;&lt;span class="p"&gt;(&lt;/span&gt;&lt;span class="p"&gt;{&lt;/span&gt; &lt;span class="n"&gt;jobs&lt;/span&gt;&lt;span class="p"&gt;,&lt;/span&gt; &lt;span class="n"&gt;totalReported&lt;/span&gt;&lt;span class="p"&gt;,&lt;/span&gt; &lt;span class="n"&gt;pagesFetched&lt;/span&gt;&lt;span class="p"&gt;,&lt;/span&gt; &lt;span class="n"&gt;stopReason&lt;/span&gt; &lt;span class="p"&gt;}&lt;/span&gt;&lt;span class="p"&gt;)&lt;/span&gt;&lt;span class="p"&gt;)&lt;/span&gt;&lt;span class="p"&gt;;&lt;/span&gt;
&lt;/code&gt;&lt;/pre&gt;

&lt;/div&gt;



&lt;p&gt;One habit that paid off: keep the records and the collection status separate. The forty paths are what I collected, the preserved total is what the first response claimed, and the stop reason says why I stopped. Hitting my own page cap is not evidence that the board has no more jobs.&lt;/p&gt;

&lt;h2&gt;
  
  
  Dates I could not safely normalize
&lt;/h2&gt;

&lt;p&gt;The Workday fixtures include &lt;code&gt;postedOn&lt;/code&gt; labels such as “Posted Today” and “Posted Yesterday.” I did not find an absolute publication timestamp in those saved posting objects. Converting those labels into precise timestamps would require assumptions about capture time and timezone that the labels themselves do not supply.&lt;/p&gt;

&lt;p&gt;I therefore leave &lt;code&gt;publishedAt&lt;/code&gt; empty in my normalized records. I can still retain the job's title, location, and public path. For monitoring, I use that path as the identifier within the company board. Missing precision is preferable to a date that appears authoritative but was manufactured during normalization.&lt;/p&gt;

&lt;h2&gt;
  
  
  Rippling: one GET and an array
&lt;/h2&gt;

&lt;p&gt;Rippling's saved response is simpler to traverse. My adapter requests &lt;code&gt;/platform/api/ats/v1/board/rippling/jobs&lt;/code&gt; on &lt;code&gt;api.rippling.com&lt;/code&gt; with GET and receives a top-level array. The local fixture contains 717 records. I iterate that array directly; there is no Workday-style offset loop in this example.&lt;/p&gt;

&lt;p&gt;The fields I map are &lt;code&gt;uuid&lt;/code&gt;, &lt;code&gt;name&lt;/code&gt;, &lt;code&gt;workLocation.label&lt;/code&gt;, &lt;code&gt;department.label&lt;/code&gt;, and &lt;code&gt;url&lt;/code&gt;. I keep missing optional labels as null. This describes the response I saved, not a promise that every Rippling board always returns a complete collection in one response. I validate the array shape before mapping it.&lt;br&gt;
&lt;/p&gt;

&lt;div class="highlight js-code-highlight"&gt;
&lt;pre class="highlight javascript"&gt;&lt;code&gt;&lt;span class="kd"&gt;const&lt;/span&gt; &lt;span class="n"&gt;company&lt;/span&gt; &lt;span class="o"&gt;=&lt;/span&gt; &lt;span class="s1"&gt;'rippling'&lt;/span&gt;&lt;span class="p"&gt;;&lt;/span&gt;
&lt;span class="kd"&gt;const&lt;/span&gt; &lt;span class="n"&gt;endpoint&lt;/span&gt; &lt;span class="o"&gt;=&lt;/span&gt; &lt;span class="sb"&gt;`https://api.rippling.com/platform/api/ats/v1/board/${encodeURIComponent(company)}/jobs`&lt;/span&gt;&lt;span class="p"&gt;;&lt;/span&gt;
&lt;span class="kd"&gt;const&lt;/span&gt; &lt;span class="n"&gt;response&lt;/span&gt; &lt;span class="o"&gt;=&lt;/span&gt; &lt;span class="k"&gt;await&lt;/span&gt; &lt;span class="nb"&gt;fetch&lt;/span&gt;&lt;span class="p"&gt;(&lt;/span&gt;&lt;span class="n"&gt;endpoint&lt;/span&gt;&lt;span class="p"&gt;)&lt;/span&gt;&lt;span class="p"&gt;;&lt;/span&gt;
&lt;span class="k"&gt;if&lt;/span&gt; &lt;span class="p"&gt;(&lt;/span&gt;&lt;span class="o"&gt;!&lt;/span&gt;&lt;span class="n"&gt;response&lt;/span&gt;&lt;span class="p"&gt;.&lt;/span&gt;&lt;span class="n"&gt;ok&lt;/span&gt;&lt;span class="p"&gt;)&lt;/span&gt; &lt;span class="k"&gt;throw&lt;/span&gt; &lt;span class="k"&gt;new&lt;/span&gt; &lt;span class="nb"&gt;Error&lt;/span&gt;&lt;span class="p"&gt;(&lt;/span&gt;&lt;span class="sb"&gt;`Rippling HTTP ${response.status}`&lt;/span&gt;&lt;span class="p"&gt;)&lt;/span&gt;&lt;span class="p"&gt;;&lt;/span&gt;
&lt;span class="kd"&gt;const&lt;/span&gt; &lt;span class="n"&gt;jobs&lt;/span&gt; &lt;span class="o"&gt;=&lt;/span&gt; &lt;span class="k"&gt;await&lt;/span&gt; &lt;span class="n"&gt;response&lt;/span&gt;&lt;span class="p"&gt;.&lt;/span&gt;&lt;span class="nf"&gt;json&lt;/span&gt;&lt;span class="p"&gt;(&lt;/span&gt;&lt;span class="p"&gt;)&lt;/span&gt;&lt;span class="p"&gt;;&lt;/span&gt;
&lt;span class="k"&gt;if&lt;/span&gt; &lt;span class="p"&gt;(&lt;/span&gt;&lt;span class="o"&gt;!&lt;/span&gt;&lt;span class="nb"&gt;Array&lt;/span&gt;&lt;span class="p"&gt;.&lt;/span&gt;&lt;span class="nf"&gt;isArray&lt;/span&gt;&lt;span class="p"&gt;(&lt;/span&gt;&lt;span class="n"&gt;jobs&lt;/span&gt;&lt;span class="p"&gt;)&lt;/span&gt;&lt;span class="p"&gt;)&lt;/span&gt; &lt;span class="k"&gt;throw&lt;/span&gt; &lt;span class="k"&gt;new&lt;/span&gt; &lt;span class="nb"&gt;Error&lt;/span&gt;&lt;span class="p"&gt;(&lt;/span&gt;&lt;span class="s1"&gt;'Expected a Rippling job array'&lt;/span&gt;&lt;span class="p"&gt;)&lt;/span&gt;&lt;span class="p"&gt;;&lt;/span&gt;
&lt;span class="nb"&gt;console&lt;/span&gt;&lt;span class="p"&gt;.&lt;/span&gt;&lt;span class="nf"&gt;log&lt;/span&gt;&lt;span class="p"&gt;(&lt;/span&gt;&lt;span class="nb"&gt;JSON&lt;/span&gt;&lt;span class="p"&gt;.&lt;/span&gt;&lt;span class="nf"&gt;stringify&lt;/span&gt;&lt;span class="p"&gt;(&lt;/span&gt;&lt;span class="n"&gt;jobs&lt;/span&gt;&lt;span class="p"&gt;.&lt;/span&gt;&lt;span class="nf"&gt;map&lt;/span&gt;&lt;span class="p"&gt;(&lt;/span&gt;&lt;span class="n"&gt;job&lt;/span&gt; &lt;span class="o"&gt;=&amp;gt;&lt;/span&gt; &lt;span class="p"&gt;(&lt;/span&gt;&lt;span class="p"&gt;{&lt;/span&gt;
  &lt;span class="n"&gt;id&lt;/span&gt;&lt;span class="p"&gt;:&lt;/span&gt; &lt;span class="n"&gt;job&lt;/span&gt;&lt;span class="p"&gt;.&lt;/span&gt;&lt;span class="n"&gt;uuid&lt;/span&gt;&lt;span class="p"&gt;,&lt;/span&gt; &lt;span class="n"&gt;title&lt;/span&gt;&lt;span class="p"&gt;:&lt;/span&gt; &lt;span class="n"&gt;job&lt;/span&gt;&lt;span class="p"&gt;.&lt;/span&gt;&lt;span class="n"&gt;name&lt;/span&gt;&lt;span class="p"&gt;,&lt;/span&gt; &lt;span class="n"&gt;location&lt;/span&gt;&lt;span class="p"&gt;:&lt;/span&gt; &lt;span class="n"&gt;job&lt;/span&gt;&lt;span class="p"&gt;.&lt;/span&gt;&lt;span class="n"&gt;workLocation&lt;/span&gt;&lt;span class="o"&gt;?&lt;/span&gt;&lt;span class="p"&gt;.&lt;/span&gt;&lt;span class="n"&gt;label&lt;/span&gt; &lt;span class="o"&gt;??&lt;/span&gt; &lt;span class="kc"&gt;null&lt;/span&gt;&lt;span class="p"&gt;,&lt;/span&gt;
  &lt;span class="n"&gt;department&lt;/span&gt;&lt;span class="p"&gt;:&lt;/span&gt; &lt;span class="n"&gt;job&lt;/span&gt;&lt;span class="p"&gt;.&lt;/span&gt;&lt;span class="n"&gt;department&lt;/span&gt;&lt;span class="o"&gt;?&lt;/span&gt;&lt;span class="p"&gt;.&lt;/span&gt;&lt;span class="n"&gt;label&lt;/span&gt; &lt;span class="o"&gt;??&lt;/span&gt; &lt;span class="kc"&gt;null&lt;/span&gt;&lt;span class="p"&gt;,&lt;/span&gt; &lt;span class="n"&gt;url&lt;/span&gt;&lt;span class="p"&gt;:&lt;/span&gt; &lt;span class="n"&gt;job&lt;/span&gt;&lt;span class="p"&gt;.&lt;/span&gt;&lt;span class="n"&gt;url&lt;/span&gt; &lt;span class="o"&gt;??&lt;/span&gt; &lt;span class="kc"&gt;null&lt;/span&gt;&lt;span class="p"&gt;,&lt;/span&gt;
&lt;span class="p"&gt;}&lt;/span&gt;&lt;span class="p"&gt;)&lt;/span&gt;&lt;span class="p"&gt;)&lt;/span&gt;&lt;span class="p"&gt;)&lt;/span&gt;&lt;span class="p"&gt;)&lt;/span&gt;&lt;span class="p"&gt;;&lt;/span&gt;
&lt;/code&gt;&lt;/pre&gt;

&lt;/div&gt;



&lt;h2&gt;
  
  
  What I verified, and what stayed unconfirmed
&lt;/h2&gt;

&lt;p&gt;Both snippets are covered by tests that replay saved responses: the Workday one checks the POST body, the offsets, the preserved total and the two-page stop, and the Rippling one checks the endpoint and the mapped fields.&lt;/p&gt;

&lt;p&gt;BambooHR did not reach the same standard of evidence. My six recorded requests produced four HTTP 302 responses and two empty JSON lists. I followed no redirects and made no retries. I exhausted the request allowance without obtaining a positive jobs fixture, so I left BambooHR out. These results do not establish that BambooHR cannot be read; they establish that I could not confirm it in this investigation.&lt;/p&gt;

&lt;p&gt;I have incorporated the Workday and Rippling handling into &lt;a href="https://apify.com/textokit/new-jobs-monitor" rel="noopener noreferrer"&gt;New Jobs Monitor on Apify&lt;/a&gt;. The implementation is also available in its &lt;a href="https://github.com/martinPisacreta/new-jobs-monitor" rel="noopener noreferrer"&gt;public GitHub repository&lt;/a&gt;.&lt;/p&gt;

</description>
      <category>webdev</category>
      <category>javascript</category>
      <category>api</category>
      <category>tutorial</category>
    </item>
    <item>
      <title>Eight public job-board APIs you can read without scraping (and the trap in each)</title>
      <dc:creator>Martin Pisacreta</dc:creator>
      <pubDate>Sun, 15 Nov 2026 13:24:13 +0000</pubDate>
      <link>https://dev.to/martinpisacreta/eight-public-job-board-apis-you-can-read-without-scraping-and-the-trap-in-each-2bf6</link>
      <guid>https://dev.to/martinpisacreta/eight-public-job-board-apis-you-can-read-without-scraping-and-the-trap-in-each-2bf6</guid>
      <description>&lt;p&gt;I wrote adapters for eight public job-board systems by reading JSON endpoints or an XML feed directly. I start with those responses when I need postings from a company careers page. The HTTP request is straightforward; the useful work comes afterward, when I choose identifiers, interpret compensation, follow pagination, and compare posting IDs with saved history.&lt;/p&gt;

&lt;p&gt;When I ran this on 15 November 2026, all eight endpoints returned postings. I organize the mappings around the traps I need to handle when turning those responses into a consistent job record.&lt;/p&gt;

&lt;h2&gt;
  
  
  Find the board identifier first
&lt;/h2&gt;

&lt;p&gt;I use a board identifier as the company argument. I take that identifier from the corresponding hosted board or feed and substitute it for &lt;code&gt;{board}&lt;/code&gt;. A display name or a full careers URL belongs to a different part of the workflow: I resolve the identifier before constructing the request.&lt;/p&gt;

&lt;div class="table-wrapper-paragraph"&gt;&lt;table&gt;
&lt;thead&gt;
&lt;tr&gt;
&lt;th&gt;System&lt;/th&gt;
&lt;th&gt;Public URL used&lt;/th&gt;
&lt;th&gt;Format&lt;/th&gt;
&lt;th&gt;Pagination in this implementation&lt;/th&gt;
&lt;th&gt;Identifier location&lt;/th&gt;
&lt;/tr&gt;
&lt;/thead&gt;
&lt;tbody&gt;
&lt;tr&gt;
&lt;td&gt;Greenhouse&lt;/td&gt;
&lt;td&gt;&lt;code&gt;https://boards-api.greenhouse.io/v1/boards/{board}/jobs?content=true&lt;/code&gt;&lt;/td&gt;
&lt;td&gt;JSON&lt;/td&gt;
&lt;td&gt;One request&lt;/td&gt;
&lt;td&gt;Hosted board path&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;Lever&lt;/td&gt;
&lt;td&gt;&lt;code&gt;https://api.lever.co/v0/postings/{board}?mode=json&lt;/code&gt;&lt;/td&gt;
&lt;td&gt;JSON&lt;/td&gt;
&lt;td&gt;One request&lt;/td&gt;
&lt;td&gt;Path after &lt;code&gt;jobs.lever.co/&lt;/code&gt;
&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;Ashby&lt;/td&gt;
&lt;td&gt;&lt;code&gt;https://api.ashbyhq.com/posting-api/job-board/{board}?includeCompensation=true&lt;/code&gt;&lt;/td&gt;
&lt;td&gt;JSON&lt;/td&gt;
&lt;td&gt;One request&lt;/td&gt;
&lt;td&gt;Path after &lt;code&gt;jobs.ashbyhq.com/&lt;/code&gt;
&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;Workable&lt;/td&gt;
&lt;td&gt;&lt;code&gt;https://apply.workable.com/api/v1/widget/accounts/{board}&lt;/code&gt;&lt;/td&gt;
&lt;td&gt;JSON&lt;/td&gt;
&lt;td&gt;One request&lt;/td&gt;
&lt;td&gt;Account path on &lt;code&gt;apply.workable.com&lt;/code&gt;
&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;Recruitee&lt;/td&gt;
&lt;td&gt;&lt;code&gt;https://{board}.recruitee.com/api/offers/&lt;/code&gt;&lt;/td&gt;
&lt;td&gt;JSON&lt;/td&gt;
&lt;td&gt;One request&lt;/td&gt;
&lt;td&gt;Hosted careers subdomain&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;SmartRecruiters&lt;/td&gt;
&lt;td&gt;&lt;code&gt;https://api.smartrecruiters.com/v1/companies/{board}/postings?limit=100&amp;amp;offset=0&lt;/code&gt;&lt;/td&gt;
&lt;td&gt;JSON&lt;/td&gt;
&lt;td&gt;Offset advances by returned posting count&lt;/td&gt;
&lt;td&gt;Public API company segment&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;Breezy HR&lt;/td&gt;
&lt;td&gt;&lt;code&gt;https://{board}.breezy.hr/json&lt;/code&gt;&lt;/td&gt;
&lt;td&gt;JSON&lt;/td&gt;
&lt;td&gt;One request&lt;/td&gt;
&lt;td&gt;Hosted careers subdomain&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;Personio&lt;/td&gt;
&lt;td&gt;&lt;code&gt;https://{board}.jobs.personio.de/xml&lt;/code&gt;&lt;/td&gt;
&lt;td&gt;XML&lt;/td&gt;
&lt;td&gt;One request&lt;/td&gt;
&lt;td&gt;Hosted careers subdomain&lt;/td&gt;
&lt;/tr&gt;
&lt;/tbody&gt;
&lt;/table&gt;&lt;/div&gt;

&lt;p&gt;I keep the following examples as standalone &lt;code&gt;.mjs&lt;/code&gt; files for Node 22. Each shows the request and the shape I read before applying the full normalization layer.&lt;/p&gt;

&lt;h2&gt;
  
  
  Greenhouse, Lever, and Ashby
&lt;/h2&gt;

&lt;p&gt;&lt;strong&gt;Greenhouse:&lt;/strong&gt; an updated posting can still be an old posting. I map &lt;code&gt;first_published&lt;/code&gt; to publication time and &lt;code&gt;updated_at&lt;/code&gt; to modification time, keeping those meanings separate. I use &lt;code&gt;id&lt;/code&gt; for identity and leave salary fields null. For monitoring, I compare that identity with saved history instead of treating every recent modification as a new vacancy.&lt;/p&gt;

&lt;p&gt;&lt;strong&gt;Lever:&lt;/strong&gt; the response is a bare array, so &lt;code&gt;body.jobs&lt;/code&gt; is undefined. I read the body directly, take the title from &lt;code&gt;text&lt;/code&gt;, and combine &lt;code&gt;categories.location&lt;/code&gt; with &lt;code&gt;categories.allLocations&lt;/code&gt;. I map structured compensation from &lt;code&gt;salaryRange&lt;/code&gt;, keeping absent values null. That gives me a consistent record while preserving the distinction between a supplied salary and a missing one.&lt;/p&gt;

&lt;p&gt;&lt;strong&gt;Ashby:&lt;/strong&gt; a compensation response can contain more than one Salary component. I select a single Salary component from &lt;code&gt;compensation.summaryComponents&lt;/code&gt; and respect its display flag. I leave salary fields null when the choice is ambiguous or hidden. I also filter out records explicitly marked &lt;code&gt;isListed: false&lt;/code&gt; before returning the visible postings.&lt;br&gt;
&lt;/p&gt;

&lt;div class="highlight js-code-highlight"&gt;
&lt;pre class="highlight javascript"&gt;&lt;code&gt;&lt;span class="kd"&gt;const&lt;/span&gt; &lt;span class="n"&gt;boards&lt;/span&gt; &lt;span class="o"&gt;=&lt;/span&gt; &lt;span class="p"&gt;[&lt;/span&gt;
  &lt;span class="p"&gt;[&lt;/span&gt;&lt;span class="s1"&gt;'greenhouse'&lt;/span&gt;&lt;span class="p"&gt;,&lt;/span&gt; &lt;span class="s1"&gt;'https://boards-api.greenhouse.io/v1/boards/stripe/jobs?content=true'&lt;/span&gt;&lt;span class="p"&gt;]&lt;/span&gt;&lt;span class="p"&gt;,&lt;/span&gt;
  &lt;span class="p"&gt;[&lt;/span&gt;&lt;span class="s1"&gt;'lever'&lt;/span&gt;&lt;span class="p"&gt;,&lt;/span&gt; &lt;span class="s1"&gt;'https://api.lever.co/v0/postings/palantir?mode=json'&lt;/span&gt;&lt;span class="p"&gt;]&lt;/span&gt;&lt;span class="p"&gt;,&lt;/span&gt;
  &lt;span class="p"&gt;[&lt;/span&gt;&lt;span class="s1"&gt;'ashby'&lt;/span&gt;&lt;span class="p"&gt;,&lt;/span&gt; &lt;span class="s1"&gt;'https://api.ashbyhq.com/posting-api/job-board/ramp?includeCompensation=true'&lt;/span&gt;&lt;span class="p"&gt;]&lt;/span&gt;&lt;span class="p"&gt;,&lt;/span&gt;
&lt;span class="p"&gt;]&lt;/span&gt;&lt;span class="p"&gt;;&lt;/span&gt;
&lt;span class="k"&gt;for&lt;/span&gt; &lt;span class="p"&gt;(&lt;/span&gt;&lt;span class="kd"&gt;const&lt;/span&gt; &lt;span class="p"&gt;[&lt;/span&gt;&lt;span class="n"&gt;source&lt;/span&gt;&lt;span class="p"&gt;,&lt;/span&gt; &lt;span class="n"&gt;url&lt;/span&gt;&lt;span class="p"&gt;]&lt;/span&gt; &lt;span class="k"&gt;of&lt;/span&gt; &lt;span class="n"&gt;boards&lt;/span&gt;&lt;span class="p"&gt;)&lt;/span&gt; &lt;span class="p"&gt;{&lt;/span&gt;
  &lt;span class="kd"&gt;const&lt;/span&gt; &lt;span class="n"&gt;response&lt;/span&gt; &lt;span class="o"&gt;=&lt;/span&gt; &lt;span class="k"&gt;await&lt;/span&gt; &lt;span class="nb"&gt;fetch&lt;/span&gt;&lt;span class="p"&gt;(&lt;/span&gt;&lt;span class="n"&gt;url&lt;/span&gt;&lt;span class="p"&gt;)&lt;/span&gt;&lt;span class="p"&gt;;&lt;/span&gt;
  &lt;span class="k"&gt;if&lt;/span&gt; &lt;span class="p"&gt;(&lt;/span&gt;&lt;span class="o"&gt;!&lt;/span&gt;&lt;span class="n"&gt;response&lt;/span&gt;&lt;span class="p"&gt;.&lt;/span&gt;&lt;span class="n"&gt;ok&lt;/span&gt;&lt;span class="p"&gt;)&lt;/span&gt; &lt;span class="k"&gt;throw&lt;/span&gt; &lt;span class="k"&gt;new&lt;/span&gt; &lt;span class="nb"&gt;Error&lt;/span&gt;&lt;span class="p"&gt;(&lt;/span&gt;&lt;span class="sb"&gt;`${source}: HTTP ${response.status}`&lt;/span&gt;&lt;span class="p"&gt;)&lt;/span&gt;&lt;span class="p"&gt;;&lt;/span&gt;
  &lt;span class="kd"&gt;const&lt;/span&gt; &lt;span class="n"&gt;body&lt;/span&gt; &lt;span class="o"&gt;=&lt;/span&gt; &lt;span class="k"&gt;await&lt;/span&gt; &lt;span class="n"&gt;response&lt;/span&gt;&lt;span class="p"&gt;.&lt;/span&gt;&lt;span class="nf"&gt;json&lt;/span&gt;&lt;span class="p"&gt;(&lt;/span&gt;&lt;span class="p"&gt;)&lt;/span&gt;&lt;span class="p"&gt;;&lt;/span&gt;
  &lt;span class="kd"&gt;const&lt;/span&gt; &lt;span class="n"&gt;jobs&lt;/span&gt; &lt;span class="o"&gt;=&lt;/span&gt; &lt;span class="n"&gt;source&lt;/span&gt; &lt;span class="o"&gt;===&lt;/span&gt; &lt;span class="s1"&gt;'lever'&lt;/span&gt; &lt;span class="o"&gt;?&lt;/span&gt; &lt;span class="n"&gt;body&lt;/span&gt; &lt;span class="p"&gt;:&lt;/span&gt; &lt;span class="n"&gt;body&lt;/span&gt;&lt;span class="p"&gt;.&lt;/span&gt;&lt;span class="n"&gt;jobs&lt;/span&gt;&lt;span class="p"&gt;;&lt;/span&gt;
  &lt;span class="k"&gt;if&lt;/span&gt; &lt;span class="p"&gt;(&lt;/span&gt;&lt;span class="o"&gt;!&lt;/span&gt;&lt;span class="nb"&gt;Array&lt;/span&gt;&lt;span class="p"&gt;.&lt;/span&gt;&lt;span class="nf"&gt;isArray&lt;/span&gt;&lt;span class="p"&gt;(&lt;/span&gt;&lt;span class="n"&gt;jobs&lt;/span&gt;&lt;span class="p"&gt;)&lt;/span&gt;&lt;span class="p"&gt;)&lt;/span&gt; &lt;span class="k"&gt;throw&lt;/span&gt; &lt;span class="k"&gt;new&lt;/span&gt; &lt;span class="nb"&gt;Error&lt;/span&gt;&lt;span class="p"&gt;(&lt;/span&gt;&lt;span class="sb"&gt;`${source}: unexpected list`&lt;/span&gt;&lt;span class="p"&gt;)&lt;/span&gt;&lt;span class="p"&gt;;&lt;/span&gt;
  &lt;span class="kd"&gt;const&lt;/span&gt; &lt;span class="n"&gt;visible&lt;/span&gt; &lt;span class="o"&gt;=&lt;/span&gt; &lt;span class="n"&gt;source&lt;/span&gt; &lt;span class="o"&gt;===&lt;/span&gt; &lt;span class="s1"&gt;'ashby'&lt;/span&gt;
    &lt;span class="o"&gt;?&lt;/span&gt; &lt;span class="n"&gt;jobs&lt;/span&gt;&lt;span class="p"&gt;.&lt;/span&gt;&lt;span class="nf"&gt;filter&lt;/span&gt;&lt;span class="p"&gt;(&lt;/span&gt;&lt;span class="n"&gt;job&lt;/span&gt; &lt;span class="o"&gt;=&amp;gt;&lt;/span&gt; &lt;span class="n"&gt;job&lt;/span&gt;&lt;span class="p"&gt;.&lt;/span&gt;&lt;span class="n"&gt;isListed&lt;/span&gt; &lt;span class="o"&gt;!==&lt;/span&gt; &lt;span class="kc"&gt;false&lt;/span&gt;&lt;span class="p"&gt;)&lt;/span&gt; &lt;span class="p"&gt;:&lt;/span&gt; &lt;span class="n"&gt;jobs&lt;/span&gt;&lt;span class="p"&gt;;&lt;/span&gt;
  &lt;span class="nb"&gt;console&lt;/span&gt;&lt;span class="p"&gt;.&lt;/span&gt;&lt;span class="nf"&gt;log&lt;/span&gt;&lt;span class="p"&gt;(&lt;/span&gt;&lt;span class="n"&gt;source&lt;/span&gt;&lt;span class="p"&gt;,&lt;/span&gt; &lt;span class="n"&gt;visible&lt;/span&gt;&lt;span class="p"&gt;.&lt;/span&gt;&lt;span class="nf"&gt;map&lt;/span&gt;&lt;span class="p"&gt;(&lt;/span&gt;&lt;span class="n"&gt;job&lt;/span&gt; &lt;span class="o"&gt;=&amp;gt;&lt;/span&gt; &lt;span class="n"&gt;job&lt;/span&gt;&lt;span class="p"&gt;.&lt;/span&gt;&lt;span class="n"&gt;id&lt;/span&gt;&lt;span class="p"&gt;)&lt;/span&gt;&lt;span class="p"&gt;)&lt;/span&gt;&lt;span class="p"&gt;;&lt;/span&gt;
&lt;span class="p"&gt;}&lt;/span&gt;
&lt;/code&gt;&lt;/pre&gt;

&lt;/div&gt;



&lt;h2&gt;
  
  
  Workable, Recruitee, and SmartRecruiters
&lt;/h2&gt;

&lt;p&gt;&lt;strong&gt;Workable:&lt;/strong&gt; posting identity lives in &lt;code&gt;shortcode&lt;/code&gt;, so reading a generic &lt;code&gt;id&lt;/code&gt; loses the key I need. I exclude hidden location entries and fall back to the older city, state, and country fields when necessary. I map &lt;code&gt;published_on&lt;/code&gt; to publication time and keep compensation fields null.&lt;/p&gt;

&lt;p&gt;&lt;strong&gt;Recruitee:&lt;/strong&gt; treating every salary value as an object breaks the field mapping. I accept &lt;code&gt;salary&lt;/code&gt; when it is an object, then map its numeric bounds, currency, and period independently. I read location from &lt;code&gt;location&lt;/code&gt; or assemble it from city, state, and country.&lt;/p&gt;

&lt;p&gt;&lt;strong&gt;SmartRecruiters:&lt;/strong&gt; HTTP 200 with an empty list can hide a mistyped company identifier. When I ran this on 15 November 2026, an identifier that did not exist returned HTTP 200 with &lt;code&gt;totalFound: 0&lt;/code&gt; and an empty &lt;code&gt;content&lt;/code&gt; array. A typo therefore looked the same as a company with no openings. I preserve identifier casing and check the intended company when that result is surprising.&lt;/p&gt;

&lt;p&gt;Pagination matters here too. In that run, SmartRecruiters returned 100 postings out of 190 on the first page. I read &lt;code&gt;content&lt;/code&gt;, validate the nonnegative integer &lt;code&gt;totalFound&lt;/code&gt;, and advance the offset by the number actually returned. In the full implementation, I deduplicate IDs across pages and reject an empty page before the reported total.&lt;br&gt;
&lt;/p&gt;

&lt;div class="highlight js-code-highlight"&gt;
&lt;pre class="highlight javascript"&gt;&lt;code&gt;&lt;span class="kd"&gt;const&lt;/span&gt; &lt;span class="n"&gt;endpoints&lt;/span&gt; &lt;span class="o"&gt;=&lt;/span&gt; &lt;span class="p"&gt;[&lt;/span&gt;
  &lt;span class="p"&gt;[&lt;/span&gt;&lt;span class="s1"&gt;'workable'&lt;/span&gt;&lt;span class="p"&gt;,&lt;/span&gt; &lt;span class="s1"&gt;'https://apply.workable.com/api/v1/widget/accounts/huggingface'&lt;/span&gt;&lt;span class="p"&gt;,&lt;/span&gt; &lt;span class="s1"&gt;'jobs'&lt;/span&gt;&lt;span class="p"&gt;]&lt;/span&gt;&lt;span class="p"&gt;,&lt;/span&gt;
  &lt;span class="p"&gt;[&lt;/span&gt;&lt;span class="s1"&gt;'recruitee'&lt;/span&gt;&lt;span class="p"&gt;,&lt;/span&gt; &lt;span class="s1"&gt;'https://bunq.recruitee.com/api/offers/'&lt;/span&gt;&lt;span class="p"&gt;,&lt;/span&gt; &lt;span class="s1"&gt;'offers'&lt;/span&gt;&lt;span class="p"&gt;]&lt;/span&gt;&lt;span class="p"&gt;,&lt;/span&gt;
  &lt;span class="p"&gt;[&lt;/span&gt;&lt;span class="s1"&gt;'smartrecruiters'&lt;/span&gt;&lt;span class="p"&gt;,&lt;/span&gt; &lt;span class="s1"&gt;'https://api.smartrecruiters.com/v1/companies/LinkedIn3/postings'&lt;/span&gt;&lt;span class="p"&gt;,&lt;/span&gt; &lt;span class="s1"&gt;'content'&lt;/span&gt;&lt;span class="p"&gt;]&lt;/span&gt;&lt;span class="p"&gt;,&lt;/span&gt;
&lt;span class="p"&gt;]&lt;/span&gt;&lt;span class="p"&gt;;&lt;/span&gt;
&lt;span class="k"&gt;for&lt;/span&gt; &lt;span class="p"&gt;(&lt;/span&gt;&lt;span class="kd"&gt;const&lt;/span&gt; &lt;span class="p"&gt;[&lt;/span&gt;&lt;span class="n"&gt;source&lt;/span&gt;&lt;span class="p"&gt;,&lt;/span&gt; &lt;span class="n"&gt;base&lt;/span&gt;&lt;span class="p"&gt;,&lt;/span&gt; &lt;span class="n"&gt;field&lt;/span&gt;&lt;span class="p"&gt;]&lt;/span&gt; &lt;span class="k"&gt;of&lt;/span&gt; &lt;span class="n"&gt;endpoints&lt;/span&gt;&lt;span class="p"&gt;)&lt;/span&gt; &lt;span class="p"&gt;{&lt;/span&gt;
  &lt;span class="kd"&gt;let&lt;/span&gt; &lt;span class="n"&gt;offset&lt;/span&gt; &lt;span class="o"&gt;=&lt;/span&gt; &lt;span class="mi"&gt;0&lt;/span&gt;&lt;span class="p"&gt;;&lt;/span&gt;
  &lt;span class="k"&gt;for&lt;/span&gt; &lt;span class="p"&gt;(&lt;/span&gt;&lt;span class="kd"&gt;let&lt;/span&gt; &lt;span class="n"&gt;page&lt;/span&gt; &lt;span class="o"&gt;=&lt;/span&gt; &lt;span class="mi"&gt;0&lt;/span&gt;&lt;span class="p"&gt;;&lt;/span&gt; &lt;span class="n"&gt;page&lt;/span&gt; &lt;span class="o"&gt;&amp;lt;&lt;/span&gt; &lt;span class="mi"&gt;60&lt;/span&gt;&lt;span class="p"&gt;;&lt;/span&gt; &lt;span class="n"&gt;page&lt;/span&gt;&lt;span class="o"&gt;++&lt;/span&gt;&lt;span class="p"&gt;)&lt;/span&gt; &lt;span class="p"&gt;{&lt;/span&gt;
    &lt;span class="kd"&gt;const&lt;/span&gt; &lt;span class="n"&gt;url&lt;/span&gt; &lt;span class="o"&gt;=&lt;/span&gt; &lt;span class="n"&gt;source&lt;/span&gt; &lt;span class="o"&gt;===&lt;/span&gt; &lt;span class="s1"&gt;'smartrecruiters'&lt;/span&gt; &lt;span class="o"&gt;?&lt;/span&gt; &lt;span class="sb"&gt;`${base}?limit=100&amp;amp;offset=${offset}`&lt;/span&gt; &lt;span class="p"&gt;:&lt;/span&gt; &lt;span class="n"&gt;base&lt;/span&gt;&lt;span class="p"&gt;;&lt;/span&gt;
    &lt;span class="kd"&gt;const&lt;/span&gt; &lt;span class="n"&gt;response&lt;/span&gt; &lt;span class="o"&gt;=&lt;/span&gt; &lt;span class="k"&gt;await&lt;/span&gt; &lt;span class="nb"&gt;fetch&lt;/span&gt;&lt;span class="p"&gt;(&lt;/span&gt;&lt;span class="n"&gt;url&lt;/span&gt;&lt;span class="p"&gt;)&lt;/span&gt;&lt;span class="p"&gt;;&lt;/span&gt;
    &lt;span class="k"&gt;if&lt;/span&gt; &lt;span class="p"&gt;(&lt;/span&gt;&lt;span class="o"&gt;!&lt;/span&gt;&lt;span class="n"&gt;response&lt;/span&gt;&lt;span class="p"&gt;.&lt;/span&gt;&lt;span class="n"&gt;ok&lt;/span&gt;&lt;span class="p"&gt;)&lt;/span&gt; &lt;span class="k"&gt;throw&lt;/span&gt; &lt;span class="k"&gt;new&lt;/span&gt; &lt;span class="nb"&gt;Error&lt;/span&gt;&lt;span class="p"&gt;(&lt;/span&gt;&lt;span class="sb"&gt;`HTTP ${response.status}`&lt;/span&gt;&lt;span class="p"&gt;)&lt;/span&gt;&lt;span class="p"&gt;;&lt;/span&gt;
    &lt;span class="kd"&gt;const&lt;/span&gt; &lt;span class="n"&gt;body&lt;/span&gt; &lt;span class="o"&gt;=&lt;/span&gt; &lt;span class="k"&gt;await&lt;/span&gt; &lt;span class="n"&gt;response&lt;/span&gt;&lt;span class="p"&gt;.&lt;/span&gt;&lt;span class="nf"&gt;json&lt;/span&gt;&lt;span class="p"&gt;(&lt;/span&gt;&lt;span class="p"&gt;)&lt;/span&gt;&lt;span class="p"&gt;,&lt;/span&gt; &lt;span class="n"&gt;jobs&lt;/span&gt; &lt;span class="o"&gt;=&lt;/span&gt; &lt;span class="n"&gt;body&lt;/span&gt;&lt;span class="p"&gt;[&lt;/span&gt;&lt;span class="n"&gt;field&lt;/span&gt;&lt;span class="p"&gt;]&lt;/span&gt;&lt;span class="p"&gt;;&lt;/span&gt;
    &lt;span class="k"&gt;if&lt;/span&gt; &lt;span class="p"&gt;(&lt;/span&gt;&lt;span class="o"&gt;!&lt;/span&gt;&lt;span class="nb"&gt;Array&lt;/span&gt;&lt;span class="p"&gt;.&lt;/span&gt;&lt;span class="nf"&gt;isArray&lt;/span&gt;&lt;span class="p"&gt;(&lt;/span&gt;&lt;span class="n"&gt;jobs&lt;/span&gt;&lt;span class="p"&gt;)&lt;/span&gt;&lt;span class="p"&gt;)&lt;/span&gt; &lt;span class="k"&gt;throw&lt;/span&gt; &lt;span class="k"&gt;new&lt;/span&gt; &lt;span class="nb"&gt;Error&lt;/span&gt;&lt;span class="p"&gt;(&lt;/span&gt;&lt;span class="s1"&gt;'Unexpected list'&lt;/span&gt;&lt;span class="p"&gt;)&lt;/span&gt;&lt;span class="p"&gt;;&lt;/span&gt;
    &lt;span class="nb"&gt;console&lt;/span&gt;&lt;span class="p"&gt;.&lt;/span&gt;&lt;span class="nf"&gt;log&lt;/span&gt;&lt;span class="p"&gt;(&lt;/span&gt;&lt;span class="n"&gt;source&lt;/span&gt;&lt;span class="p"&gt;,&lt;/span&gt; &lt;span class="n"&gt;jobs&lt;/span&gt;&lt;span class="p"&gt;)&lt;/span&gt;&lt;span class="p"&gt;;&lt;/span&gt; &lt;span class="k"&gt;if&lt;/span&gt; &lt;span class="p"&gt;(&lt;/span&gt;&lt;span class="n"&gt;source&lt;/span&gt; &lt;span class="o"&gt;!==&lt;/span&gt; &lt;span class="s1"&gt;'smartrecruiters'&lt;/span&gt;&lt;span class="p"&gt;)&lt;/span&gt; &lt;span class="k"&gt;break&lt;/span&gt;&lt;span class="p"&gt;;&lt;/span&gt;
    &lt;span class="k"&gt;if&lt;/span&gt; &lt;span class="p"&gt;(&lt;/span&gt;&lt;span class="o"&gt;!&lt;/span&gt;&lt;span class="nb"&gt;Number&lt;/span&gt;&lt;span class="p"&gt;.&lt;/span&gt;&lt;span class="nf"&gt;isSafeInteger&lt;/span&gt;&lt;span class="p"&gt;(&lt;/span&gt;&lt;span class="n"&gt;body&lt;/span&gt;&lt;span class="p"&gt;.&lt;/span&gt;&lt;span class="n"&gt;totalFound&lt;/span&gt;&lt;span class="p"&gt;)&lt;/span&gt; &lt;span class="o"&gt;||&lt;/span&gt; &lt;span class="n"&gt;body&lt;/span&gt;&lt;span class="p"&gt;.&lt;/span&gt;&lt;span class="n"&gt;totalFound&lt;/span&gt; &lt;span class="o"&gt;&amp;lt;&lt;/span&gt; &lt;span class="mi"&gt;0&lt;/span&gt;&lt;span class="p"&gt;)&lt;/span&gt; &lt;span class="k"&gt;throw&lt;/span&gt; &lt;span class="k"&gt;new&lt;/span&gt; &lt;span class="nb"&gt;Error&lt;/span&gt;&lt;span class="p"&gt;(&lt;/span&gt;&lt;span class="s1"&gt;'Invalid total'&lt;/span&gt;&lt;span class="p"&gt;)&lt;/span&gt;&lt;span class="p"&gt;;&lt;/span&gt;
    &lt;span class="n"&gt;offset&lt;/span&gt; &lt;span class="o"&gt;+=&lt;/span&gt; &lt;span class="n"&gt;jobs&lt;/span&gt;&lt;span class="p"&gt;.&lt;/span&gt;&lt;span class="n"&gt;length&lt;/span&gt;&lt;span class="p"&gt;;&lt;/span&gt;
    &lt;span class="k"&gt;if&lt;/span&gt; &lt;span class="p"&gt;(&lt;/span&gt;&lt;span class="n"&gt;offset&lt;/span&gt; &lt;span class="o"&gt;&amp;gt;=&lt;/span&gt; &lt;span class="n"&gt;body&lt;/span&gt;&lt;span class="p"&gt;.&lt;/span&gt;&lt;span class="n"&gt;totalFound&lt;/span&gt;&lt;span class="p"&gt;)&lt;/span&gt; &lt;span class="k"&gt;break&lt;/span&gt;&lt;span class="p"&gt;;&lt;/span&gt;
    &lt;span class="k"&gt;if&lt;/span&gt; &lt;span class="p"&gt;(&lt;/span&gt;&lt;span class="o"&gt;!&lt;/span&gt;&lt;span class="n"&gt;jobs&lt;/span&gt;&lt;span class="p"&gt;.&lt;/span&gt;&lt;span class="n"&gt;length&lt;/span&gt;&lt;span class="p"&gt;)&lt;/span&gt; &lt;span class="k"&gt;throw&lt;/span&gt; &lt;span class="k"&gt;new&lt;/span&gt; &lt;span class="nb"&gt;Error&lt;/span&gt;&lt;span class="p"&gt;(&lt;/span&gt;&lt;span class="s1"&gt;'Premature empty page'&lt;/span&gt;&lt;span class="p"&gt;)&lt;/span&gt;&lt;span class="p"&gt;;&lt;/span&gt;
  &lt;span class="p"&gt;}&lt;/span&gt;
&lt;span class="p"&gt;}&lt;/span&gt;
&lt;/code&gt;&lt;/pre&gt;

&lt;/div&gt;



&lt;h2&gt;
  
  
  Breezy HR and Personio
&lt;/h2&gt;

&lt;p&gt;&lt;strong&gt;Breezy HR:&lt;/strong&gt; free-text compensation needs interpretation before it can become a currency and range. I read the public &lt;code&gt;/json&lt;/code&gt; feed and keep salary fields null in my normalized output. I also convert invalid publication dates to null, a behavior covered by the local tests.&lt;/p&gt;

&lt;p&gt;&lt;strong&gt;Personio:&lt;/strong&gt; calling &lt;code&gt;response.json()&lt;/code&gt; on this feed fails because the response is XML. I retrieve text and use a dedicated reader written without dependencies. Its tests include rejection of external entity declarations. I map &lt;code&gt;createdAt&lt;/code&gt; to publication time and keep update and salary fields null.&lt;br&gt;
&lt;/p&gt;

&lt;div class="highlight js-code-highlight"&gt;
&lt;pre class="highlight javascript"&gt;&lt;code&gt;&lt;span class="kd"&gt;const&lt;/span&gt; &lt;span class="n"&gt;feeds&lt;/span&gt; &lt;span class="o"&gt;=&lt;/span&gt; &lt;span class="p"&gt;[&lt;/span&gt;
  &lt;span class="p"&gt;[&lt;/span&gt;&lt;span class="s1"&gt;'breezy'&lt;/span&gt;&lt;span class="p"&gt;,&lt;/span&gt; &lt;span class="s1"&gt;'https://breezy.breezy.hr/json'&lt;/span&gt;&lt;span class="p"&gt;]&lt;/span&gt;&lt;span class="p"&gt;,&lt;/span&gt;
  &lt;span class="p"&gt;[&lt;/span&gt;&lt;span class="s1"&gt;'personio'&lt;/span&gt;&lt;span class="p"&gt;,&lt;/span&gt; &lt;span class="s1"&gt;'https://personio.jobs.personio.de/xml'&lt;/span&gt;&lt;span class="p"&gt;]&lt;/span&gt;&lt;span class="p"&gt;,&lt;/span&gt;
&lt;span class="p"&gt;]&lt;/span&gt;&lt;span class="p"&gt;;&lt;/span&gt;
&lt;span class="k"&gt;for&lt;/span&gt; &lt;span class="p"&gt;(&lt;/span&gt;&lt;span class="kd"&gt;const&lt;/span&gt; &lt;span class="p"&gt;[&lt;/span&gt;&lt;span class="n"&gt;source&lt;/span&gt;&lt;span class="p"&gt;,&lt;/span&gt; &lt;span class="n"&gt;url&lt;/span&gt;&lt;span class="p"&gt;]&lt;/span&gt; &lt;span class="k"&gt;of&lt;/span&gt; &lt;span class="n"&gt;feeds&lt;/span&gt;&lt;span class="p"&gt;)&lt;/span&gt; &lt;span class="p"&gt;{&lt;/span&gt;
  &lt;span class="kd"&gt;const&lt;/span&gt; &lt;span class="n"&gt;response&lt;/span&gt; &lt;span class="o"&gt;=&lt;/span&gt; &lt;span class="k"&gt;await&lt;/span&gt; &lt;span class="nb"&gt;fetch&lt;/span&gt;&lt;span class="p"&gt;(&lt;/span&gt;&lt;span class="n"&gt;url&lt;/span&gt;&lt;span class="p"&gt;)&lt;/span&gt;&lt;span class="p"&gt;;&lt;/span&gt;
  &lt;span class="k"&gt;if&lt;/span&gt; &lt;span class="p"&gt;(&lt;/span&gt;&lt;span class="o"&gt;!&lt;/span&gt;&lt;span class="n"&gt;response&lt;/span&gt;&lt;span class="p"&gt;.&lt;/span&gt;&lt;span class="n"&gt;ok&lt;/span&gt;&lt;span class="p"&gt;)&lt;/span&gt; &lt;span class="k"&gt;throw&lt;/span&gt; &lt;span class="k"&gt;new&lt;/span&gt; &lt;span class="nb"&gt;Error&lt;/span&gt;&lt;span class="p"&gt;(&lt;/span&gt;&lt;span class="sb"&gt;`${source}: HTTP ${response.status}`&lt;/span&gt;&lt;span class="p"&gt;)&lt;/span&gt;&lt;span class="p"&gt;;&lt;/span&gt;
  &lt;span class="kd"&gt;const&lt;/span&gt; &lt;span class="n"&gt;body&lt;/span&gt; &lt;span class="o"&gt;=&lt;/span&gt; &lt;span class="n"&gt;source&lt;/span&gt; &lt;span class="o"&gt;===&lt;/span&gt; &lt;span class="s1"&gt;'personio'&lt;/span&gt;
    &lt;span class="o"&gt;?&lt;/span&gt; &lt;span class="k"&gt;await&lt;/span&gt; &lt;span class="n"&gt;response&lt;/span&gt;&lt;span class="p"&gt;.&lt;/span&gt;&lt;span class="nf"&gt;text&lt;/span&gt;&lt;span class="p"&gt;(&lt;/span&gt;&lt;span class="p"&gt;)&lt;/span&gt;
    &lt;span class="p"&gt;:&lt;/span&gt; &lt;span class="k"&gt;await&lt;/span&gt; &lt;span class="n"&gt;response&lt;/span&gt;&lt;span class="p"&gt;.&lt;/span&gt;&lt;span class="nf"&gt;json&lt;/span&gt;&lt;span class="p"&gt;(&lt;/span&gt;&lt;span class="p"&gt;)&lt;/span&gt;&lt;span class="p"&gt;;&lt;/span&gt;
  &lt;span class="nb"&gt;console&lt;/span&gt;&lt;span class="p"&gt;.&lt;/span&gt;&lt;span class="nf"&gt;log&lt;/span&gt;&lt;span class="p"&gt;(&lt;/span&gt;&lt;span class="n"&gt;source&lt;/span&gt;&lt;span class="p"&gt;,&lt;/span&gt; &lt;span class="n"&gt;body&lt;/span&gt;&lt;span class="p"&gt;)&lt;/span&gt;&lt;span class="p"&gt;;&lt;/span&gt;
&lt;span class="p"&gt;}&lt;/span&gt;
&lt;/code&gt;&lt;/pre&gt;

&lt;/div&gt;



&lt;h2&gt;
  
  
  Detecting only new postings
&lt;/h2&gt;

&lt;p&gt;I persist a cumulative set keyed by source, board, and posting ID. Including the source and board keeps identical raw IDs from different boards separate. After each successful collection, I compare the returned keys with that history and save the accumulated state.&lt;/p&gt;

&lt;p&gt;I define “new” as an ID observed for the first time. An edit that preserves the ID stays in history, and I keep removed IDs recorded so a later reappearance also stays old. On the first run, I can return the initial postings or establish a baseline silently and report additions afterward. I keep failed collections separate from successful empty results when updating that state.&lt;/p&gt;

&lt;h2&gt;
  
  
  What I did not verify
&lt;/h2&gt;

&lt;p&gt;I have not measured provider quotas, continuous availability, or universal coverage. A single request can leave completeness unresolved. My 60-page ceiling is an implementation safeguard; the full implementation warns about truncation, while the example simply stops. The snippets simplify transport and normalization, and the Personio example retrieves XML for later parsing.&lt;/p&gt;

&lt;p&gt;I have not independently established first-advertised dates from mapped timestamps, inferred salaries from descriptions, or demonstrated identifier discovery from every careers page. For SmartRecruiters I tried one identifier in lowercase and got the same 190 postings, which says nothing about other boards, so I still pass the casing through unchanged. The live observations above come from a single run on 15 November 2026.&lt;/p&gt;

&lt;h2&gt;
  
  
  Using the maintained adapter
&lt;/h2&gt;

&lt;p&gt;If you prefer not to maintain these mappings, the &lt;a href="https://apify.com/textokit/new-jobs-monitor" rel="noopener noreferrer"&gt;New Jobs Monitor Actor&lt;/a&gt; packages this approach. The &lt;a href="https://github.com/martinPisacreta/new-jobs-monitor" rel="noopener noreferrer"&gt;repository&lt;/a&gt; contains the implementation. Check its documented state and coverage behavior before scheduling recurring runs.&lt;/p&gt;

</description>
      <category>api</category>
      <category>node</category>
      <category>javascript</category>
      <category>webdev</category>
    </item>
    <item>
      <title>An MCP Assistant That Checks Before Repeating a Payment Reminder</title>
      <dc:creator>Martin Pisacreta</dc:creator>
      <pubDate>Wed, 04 Nov 2026 13:10:12 +0000</pubDate>
      <link>https://dev.to/martinpisacreta/an-mcp-assistant-that-checks-before-repeating-a-payment-reminder-2chm</link>
      <guid>https://dev.to/martinpisacreta/an-mcp-assistant-that-checks-before-repeating-a-payment-reminder-2chm</guid>
      <description>&lt;p&gt;An unpaid invoice does not tell an assistant whether someone has already followed up. If the assistant sorts only by due date, it can recommend contacting the same customer again while another overdue invoice has received no reminder.&lt;/p&gt;

&lt;p&gt;A recorded Claude Code session on November 3 demonstrates how reminder history changes that recommendation. The client connected to the PayPal sandbox MCP server, read two fictional invoices, retrieved a local reminder history, and requested a deterministic priority explanation. The older invoice moved behind the newer one because the supplied snapshot assigned it one previous reminder.&lt;/p&gt;

&lt;p&gt;The session produced a recommendation. The model did not approve or send a reminder.&lt;/p&gt;

&lt;h2&gt;
  
  
  What the invoice tool returned
&lt;/h2&gt;

&lt;p&gt;&lt;code&gt;list_unpaid_invoices&lt;/code&gt; returned two invoices with status &lt;code&gt;UNPAID&lt;/code&gt;, each showing an outstanding amount of &lt;code&gt;10.00&lt;/code&gt; USD. One was due October 24; the other was due October 28. I will call them the older and newer invoice, respectively, without reproducing recipient details.&lt;/p&gt;

&lt;p&gt;Both included the note: “Fictional sandbox validation invoice. No real goods or payment required.” These were demonstration records, not evidence of real customers owing money.&lt;/p&gt;

&lt;p&gt;The response reported &lt;code&gt;scannedCount: 3&lt;/code&gt;, &lt;code&gt;returnedCount: 2&lt;/code&gt;, and pagination scoped to &lt;code&gt;single_page&lt;/code&gt;, with &lt;code&gt;hasMore: false&lt;/code&gt;. That describes this response; it does not establish a general guarantee about complete account history.&lt;/p&gt;

&lt;p&gt;The list supplied dates, amounts, and invoice statuses. It did not supply a verified count of previous reminders. That context came from a separate history lookup and the snapshot subsequently passed to the priority tool.&lt;/p&gt;

&lt;h2&gt;
  
  
  Reading history without authorizing another action
&lt;/h2&gt;

&lt;p&gt;&lt;code&gt;get_reminder_history&lt;/code&gt; accepted a proposal ID. Its response explicitly identified its source: “Local SQLite snapshot only; no PayPal API request.”&lt;/p&gt;

&lt;p&gt;For the older invoice, it returned four recorded workflow stages: proposal, approval, execution attempt, and result. All were dated October 25. The approval recorded &lt;code&gt;approved_via&lt;/code&gt; as &lt;code&gt;local-terminal&lt;/code&gt;; the execution attempt recorded &lt;code&gt;state&lt;/code&gt; as &lt;code&gt;succeeded&lt;/code&gt;.&lt;/p&gt;

&lt;p&gt;The result explained exactly what that success meant: “The local executor recorded a successful sender return. Recipient delivery remains unverified.”&lt;/p&gt;

&lt;p&gt;That distinction matters. A successful local sender return is evidence about execution, not proof that a person received or read a message. The response also warned that current PayPal invoice status, payment status, and remote reminder history were not checked by this lookup.&lt;/p&gt;

&lt;p&gt;Its effects reported no persistence, approval, execution, or API call. Reading the history did not repeat the historical operation. The reconciliation section also returned &lt;code&gt;resendAllowed: false&lt;/code&gt;.&lt;/p&gt;

&lt;p&gt;There is another useful detail: the stored proposal still contained &lt;code&gt;status: pending&lt;/code&gt;. The history response explicitly warned that this draft field did not override the execution record. Looking at the proposal alone would therefore miss relevant workflow context.&lt;/p&gt;

&lt;h2&gt;
  
  
  Reading, approving, and sending are separate steps
&lt;/h2&gt;

&lt;p&gt;The recorded history makes those boundaries visible. A proposal captures a proposed reminder. An approval is a separate record. An execution attempt has its own start, completion, and state. The result describes what the executor recorded.&lt;/p&gt;

&lt;p&gt;The current AI session inspected those records rather than performing those steps again. Its connected PayPal tool list contained invoice reading, dispute reading, reminder-history reading, and priority explanation. It did not expose an approval or sending tool to this client.&lt;/p&gt;

&lt;p&gt;The source reinforces the explanation boundary. In &lt;code&gt;src/mcp-tools.ts&lt;/code&gt;, the description of &lt;code&gt;explain_invoice_priority&lt;/code&gt; says: “Does not call APIs, persist proposals, approve or send reminders.”&lt;/p&gt;

&lt;p&gt;The explanation response included a reminder draft with &lt;code&gt;requiresHumanApproval: true&lt;/code&gt;, while its effects were &lt;code&gt;persisted: false&lt;/code&gt;, &lt;code&gt;approved: false&lt;/code&gt;, and &lt;code&gt;sent: false&lt;/code&gt;. Generating wording did not create authorization. The history response likewise stated: “A historical approval does not establish current permission to execute.”&lt;/p&gt;

&lt;h2&gt;
  
  
  Why the order changed
&lt;/h2&gt;

&lt;p&gt;The client called &lt;code&gt;explain_invoice_priority&lt;/code&gt; for each invoice using the same two-invoice snapshot. It supplied one previous reminder for the older invoice and zero for the newer one, equal outstanding amounts, and &lt;code&gt;openDispute: false&lt;/code&gt; for both.&lt;/p&gt;

&lt;p&gt;The comparison used an explicit &lt;code&gt;now&lt;/code&gt; value of &lt;code&gt;2026-11-03T12:00:00Z&lt;/code&gt;. That is the supplied evaluation time, not a claim about the actual moment of execution. Both invoices were classified as overdue under that input.&lt;/p&gt;

&lt;p&gt;The older invoice's response contained this reason: “Previous reminders: 1. Within an urgency category, fewer prior reminders rank first.”&lt;/p&gt;

&lt;p&gt;Its contextual position was 2, compared with position 1 under date-only ordering. The newer invoice received contextual position 1. Because both shared the urgency category, reminder count decided their relative order before outstanding amount or due date could break a tie.&lt;/p&gt;

&lt;p&gt;The relevant comparator is in &lt;code&gt;src/priorities.ts&lt;/code&gt;:&lt;br&gt;
&lt;/p&gt;

&lt;div class="highlight js-code-highlight"&gt;
&lt;pre class="highlight typescript"&gt;&lt;code&gt;  &lt;span class="kd"&gt;const&lt;/span&gt; &lt;span class="nx"&gt;contextual&lt;/span&gt; &lt;span class="o"&gt;=&lt;/span&gt; &lt;span class="p"&gt;[...&lt;/span&gt;&lt;span class="nx"&gt;invoices&lt;/span&gt;&lt;span class="p"&gt;].&lt;/span&gt;&lt;span class="nf"&gt;sort&lt;/span&gt;&lt;span class="p"&gt;((&lt;/span&gt;&lt;span class="nx"&gt;a&lt;/span&gt;&lt;span class="p"&gt;,&lt;/span&gt; &lt;span class="nx"&gt;b&lt;/span&gt;&lt;span class="p"&gt;)&lt;/span&gt; &lt;span class="o"&gt;=&amp;gt;&lt;/span&gt; &lt;span class="p"&gt;{&lt;/span&gt;
    &lt;span class="kd"&gt;const&lt;/span&gt; &lt;span class="nx"&gt;reviewOrder&lt;/span&gt; &lt;span class="o"&gt;=&lt;/span&gt; &lt;span class="nc"&gt;Number&lt;/span&gt;&lt;span class="p"&gt;(&lt;/span&gt;&lt;span class="nf"&gt;needsReview&lt;/span&gt;&lt;span class="p"&gt;(&lt;/span&gt;&lt;span class="nx"&gt;a&lt;/span&gt;&lt;span class="p"&gt;))&lt;/span&gt; &lt;span class="o"&gt;-&lt;/span&gt; &lt;span class="nc"&gt;Number&lt;/span&gt;&lt;span class="p"&gt;(&lt;/span&gt;&lt;span class="nf"&gt;needsReview&lt;/span&gt;&lt;span class="p"&gt;(&lt;/span&gt;&lt;span class="nx"&gt;b&lt;/span&gt;&lt;span class="p"&gt;));&lt;/span&gt;
    &lt;span class="k"&gt;if &lt;/span&gt;&lt;span class="p"&gt;(&lt;/span&gt;&lt;span class="nx"&gt;reviewOrder&lt;/span&gt;&lt;span class="p"&gt;)&lt;/span&gt; &lt;span class="k"&gt;return&lt;/span&gt; &lt;span class="nx"&gt;reviewOrder&lt;/span&gt;&lt;span class="p"&gt;;&lt;/span&gt;
    &lt;span class="kd"&gt;const&lt;/span&gt; &lt;span class="nx"&gt;urgencyOrder&lt;/span&gt; &lt;span class="o"&gt;=&lt;/span&gt; &lt;span class="nx"&gt;rank&lt;/span&gt;&lt;span class="p"&gt;[&lt;/span&gt;&lt;span class="nx"&gt;classification&lt;/span&gt;&lt;span class="p"&gt;.&lt;/span&gt;&lt;span class="nf"&gt;get&lt;/span&gt;&lt;span class="p"&gt;(&lt;/span&gt;&lt;span class="nx"&gt;a&lt;/span&gt;&lt;span class="p"&gt;.&lt;/span&gt;&lt;span class="nx"&gt;id&lt;/span&gt;&lt;span class="p"&gt;)&lt;/span&gt;&lt;span class="o"&gt;!&lt;/span&gt;&lt;span class="p"&gt;.&lt;/span&gt;&lt;span class="nx"&gt;priority&lt;/span&gt;&lt;span class="p"&gt;]&lt;/span&gt; &lt;span class="o"&gt;-&lt;/span&gt; &lt;span class="nx"&gt;rank&lt;/span&gt;&lt;span class="p"&gt;[&lt;/span&gt;&lt;span class="nx"&gt;classification&lt;/span&gt;&lt;span class="p"&gt;.&lt;/span&gt;&lt;span class="nf"&gt;get&lt;/span&gt;&lt;span class="p"&gt;(&lt;/span&gt;&lt;span class="nx"&gt;b&lt;/span&gt;&lt;span class="p"&gt;.&lt;/span&gt;&lt;span class="nx"&gt;id&lt;/span&gt;&lt;span class="p"&gt;)&lt;/span&gt;&lt;span class="o"&gt;!&lt;/span&gt;&lt;span class="p"&gt;.&lt;/span&gt;&lt;span class="nx"&gt;priority&lt;/span&gt;&lt;span class="p"&gt;];&lt;/span&gt;
    &lt;span class="k"&gt;if &lt;/span&gt;&lt;span class="p"&gt;(&lt;/span&gt;&lt;span class="nx"&gt;urgencyOrder&lt;/span&gt;&lt;span class="p"&gt;)&lt;/span&gt; &lt;span class="k"&gt;return&lt;/span&gt; &lt;span class="nx"&gt;urgencyOrder&lt;/span&gt;&lt;span class="p"&gt;;&lt;/span&gt;
    &lt;span class="kd"&gt;const&lt;/span&gt; &lt;span class="nx"&gt;reminderOrder&lt;/span&gt; &lt;span class="o"&gt;=&lt;/span&gt; &lt;span class="nx"&gt;a&lt;/span&gt;&lt;span class="p"&gt;.&lt;/span&gt;&lt;span class="nx"&gt;previousReminders&lt;/span&gt; &lt;span class="o"&gt;-&lt;/span&gt; &lt;span class="nx"&gt;b&lt;/span&gt;&lt;span class="p"&gt;.&lt;/span&gt;&lt;span class="nx"&gt;previousReminders&lt;/span&gt;&lt;span class="p"&gt;;&lt;/span&gt;
    &lt;span class="k"&gt;if &lt;/span&gt;&lt;span class="p"&gt;(&lt;/span&gt;&lt;span class="nx"&gt;reminderOrder&lt;/span&gt;&lt;span class="p"&gt;)&lt;/span&gt; &lt;span class="k"&gt;return&lt;/span&gt; &lt;span class="nx"&gt;reminderOrder&lt;/span&gt;&lt;span class="p"&gt;;&lt;/span&gt;
    &lt;span class="kd"&gt;const&lt;/span&gt; &lt;span class="nx"&gt;aAmount&lt;/span&gt; &lt;span class="o"&gt;=&lt;/span&gt; &lt;span class="nx"&gt;amounts&lt;/span&gt;&lt;span class="p"&gt;.&lt;/span&gt;&lt;span class="nf"&gt;get&lt;/span&gt;&lt;span class="p"&gt;(&lt;/span&gt;&lt;span class="nx"&gt;a&lt;/span&gt;&lt;span class="p"&gt;.&lt;/span&gt;&lt;span class="nx"&gt;id&lt;/span&gt;&lt;span class="p"&gt;)&lt;/span&gt;&lt;span class="o"&gt;!&lt;/span&gt;&lt;span class="p"&gt;;&lt;/span&gt;
    &lt;span class="kd"&gt;const&lt;/span&gt; &lt;span class="nx"&gt;bAmount&lt;/span&gt; &lt;span class="o"&gt;=&lt;/span&gt; &lt;span class="nx"&gt;amounts&lt;/span&gt;&lt;span class="p"&gt;.&lt;/span&gt;&lt;span class="nf"&gt;get&lt;/span&gt;&lt;span class="p"&gt;(&lt;/span&gt;&lt;span class="nx"&gt;b&lt;/span&gt;&lt;span class="p"&gt;.&lt;/span&gt;&lt;span class="nx"&gt;id&lt;/span&gt;&lt;span class="p"&gt;)&lt;/span&gt;&lt;span class="o"&gt;!&lt;/span&gt;&lt;span class="p"&gt;;&lt;/span&gt;
    &lt;span class="k"&gt;if &lt;/span&gt;&lt;span class="p"&gt;(&lt;/span&gt;&lt;span class="nx"&gt;aAmount&lt;/span&gt; &lt;span class="o"&gt;!==&lt;/span&gt; &lt;span class="nx"&gt;bAmount&lt;/span&gt;&lt;span class="p"&gt;)&lt;/span&gt; &lt;span class="k"&gt;return&lt;/span&gt; &lt;span class="nx"&gt;aAmount&lt;/span&gt; &lt;span class="o"&gt;&amp;gt;&lt;/span&gt; &lt;span class="nx"&gt;bAmount&lt;/span&gt; &lt;span class="p"&gt;?&lt;/span&gt; &lt;span class="o"&gt;-&lt;/span&gt;&lt;span class="mi"&gt;1&lt;/span&gt; &lt;span class="p"&gt;:&lt;/span&gt; &lt;span class="mi"&gt;1&lt;/span&gt;&lt;span class="p"&gt;;&lt;/span&gt;
    &lt;span class="k"&gt;return&lt;/span&gt; &lt;span class="nf"&gt;explanationDateOrder&lt;/span&gt;&lt;span class="p"&gt;(&lt;/span&gt;&lt;span class="nx"&gt;a&lt;/span&gt;&lt;span class="p"&gt;,&lt;/span&gt; &lt;span class="nx"&gt;b&lt;/span&gt;&lt;span class="p"&gt;);&lt;/span&gt;
  &lt;span class="p"&gt;});&lt;/span&gt;
&lt;/code&gt;&lt;/pre&gt;

&lt;/div&gt;



&lt;p&gt;The subtraction of reminder counts puts zero before one. It runs after manual-review ordering and urgency, so fewer reminders do not universally override every other concern. The policy also routes invoices with an open dispute or missing deadline into manual review after the other invoices.&lt;/p&gt;

&lt;p&gt;The response describes this as a deterministic heuristic. The calling AI client writes the final explanation and reminder wording; the comparator supplies the declared ordering.&lt;/p&gt;

&lt;h2&gt;
  
  
  What this demonstration does not prove
&lt;/h2&gt;

&lt;p&gt;The priority response labels its source: “Caller-supplied snapshot; not verified against PayPal.” It does not fetch reminder history itself or independently validate the client's counts.&lt;/p&gt;

&lt;p&gt;The retrieved local history supports a recorded prior execution for the older invoice. The zero assigned to the newer invoice is an input in the transcript, not a separately demonstrated exhaustive history search. Likewise, the supplied dispute flags should not be presented as independently verified facts.&lt;/p&gt;

&lt;p&gt;This rule changes which invoice comes first. It does not permanently exclude previously contacted invoices or implement a blanket duplicate-send prohibition. The older invoice remained second in the queue. Any actual follow-up would still require a separate decision and authorization.&lt;/p&gt;

&lt;p&gt;The demonstrated outcome is narrower and useful: with explicit reminder context, an assistant can recommend reviewing the untouched invoice first instead of repeatedly prioritizing the oldest debt. In this session, no new reminder was approved or sent, and recipient delivery of the historical reminder remains unverified.&lt;/p&gt;

&lt;p&gt;Source code: &lt;a href="https://github.com/martinPisacreta/paypal-ops-agent" rel="noopener noreferrer"&gt;paypal-ops-agent&lt;/a&gt;. A sanitized transcript of this session is in the repository: &lt;a href="https://github.com/martinPisacreta/paypal-ops-agent/blob/main/evidence/ai-client-session-20261103/TRANSCRIPT.md" rel="noopener noreferrer"&gt;evidence/ai-client-session-20261103/TRANSCRIPT.md&lt;/a&gt;.&lt;/p&gt;

&lt;p&gt;If you want an MCP server that lets an AI assistant use your own API, I offer a fixed-scope service on Upwork: &lt;a href="https://www.upwork.com/services/product/2113736906027325062" rel="noopener noreferrer"&gt;https://www.upwork.com/services/product/2113736906027325062&lt;/a&gt; .&lt;/p&gt;

</description>
      <category>mcp</category>
      <category>ai</category>
      <category>typescript</category>
      <category>fintech</category>
    </item>
    <item>
      <title>CI for an ASP.NET MVC 5 app on .NET Framework 4.8 with GitHub Actions</title>
      <dc:creator>Martin Pisacreta</dc:creator>
      <pubDate>Tue, 03 Nov 2026 13:09:00 +0000</pubDate>
      <link>https://dev.to/martinpisacreta/ci-for-an-aspnet-mvc-5-app-on-net-framework-48-with-github-actions-3gah</link>
      <guid>https://dev.to/martinpisacreta/ci-for-an-aspnet-mvc-5-app-on-net-framework-48-with-github-actions-3gah</guid>
      <description>&lt;p&gt;An ASP.NET MVC 5 application on .NET Framework 4.8 needs a repeatable build long before it needs automated production deployment. A good first workflow restores packages, compiles the solution, runs the existing tests and produces a web deployment package someone can inspect.&lt;/p&gt;

&lt;p&gt;This is a setup guide built from official documentation that I re-read on November 2, 2026, and the workflow below is not just a sketch: I ran it unchanged against a minimal public sample, &lt;a href="https://github.com/martinPisacreta/legacy-mvc48-ci-sample" rel="noopener noreferrer"&gt;legacy-mvc48-ci-sample&lt;/a&gt;, an MVC 5 app on .NET Framework 4.8 with two MSTest tests. &lt;a href="https://github.com/martinPisacreta/legacy-mvc48-ci-sample/actions/runs/41474760557" rel="noopener noreferrer"&gt;That run&lt;/a&gt; on &lt;code&gt;windows-2022&lt;/code&gt; restored, built, passed 2 of 2 tests, produced the package and uploaded both artifacts. A real application will have more projects and dependencies, so the paths and names are still yours to adapt.&lt;/p&gt;

&lt;h2&gt;
  
  
  The Windows runner
&lt;/h2&gt;

&lt;p&gt;GitHub's &lt;a href="https://docs.github.com/en/actions/use-cases-and-examples/building-and-testing/building-and-testing-net" rel="noopener noreferrer"&gt;.NET build and test guide&lt;/a&gt; is written around the &lt;code&gt;dotnet&lt;/code&gt; CLI, which fits SDK-style projects. A classic MVC 5 web project with &lt;code&gt;packages.config&lt;/code&gt; is usually built with Visual Studio's MSBuild instead, so the runner matters.&lt;/p&gt;

&lt;p&gt;The &lt;a href="https://github.com/actions/runner-images/blob/main/images/windows/Windows2022-Readme.md" rel="noopener noreferrer"&gt;Windows Server 2022 runner inventory&lt;/a&gt;, image version &lt;code&gt;20261026.347.1&lt;/code&gt; when I checked, lists Visual Studio Enterprise 2022, the MSBuild component, vswhere and .NET Framework 4.7.2, 4.8 and 4.8.1. That inventory changes over time, so record the image version shown in each workflow run.&lt;/p&gt;

&lt;h2&gt;
  
  
  Write down the repository contract first
&lt;/h2&gt;

&lt;p&gt;Before touching YAML, list the solution path, the web project, the test projects, the build configuration and what the package should be called. Also list everything that only exists on a developer's machine: private NuGet feeds, DLLs referenced from a local folder, generated files, custom build tasks. Each one either gets an authorized way into CI or becomes an explicit blocker. That list is usually where legacy builds fail, not in the YAML.&lt;/p&gt;

&lt;h2&gt;
  
  
  Restore packages.config dependencies
&lt;/h2&gt;

&lt;p&gt;Microsoft's &lt;a href="https://learn.microsoft.com/en-us/nuget/reference/cli-reference/cli-ref-restore" rel="noopener noreferrer"&gt;nuget restore reference&lt;/a&gt; accepts a solution file, so &lt;code&gt;nuget restore LegacyMvc.sln&lt;/code&gt; restores every &lt;code&gt;packages.config&lt;/code&gt; in it. Keep it as its own step so a restore failure is not confused with a compile error.&lt;/p&gt;

&lt;p&gt;&lt;a href="https://github.com/NuGet/setup-nuget" rel="noopener noreferrer"&gt;NuGet/setup-nuget&lt;/a&gt; puts &lt;code&gt;nuget.exe&lt;/code&gt; on the runner; its latest release on November 2 was v4.0. &lt;code&gt;nuget-version: latest&lt;/code&gt; is convenient for a first run, and the README also accepts version ranges such as &lt;code&gt;5.x&lt;/code&gt;, so pin one once the build works. If you use private feeds, supply credentials through secrets and the feed's documented method, never by committing a developer's &lt;code&gt;NuGet.Config&lt;/code&gt;.&lt;/p&gt;

&lt;h2&gt;
  
  
  Build, test, package
&lt;/h2&gt;

&lt;p&gt;&lt;a href="https://github.com/microsoft/setup-msbuild" rel="noopener noreferrer"&gt;microsoft/setup-msbuild&lt;/a&gt; finds MSBuild with vswhere and adds it to &lt;code&gt;PATH&lt;/code&gt;. Its README is explicit that it "does NOT add other Visual Studio tools (like VSTest…)" to &lt;code&gt;PATH&lt;/code&gt;, so the test step below locates &lt;code&gt;vstest.console.exe&lt;/code&gt; itself. The MSBuild switches used here (&lt;code&gt;/m&lt;/code&gt;, &lt;code&gt;/t&lt;/code&gt;, &lt;code&gt;/p&lt;/code&gt;) are in the &lt;a href="https://learn.microsoft.com/en-us/visualstudio/msbuild/msbuild-command-line-reference?view=vs-2022" rel="noopener noreferrer"&gt;MSBuild command-line reference&lt;/a&gt;.&lt;/p&gt;

&lt;p&gt;For packaging, Microsoft's guide to &lt;a href="https://learn.microsoft.com/en-us/aspnet/web-forms/overview/deployment/web-deployment-in-the-enterprise/building-and-packaging-web-application-projects" rel="noopener noreferrer"&gt;building and packaging web application projects&lt;/a&gt; shows &lt;code&gt;DeployOnBuild=true&lt;/code&gt; with &lt;code&gt;DeployTarget=Package&lt;/code&gt;: after the build, the Web Publishing Pipeline runs its &lt;code&gt;Package&lt;/code&gt; target and produces &lt;code&gt;[project name].zip&lt;/code&gt;, plus a &lt;code&gt;.deploy.cmd&lt;/code&gt; and a &lt;code&gt;SetParameters.xml&lt;/code&gt; next to it. The template searches for that zip instead of assuming a folder.&lt;/p&gt;

&lt;p&gt;&lt;strong&gt;Tested on the sample repository above, on November 2, 2026.&lt;/strong&gt; Save an adapted copy as &lt;code&gt;.github/workflows/legacy-mvc.yml&lt;/code&gt;. The action versions are the current major tags from each project's README or releases page; for stricter supply-chain control, pin reviewed commit SHAs.&lt;br&gt;
&lt;/p&gt;

&lt;div class="highlight js-code-highlight"&gt;
&lt;pre class="highlight yaml"&gt;&lt;code&gt;&lt;span class="na"&gt;name&lt;/span&gt;&lt;span class="pi"&gt;:&lt;/span&gt; &lt;span class="s"&gt;Legacy MVC CI&lt;/span&gt;
&lt;span class="na"&gt;on&lt;/span&gt;&lt;span class="pi"&gt;:&lt;/span&gt;
  &lt;span class="na"&gt;pull_request&lt;/span&gt;&lt;span class="pi"&gt;:&lt;/span&gt;
  &lt;span class="na"&gt;push&lt;/span&gt;&lt;span class="pi"&gt;:&lt;/span&gt;
    &lt;span class="na"&gt;branches&lt;/span&gt;&lt;span class="pi"&gt;:&lt;/span&gt; &lt;span class="pi"&gt;[&lt;/span&gt;&lt;span class="nv"&gt;main&lt;/span&gt;&lt;span class="pi"&gt;]&lt;/span&gt;
  &lt;span class="na"&gt;workflow_dispatch&lt;/span&gt;&lt;span class="pi"&gt;:&lt;/span&gt;
&lt;span class="na"&gt;permissions&lt;/span&gt;&lt;span class="pi"&gt;:&lt;/span&gt;
  &lt;span class="na"&gt;contents&lt;/span&gt;&lt;span class="pi"&gt;:&lt;/span&gt; &lt;span class="s"&gt;read&lt;/span&gt;
&lt;span class="na"&gt;jobs&lt;/span&gt;&lt;span class="pi"&gt;:&lt;/span&gt;
  &lt;span class="na"&gt;build&lt;/span&gt;&lt;span class="pi"&gt;:&lt;/span&gt;
    &lt;span class="na"&gt;runs-on&lt;/span&gt;&lt;span class="pi"&gt;:&lt;/span&gt; &lt;span class="s"&gt;windows-2022&lt;/span&gt;
    &lt;span class="na"&gt;defaults&lt;/span&gt;&lt;span class="pi"&gt;:&lt;/span&gt;
      &lt;span class="na"&gt;run&lt;/span&gt;&lt;span class="pi"&gt;:&lt;/span&gt;
        &lt;span class="na"&gt;shell&lt;/span&gt;&lt;span class="pi"&gt;:&lt;/span&gt; &lt;span class="s"&gt;pwsh&lt;/span&gt;
    &lt;span class="na"&gt;steps&lt;/span&gt;&lt;span class="pi"&gt;:&lt;/span&gt;
      &lt;span class="pi"&gt;-&lt;/span&gt; &lt;span class="na"&gt;uses&lt;/span&gt;&lt;span class="pi"&gt;:&lt;/span&gt; &lt;span class="s"&gt;actions/checkout@v7&lt;/span&gt;
      &lt;span class="pi"&gt;-&lt;/span&gt; &lt;span class="na"&gt;uses&lt;/span&gt;&lt;span class="pi"&gt;:&lt;/span&gt; &lt;span class="s"&gt;microsoft/setup-msbuild@v3&lt;/span&gt;
      &lt;span class="pi"&gt;-&lt;/span&gt; &lt;span class="na"&gt;uses&lt;/span&gt;&lt;span class="pi"&gt;:&lt;/span&gt; &lt;span class="s"&gt;NuGet/setup-nuget@v4&lt;/span&gt;
        &lt;span class="na"&gt;with&lt;/span&gt;&lt;span class="pi"&gt;:&lt;/span&gt;
          &lt;span class="na"&gt;nuget-version&lt;/span&gt;&lt;span class="pi"&gt;:&lt;/span&gt; &lt;span class="s"&gt;latest&lt;/span&gt;
      &lt;span class="pi"&gt;-&lt;/span&gt; &lt;span class="na"&gt;name&lt;/span&gt;&lt;span class="pi"&gt;:&lt;/span&gt; &lt;span class="s"&gt;Restore packages.config dependencies&lt;/span&gt;
        &lt;span class="na"&gt;run&lt;/span&gt;&lt;span class="pi"&gt;:&lt;/span&gt; &lt;span class="pi"&gt;|&lt;/span&gt;
          &lt;span class="s"&gt;nuget restore LegacyMvc.sln -NonInteractive&lt;/span&gt;
          &lt;span class="s"&gt;if ($LASTEXITCODE -ne 0) { throw 'Restore failed' }&lt;/span&gt;
      &lt;span class="pi"&gt;-&lt;/span&gt; &lt;span class="na"&gt;name&lt;/span&gt;&lt;span class="pi"&gt;:&lt;/span&gt; &lt;span class="s"&gt;Build solution&lt;/span&gt;
        &lt;span class="na"&gt;run&lt;/span&gt;&lt;span class="pi"&gt;:&lt;/span&gt; &lt;span class="pi"&gt;|&lt;/span&gt;
          &lt;span class="s"&gt;msbuild LegacyMvc.sln /m /t:Build /p:Configuration=Release '/p:Platform=Any CPU'&lt;/span&gt;
          &lt;span class="s"&gt;if ($LASTEXITCODE -ne 0) { throw 'Build failed' }&lt;/span&gt;
      &lt;span class="pi"&gt;-&lt;/span&gt; &lt;span class="na"&gt;name&lt;/span&gt;&lt;span class="pi"&gt;:&lt;/span&gt; &lt;span class="s"&gt;Run existing tests&lt;/span&gt;
        &lt;span class="na"&gt;run&lt;/span&gt;&lt;span class="pi"&gt;:&lt;/span&gt; &lt;span class="pi"&gt;|&lt;/span&gt;
          &lt;span class="s"&gt;$vswhere = Join-Path ${env:ProgramFiles(x86)} 'Microsoft Visual Studio\Installer\vswhere.exe'&lt;/span&gt;
          &lt;span class="s"&gt;$vs = &amp;amp; $vswhere -latest -products '*' -property installationPath&lt;/span&gt;
          &lt;span class="s"&gt;$vstest = Join-Path $vs 'Common7\IDE\Extensions\TestPlatform\vstest.console.exe'&lt;/span&gt;
          &lt;span class="s"&gt;$tests = 'tests\LegacyMvc.Tests\bin\Release\LegacyMvc.Tests.dll'&lt;/span&gt;
          &lt;span class="s"&gt;if (!(Test-Path $vstest)) { throw 'vstest.console.exe not found' }&lt;/span&gt;
          &lt;span class="s"&gt;if (!(Test-Path $tests)) { throw 'Test assembly not found' }&lt;/span&gt;
          &lt;span class="s"&gt;&amp;amp; $vstest $tests '/Logger:trx;LogFileName=tests.trx' '/ResultsDirectory:TestResults'&lt;/span&gt;
          &lt;span class="s"&gt;if ($LASTEXITCODE -ne 0) { throw 'Tests failed' }&lt;/span&gt;
          &lt;span class="s"&gt;[xml]$trx = Get-Content 'TestResults\tests.trx' -Raw&lt;/span&gt;
          &lt;span class="s"&gt;if ([int]$trx.TestRun.ResultSummary.Counters.total -lt 1) { throw 'No tests were discovered' }&lt;/span&gt;
      &lt;span class="pi"&gt;-&lt;/span&gt; &lt;span class="na"&gt;name&lt;/span&gt;&lt;span class="pi"&gt;:&lt;/span&gt; &lt;span class="s"&gt;Create web deployment package&lt;/span&gt;
        &lt;span class="na"&gt;run&lt;/span&gt;&lt;span class="pi"&gt;:&lt;/span&gt; &lt;span class="pi"&gt;|&lt;/span&gt;
          &lt;span class="s"&gt;msbuild src\LegacyMvc\LegacyMvc.csproj /p:Configuration=Release /p:DeployOnBuild=true /p:DeployTarget=Package&lt;/span&gt;
          &lt;span class="s"&gt;if ($LASTEXITCODE -ne 0) { throw 'Packaging failed' }&lt;/span&gt;
          &lt;span class="s"&gt;$zip = Get-ChildItem src\LegacyMvc -Recurse -Filter 'LegacyMvc.zip' | Select-Object -First 1&lt;/span&gt;
          &lt;span class="s"&gt;if (!$zip) { throw 'LegacyMvc.zip not found' }&lt;/span&gt;
          &lt;span class="s"&gt;New-Item -ItemType Directory -Force artifacts | Out-Null&lt;/span&gt;
          &lt;span class="s"&gt;Get-ChildItem $zip.DirectoryName | Copy-Item -Destination artifacts&lt;/span&gt;
      &lt;span class="pi"&gt;-&lt;/span&gt; &lt;span class="na"&gt;name&lt;/span&gt;&lt;span class="pi"&gt;:&lt;/span&gt; &lt;span class="s"&gt;Upload package&lt;/span&gt;
        &lt;span class="na"&gt;uses&lt;/span&gt;&lt;span class="pi"&gt;:&lt;/span&gt; &lt;span class="s"&gt;actions/upload-artifact@v7&lt;/span&gt;
        &lt;span class="na"&gt;with&lt;/span&gt;&lt;span class="pi"&gt;:&lt;/span&gt;
          &lt;span class="na"&gt;name&lt;/span&gt;&lt;span class="pi"&gt;:&lt;/span&gt; &lt;span class="s"&gt;legacy-mvc-package&lt;/span&gt;
          &lt;span class="na"&gt;path&lt;/span&gt;&lt;span class="pi"&gt;:&lt;/span&gt; &lt;span class="s"&gt;artifacts/&lt;/span&gt;
          &lt;span class="na"&gt;if-no-files-found&lt;/span&gt;&lt;span class="pi"&gt;:&lt;/span&gt; &lt;span class="s"&gt;error&lt;/span&gt;
      &lt;span class="pi"&gt;-&lt;/span&gt; &lt;span class="na"&gt;name&lt;/span&gt;&lt;span class="pi"&gt;:&lt;/span&gt; &lt;span class="s"&gt;Upload test results&lt;/span&gt;
        &lt;span class="na"&gt;if&lt;/span&gt;&lt;span class="pi"&gt;:&lt;/span&gt; &lt;span class="s"&gt;always()&lt;/span&gt;
        &lt;span class="na"&gt;uses&lt;/span&gt;&lt;span class="pi"&gt;:&lt;/span&gt; &lt;span class="s"&gt;actions/upload-artifact@v7&lt;/span&gt;
        &lt;span class="na"&gt;with&lt;/span&gt;&lt;span class="pi"&gt;:&lt;/span&gt;
          &lt;span class="na"&gt;name&lt;/span&gt;&lt;span class="pi"&gt;:&lt;/span&gt; &lt;span class="s"&gt;legacy-mvc-test-results&lt;/span&gt;
          &lt;span class="na"&gt;path&lt;/span&gt;&lt;span class="pi"&gt;:&lt;/span&gt; &lt;span class="s"&gt;TestResults/&lt;/span&gt;
          &lt;span class="na"&gt;if-no-files-found&lt;/span&gt;&lt;span class="pi"&gt;:&lt;/span&gt; &lt;span class="s"&gt;warn&lt;/span&gt;
&lt;/code&gt;&lt;/pre&gt;

&lt;/div&gt;



&lt;h2&gt;
  
  
  Make "tests ran" an actual check
&lt;/h2&gt;

&lt;p&gt;The &lt;a href="https://learn.microsoft.com/en-us/visualstudio/test/vstest-console-options?view=vs-2022" rel="noopener noreferrer"&gt;VSTest.Console options&lt;/a&gt; cover the test assembly arguments, &lt;code&gt;/Logger&lt;/code&gt; and &lt;code&gt;/TestAdapterPath&lt;/code&gt;. MSTest, NUnit and xUnit each need their adapter available to the runner; if discovery fails on CI but works locally, &lt;code&gt;/TestAdapterPath&lt;/code&gt; is the first thing to look at. In the sample, MSTest 4.5.2 installed through &lt;code&gt;packages.config&lt;/code&gt; was discovered without it.&lt;/p&gt;

&lt;p&gt;The template reads the TRX file and fails when zero tests were reported, because a test step that exits cleanly after discovering nothing looks green and proves nothing. If the repository has no tests, say so in the handover instead of advertising coverage. And remember what unit tests don't cover: IIS configuration, the real database and browser behavior need their own checks.&lt;/p&gt;

&lt;h2&gt;
  
  
  Inspect the package before trusting it
&lt;/h2&gt;

&lt;p&gt;Download the artifact and open it. The same Microsoft guide explains that &lt;code&gt;SetParameters.xml&lt;/code&gt; is generated from the project and can include connection strings found in &lt;code&gt;web.config&lt;/code&gt;, so this is exactly where a production connection string can leak into an artifact. A zip that exists is a good sign, not proof that it deploys.&lt;/p&gt;

&lt;p&gt;If the project already has reviewed publish settings, you can keep them in a profile under &lt;code&gt;Properties/PublishProfiles/&lt;/code&gt; and select it with &lt;code&gt;PublishProfile&lt;/code&gt;, as Microsoft's &lt;a href="https://learn.microsoft.com/en-us/aspnet/web-forms/overview/deployment/visual-studio-web-deployment/command-line-deployment" rel="noopener noreferrer"&gt;command-line deployment tutorial&lt;/a&gt; does. That tutorial lists three command-line publish methods, &lt;code&gt;MSDeploy&lt;/code&gt;, &lt;code&gt;Package&lt;/code&gt; and &lt;code&gt;FileSystem&lt;/code&gt;; for CI you want the profile to use &lt;code&gt;Package&lt;/code&gt;, not publish straight to a server. It was written for Visual Studio 2012 and notes that building a single web project then also needed &lt;code&gt;/p:VisualStudioVersion&lt;/code&gt;, so if MSBuild cannot find the web application targets, that property is worth checking.&lt;/p&gt;

&lt;h2&gt;
  
  
  Keep production behind a separate gate
&lt;/h2&gt;

&lt;p&gt;No passwords, tokens or production connection strings in the YAML or in committed publish profiles. When you add a deployment job later, point it at a protected environment with required reviewers, which GitHub documents in its &lt;a href="https://docs.github.com/en/actions/reference/workflows-and-actions/deployments-and-environments" rel="noopener noreferrer"&gt;deployments and environments reference&lt;/a&gt;. Naming an environment in YAML does not create the approval rule by itself. Check your plan, too: the same page says that on GitHub Free, Pro and Team, required reviewers "are only available for public repositories", and most legacy business apps live in private ones.&lt;/p&gt;

&lt;p&gt;The first milestone is modest and useful: a package you inspected and test results you can read. Production deployment comes after staging works, access is approved and there is a rollback plan.&lt;/p&gt;

&lt;p&gt;If you have an older ASP.NET MVC or .NET Framework app that needs a build like this, or a focused fix, I offer a fixed-scope service on Upwork: &lt;a href="https://www.upwork.com/services/product/2117249840584640174" rel="noopener noreferrer"&gt;https://www.upwork.com/services/product/2117249840584640174&lt;/a&gt;&lt;/p&gt;

</description>
      <category>dotnet</category>
      <category>githubactions</category>
      <category>devops</category>
      <category>testing</category>
    </item>
    <item>
      <title>Diagnosing "utf8mb3 is not supported" in .NET Framework apps</title>
      <dc:creator>Martin Pisacreta</dc:creator>
      <pubDate>Mon, 02 Nov 2026 14:21:39 +0000</pubDate>
      <link>https://dev.to/martinpisacreta/diagnosing-utf8mb3-is-not-supported-in-net-framework-apps-4n36</link>
      <guid>https://dev.to/martinpisacreta/diagnosing-utf8mb3-is-not-supported-in-net-framework-apps-4n36</guid>
      <description>&lt;p&gt;An existing .NET Framework application moves from MySQL to MariaDB, and a connection or query now fails with a message like &lt;code&gt;Character set 'utf8mb3' is not supported&lt;/code&gt;. Where should the investigation start?&lt;/p&gt;

&lt;p&gt;My answer: with the connector that the running application actually loads, then the server's character set settings, and only after that the schema. Changing columns before you understand the failure can turn a dependency problem into a data migration.&lt;/p&gt;

&lt;p&gt;This is a diagnostic guide built from official documentation, which I re-read on November 2, 2026. I did not reproduce the exception for this article, and the snippets below are diagnostics to adapt, not tested fixes.&lt;/p&gt;

&lt;h2&gt;
  
  
  Why the name utf8mb3 appears at all
&lt;/h2&gt;

&lt;p&gt;Two documented facts fit together here.&lt;/p&gt;

&lt;p&gt;The MariaDB &lt;a href="https://mariadb.com/kb/en/old-mode/" rel="noopener noreferrer"&gt;OLD_MODE documentation&lt;/a&gt; says that from MariaDB 10.6.1, "the main name of the previous 3-byte &lt;code&gt;utf8&lt;/code&gt; character set was changed to &lt;code&gt;utf8mb3&lt;/code&gt;". With the &lt;code&gt;UTF8_IS_UTF8MB3&lt;/code&gt; flag set, &lt;code&gt;utf8&lt;/code&gt; is an alias for &lt;code&gt;utf8mb3&lt;/code&gt;. That flag was on by default until MariaDB 13.1, where the default changed and &lt;code&gt;utf8&lt;/code&gt; became an alias for &lt;code&gt;utf8mb4&lt;/code&gt;.&lt;/p&gt;

&lt;p&gt;The &lt;a href="https://dev.mysql.com/doc/relnotes/connector-net/en/news-8-0-28.html" rel="noopener noreferrer"&gt;Connector/NET 8.0.28 release notes&lt;/a&gt; include this fix: "Connector/NET did not support the &lt;code&gt;utf8mb3&lt;/code&gt; character set" (Bug #105516, Bug #33556024).&lt;/p&gt;

&lt;p&gt;So an application that still ships an older MySql.Data, connected to a MariaDB server that reports &lt;code&gt;utf8mb3&lt;/code&gt;, is a concrete hypothesis worth testing first. It is a hypothesis, not a diagnosis. The same wording can come from other paths, and the evidence below is what confirms or rules it out.&lt;/p&gt;

&lt;h2&gt;
  
  
  Step 1: keep the whole failure
&lt;/h2&gt;

&lt;p&gt;Save the full exception, inner exceptions and stack trace, with credentials removed. Note whether it fails when opening the connection, executing a command or reading results. Write down the .NET Framework target, the MariaDB version, the package references and what changed in the move.&lt;/p&gt;

&lt;h2&gt;
  
  
  Step 2: find the connector assembly that really loads
&lt;/h2&gt;

&lt;p&gt;A package reference tells you what the project asks for. The deployed &lt;code&gt;bin&lt;/code&gt; folder, the GAC or a binding redirect can give you something else. Log this from the running application through your normal logging path, never through a public endpoint:&lt;br&gt;
&lt;/p&gt;

&lt;div class="highlight js-code-highlight"&gt;
&lt;pre class="highlight csharp"&gt;&lt;code&gt;&lt;span class="kt"&gt;var&lt;/span&gt; &lt;span class="n"&gt;asm&lt;/span&gt; &lt;span class="p"&gt;=&lt;/span&gt; &lt;span class="k"&gt;typeof&lt;/span&gt;&lt;span class="p"&gt;(&lt;/span&gt;&lt;span class="n"&gt;MySql&lt;/span&gt;&lt;span class="p"&gt;.&lt;/span&gt;&lt;span class="n"&gt;Data&lt;/span&gt;&lt;span class="p"&gt;.&lt;/span&gt;&lt;span class="n"&gt;MySqlClient&lt;/span&gt;&lt;span class="p"&gt;.&lt;/span&gt;&lt;span class="n"&gt;MySqlConnection&lt;/span&gt;&lt;span class="p"&gt;).&lt;/span&gt;&lt;span class="n"&gt;Assembly&lt;/span&gt;&lt;span class="p"&gt;;&lt;/span&gt;
&lt;span class="n"&gt;log&lt;/span&gt;&lt;span class="p"&gt;.&lt;/span&gt;&lt;span class="nf"&gt;Info&lt;/span&gt;&lt;span class="p"&gt;(&lt;/span&gt;&lt;span class="n"&gt;asm&lt;/span&gt;&lt;span class="p"&gt;.&lt;/span&gt;&lt;span class="n"&gt;FullName&lt;/span&gt;&lt;span class="p"&gt;);&lt;/span&gt;   &lt;span class="c1"&gt;// name and version that actually loaded&lt;/span&gt;
&lt;span class="n"&gt;log&lt;/span&gt;&lt;span class="p"&gt;.&lt;/span&gt;&lt;span class="nf"&gt;Info&lt;/span&gt;&lt;span class="p"&gt;(&lt;/span&gt;&lt;span class="n"&gt;asm&lt;/span&gt;&lt;span class="p"&gt;.&lt;/span&gt;&lt;span class="n"&gt;Location&lt;/span&gt;&lt;span class="p"&gt;);&lt;/span&gt;   &lt;span class="c1"&gt;// file that was loaded&lt;/span&gt;
&lt;span class="n"&gt;log&lt;/span&gt;&lt;span class="p"&gt;.&lt;/span&gt;&lt;span class="nf"&gt;Info&lt;/span&gt;&lt;span class="p"&gt;(&lt;/span&gt;&lt;span class="n"&gt;asm&lt;/span&gt;&lt;span class="p"&gt;.&lt;/span&gt;&lt;span class="n"&gt;CodeBase&lt;/span&gt;&lt;span class="p"&gt;);&lt;/span&gt;   &lt;span class="c1"&gt;// original location before shadow copying&lt;/span&gt;
&lt;/code&gt;&lt;/pre&gt;

&lt;/div&gt;



&lt;p&gt;Microsoft documents &lt;a href="https://learn.microsoft.com/en-us/dotnet/api/system.reflection.assembly.location?view=netframework-4.8.1" rel="noopener noreferrer"&gt;Assembly.Location&lt;/a&gt; as the location of the loaded file that contains the manifest. On .NET Framework, if the file was shadow-copied, Location points to the copy, and the same page says to use &lt;code&gt;CodeBase&lt;/code&gt; for the location before shadow copying. ASP.NET copies Bin assemblies into its Temporary ASP.NET Files directory by default (&lt;a href="https://learn.microsoft.com/en-us/dotnet/api/system.web.configuration.hostingenvironmentsection.shadowcopybinassemblies" rel="noopener noreferrer"&gt;shadowCopyBinAssemblies&lt;/a&gt; defaults to true), so a temporary path in Location is expected and not a sign of a bad deploy by itself.&lt;/p&gt;

&lt;p&gt;If &lt;code&gt;FullName&lt;/code&gt; shows a MySql.Data version below 8.0.28, you have a strong lead.&lt;/p&gt;

&lt;h2&gt;
  
  
  Step 3: check binding redirects and publisher policy
&lt;/h2&gt;

&lt;p&gt;Microsoft's &lt;a href="https://learn.microsoft.com/en-us/dotnet/framework/configure-apps/redirect-assembly-versions" rel="noopener noreferrer"&gt;assembly redirection guide&lt;/a&gt; explains that the application configuration file, publisher policy and machine configuration can all redirect assembly versions.&lt;/p&gt;

&lt;p&gt;Open the deployed &lt;code&gt;web.config&lt;/code&gt; and look for the &lt;code&gt;assemblyBinding&lt;/code&gt; entry for MySql.Data. Compare it with source control and with the release artifact. A redirect pinned to an old version, or a forgotten copy in the GAC, explains why updating the NuGet package changed nothing. Don't paste a redirect from a forum post: the right &lt;code&gt;oldVersion&lt;/code&gt; range and &lt;code&gt;newVersion&lt;/code&gt; depend on what the app actually references.&lt;/p&gt;

&lt;h2&gt;
  
  
  Step 4: capture the server's view
&lt;/h2&gt;

&lt;p&gt;From an admin client, outside the failing app:&lt;br&gt;
&lt;/p&gt;

&lt;div class="highlight js-code-highlight"&gt;
&lt;pre class="highlight sql"&gt;&lt;code&gt;&lt;span class="k"&gt;SELECT&lt;/span&gt; &lt;span class="k"&gt;VERSION&lt;/span&gt;&lt;span class="p"&gt;();&lt;/span&gt;
&lt;span class="k"&gt;SHOW&lt;/span&gt; &lt;span class="n"&gt;VARIABLES&lt;/span&gt; &lt;span class="k"&gt;LIKE&lt;/span&gt; &lt;span class="s1"&gt;'character_set%'&lt;/span&gt;&lt;span class="p"&gt;;&lt;/span&gt;
&lt;span class="k"&gt;SHOW&lt;/span&gt; &lt;span class="n"&gt;VARIABLES&lt;/span&gt; &lt;span class="k"&gt;LIKE&lt;/span&gt; &lt;span class="s1"&gt;'collation%'&lt;/span&gt;&lt;span class="p"&gt;;&lt;/span&gt;
&lt;/code&gt;&lt;/pre&gt;

&lt;/div&gt;



&lt;p&gt;The &lt;a href="https://mariadb.com/kb/en/show-variables/" rel="noopener noreferrer"&gt;SHOW VARIABLES syntax&lt;/a&gt; accepts a &lt;code&gt;LIKE 'pattern'&lt;/code&gt; with &lt;code&gt;%&lt;/code&gt; wildcards. Then check the affected tables. The MariaDB &lt;a href="https://mariadb.com/kb/en/setting-character-sets-and-collations/" rel="noopener noreferrer"&gt;character set page&lt;/a&gt; uses both of these:&lt;br&gt;
&lt;/p&gt;

&lt;div class="highlight js-code-highlight"&gt;
&lt;pre class="highlight sql"&gt;&lt;code&gt;&lt;span class="k"&gt;SHOW&lt;/span&gt; &lt;span class="k"&gt;CREATE&lt;/span&gt; &lt;span class="k"&gt;TABLE&lt;/span&gt; &lt;span class="n"&gt;your_table&lt;/span&gt;&lt;span class="err"&gt;\&lt;/span&gt;&lt;span class="k"&gt;G&lt;/span&gt;
&lt;span class="k"&gt;SELECT&lt;/span&gt; &lt;span class="k"&gt;TABLE_NAME&lt;/span&gt;&lt;span class="p"&gt;,&lt;/span&gt; &lt;span class="k"&gt;COLUMN_NAME&lt;/span&gt;&lt;span class="p"&gt;,&lt;/span&gt; &lt;span class="k"&gt;CHARACTER_SET_NAME&lt;/span&gt;&lt;span class="p"&gt;,&lt;/span&gt; &lt;span class="k"&gt;COLLATION_NAME&lt;/span&gt;
&lt;span class="k"&gt;FROM&lt;/span&gt; &lt;span class="n"&gt;INFORMATION_SCHEMA&lt;/span&gt;&lt;span class="p"&gt;.&lt;/span&gt;&lt;span class="n"&gt;COLUMNS&lt;/span&gt;
&lt;span class="k"&gt;WHERE&lt;/span&gt; &lt;span class="n"&gt;TABLE_SCHEMA&lt;/span&gt; &lt;span class="o"&gt;=&lt;/span&gt; &lt;span class="s1"&gt;'your_db'&lt;/span&gt;&lt;span class="p"&gt;;&lt;/span&gt;
&lt;/code&gt;&lt;/pre&gt;

&lt;/div&gt;



&lt;p&gt;Don't infer each column from one server variable. Record what you see next to the exception and the loaded connector version.&lt;/p&gt;

&lt;h2&gt;
  
  
  Step 5: make one bounded change
&lt;/h2&gt;

&lt;p&gt;If the evidence points to an old Connector/NET, the smallest change is updating MySql.Data to a version that includes the 8.0.28 fix and that still supports your framework target, then fixing the redirect so that version is the one that loads. Do it in a branch, deploy to staging, log the loaded assembly again and repeat the failing operation. Keep schema changes out of this experiment so the result means something.&lt;/p&gt;

&lt;p&gt;Switching providers is a different, bigger scope. The &lt;a href="https://mysqlconnector.net/tutorials/migrating-from-connector-net/" rel="noopener noreferrer"&gt;MySqlConnector migration guide&lt;/a&gt; says to change &lt;code&gt;using MySql.Data.MySqlClient;&lt;/code&gt; to &lt;code&gt;using MySqlConnector;&lt;/code&gt;, and that &lt;code&gt;MySqlClientFactory&lt;/code&gt; is named &lt;code&gt;MySqlConnectorFactory&lt;/code&gt;, which matters if the app uses &lt;code&gt;DbProviderFactories&lt;/code&gt;. Its &lt;a href="https://mysqlconnector.net/connection-options/" rel="noopener noreferrer"&gt;connection options&lt;/a&gt; also change behavior: MySqlConnector "always uses &lt;code&gt;utf8mb4&lt;/code&gt;" and ignores &lt;code&gt;CharacterSet&lt;/code&gt;, &lt;code&gt;ConnectionReset&lt;/code&gt; defaults to &lt;code&gt;true&lt;/code&gt;, and &lt;code&gt;IgnoreCommandTransaction&lt;/code&gt; defaults to &lt;code&gt;false&lt;/code&gt;. Audit those against the app's code and ORM before calling it a drop-in replacement.&lt;/p&gt;

&lt;p&gt;Converting stored data to &lt;code&gt;utf8mb4&lt;/code&gt; is a third decision with its own risks, such as index length limits and table rebuilds. It does not belong in the same change, and it needs a rehearsal on a restored copy.&lt;/p&gt;

&lt;h2&gt;
  
  
  Step 6: verify and keep a way back
&lt;/h2&gt;

&lt;p&gt;Before touching anything, define the checks: the original failing operation, a round trip of text with accents and emoji, and the main user flow that hit the error. Keep the previous artifact and configuration so rolling back is a redeploy, and after a rollback check startup and the loaded assembly again.&lt;/p&gt;

&lt;p&gt;The useful outcome is a short chain of evidence: the failing operation, the connector that loaded, the redirect, what the server reports, one change and its measured result.&lt;/p&gt;

&lt;p&gt;If you have an older .NET Framework or ASP.NET MVC app with a problem like this, I offer a fixed-scope diagnosis and fix on Upwork: &lt;a href="https://www.upwork.com/services/product/2117249840584640174" rel="noopener noreferrer"&gt;https://www.upwork.com/services/product/2117249840584640174&lt;/a&gt;&lt;/p&gt;

</description>
      <category>dotnet</category>
      <category>csharp</category>
      <category>mysql</category>
      <category>database</category>
    </item>
    <item>
      <title>252 x402 receipts, but how many confirmed customers?</title>
      <dc:creator>Martin Pisacreta</dc:creator>
      <pubDate>Mon, 02 Nov 2026 12:19:42 +0000</pubDate>
      <link>https://dev.to/martinpisacreta/252-x402-receipts-but-how-many-confirmed-customers-4akg</link>
      <guid>https://dev.to/martinpisacreta/252-x402-receipts-but-how-many-confirmed-customers-4akg</guid>
      <description>&lt;p&gt;My store's receipt feed records 252 settled sales. That sounds like traction until I ask a different question: who paid, and what evidence shows they wanted the service?&lt;/p&gt;

&lt;p&gt;Of those sales, 56 are marked as payments from wallets controlled by the same operator. The remaining 196 come from 16 external payer addresses. Neither figure establishes sixteen customers. At the time of this review, I have no confirmed genuine customer among these payments.&lt;/p&gt;

&lt;p&gt;This is why I publish receipts and separate settlement evidence from customer evidence. A payment can be real while the demand it appears to represent remains unconfirmed.&lt;/p&gt;

&lt;h2&gt;
  
  
  A timestamped snapshot
&lt;/h2&gt;

&lt;p&gt;I fetched the public &lt;a href="https://tienda.textokit.store/receipts.json" rel="noopener noreferrer"&gt;receipts.json&lt;/a&gt; with curl on November 2, 2026. Retrieval completed at &lt;strong&gt;12:01:49.698082 UTC&lt;/strong&gt;; the feed's &lt;code&gt;generatedAt&lt;/code&gt; value was &lt;code&gt;2026-11-02T12:01:49.636Z&lt;/code&gt;. These figures describe that snapshot, not a continuously updated claim.&lt;/p&gt;

&lt;p&gt;I counted the sale rows rather than relying only on the summary fields:&lt;/p&gt;

&lt;div class="table-wrapper-paragraph"&gt;&lt;table&gt;
&lt;thead&gt;
&lt;tr&gt;
&lt;th&gt;Receipt category&lt;/th&gt;
&lt;th&gt;Sales&lt;/th&gt;
&lt;/tr&gt;
&lt;/thead&gt;
&lt;tbody&gt;
&lt;tr&gt;
&lt;td&gt;All recorded sales&lt;/td&gt;
&lt;td&gt;252&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;&lt;code&gt;ownWallet: true&lt;/code&gt;&lt;/td&gt;
&lt;td&gt;56&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;&lt;code&gt;ownWallet: false&lt;/code&gt;&lt;/td&gt;
&lt;td&gt;196&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;Missing or non-boolean ownership classification&lt;/td&gt;
&lt;td&gt;0&lt;/td&gt;
&lt;/tr&gt;
&lt;/tbody&gt;
&lt;/table&gt;&lt;/div&gt;

&lt;p&gt;The computed totals agree with the feed's declared sales and own-wallet totals. Lowercasing external payer addresses before grouping produces 16 distinct addresses, also matching its declared &lt;code&gt;externalPayers&lt;/code&gt; value.&lt;/p&gt;

&lt;p&gt;The feed separately contains a purchases array. I excluded that array from the sales calculation. Mixing outgoing purchases with incoming sales would answer a different question.&lt;/p&gt;

&lt;h2&gt;
  
  
  How the external payments are distributed
&lt;/h2&gt;

&lt;p&gt;Here is the complete distribution of the 196 external sale rows. Addresses are abbreviated for readability; the public feed contains their full values.&lt;/p&gt;

&lt;div class="table-wrapper-paragraph"&gt;&lt;table&gt;
&lt;thead&gt;
&lt;tr&gt;
&lt;th&gt;External payer&lt;/th&gt;
&lt;th&gt;Recorded sales&lt;/th&gt;
&lt;/tr&gt;
&lt;/thead&gt;
&lt;tbody&gt;
&lt;tr&gt;
&lt;td&gt;&lt;code&gt;0xe827…50a0&lt;/code&gt;&lt;/td&gt;
&lt;td&gt;44&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;&lt;code&gt;0x2465…a1b1&lt;/code&gt;&lt;/td&gt;
&lt;td&gt;37&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;&lt;code&gt;0xb83b…75cc&lt;/code&gt;&lt;/td&gt;
&lt;td&gt;35&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;&lt;code&gt;0x38b6…8b9f&lt;/code&gt;&lt;/td&gt;
&lt;td&gt;27&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;&lt;code&gt;0xec2a…d034&lt;/code&gt;&lt;/td&gt;
&lt;td&gt;22&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;&lt;code&gt;0x54e1…f4e0&lt;/code&gt;&lt;/td&gt;
&lt;td&gt;10&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;&lt;code&gt;0xc9c7…1670&lt;/code&gt;&lt;/td&gt;
&lt;td&gt;6&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;&lt;code&gt;0x7706…4121&lt;/code&gt;&lt;/td&gt;
&lt;td&gt;3&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;&lt;code&gt;0x6031…c2d1&lt;/code&gt;&lt;/td&gt;
&lt;td&gt;3&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;&lt;code&gt;0xaa34…1db3&lt;/code&gt;&lt;/td&gt;
&lt;td&gt;2&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;&lt;code&gt;0x8ec9…9d15&lt;/code&gt;&lt;/td&gt;
&lt;td&gt;2&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;&lt;code&gt;0x097e…475a&lt;/code&gt;&lt;/td&gt;
&lt;td&gt;1&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;&lt;code&gt;0xa1d1…1978&lt;/code&gt;&lt;/td&gt;
&lt;td&gt;1&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;&lt;code&gt;0x7722…5e4a&lt;/code&gt;&lt;/td&gt;
&lt;td&gt;1&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;&lt;code&gt;0xe629…4b54&lt;/code&gt;&lt;/td&gt;
&lt;td&gt;1&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;&lt;code&gt;0x03ca…55f3&lt;/code&gt;&lt;/td&gt;
&lt;td&gt;1&lt;/td&gt;
&lt;/tr&gt;
&lt;/tbody&gt;
&lt;/table&gt;&lt;/div&gt;

&lt;p&gt;These are address groups, not identified people or independent organizations. One operator can control multiple addresses, and one address can represent automated infrastructure. The JSON does not resolve those relationships.&lt;/p&gt;

&lt;h2&gt;
  
  
  Why I expose own-wallet payments
&lt;/h2&gt;

&lt;p&gt;The feed explains that wallets controlled by the same operator, including agent, auxiliary, and creator wallets, are flagged &lt;code&gt;ownWallet: true&lt;/code&gt;. That prevents self-settlements from being presented as outside customers.&lt;/p&gt;

&lt;p&gt;The flag is an operator disclosure. It is not something a USDC transfer automatically reveals. A blockchain address does not carry a trustworthy label saying who controls it or why it paid.&lt;/p&gt;

&lt;p&gt;Conversely, &lt;code&gt;ownWallet: false&lt;/code&gt; means the feed has not classified that payment as coming from an operator-controlled wallet. It does not mean a buyer independently discovered the product, needed its output, or intends to return.&lt;/p&gt;

&lt;p&gt;Keeping that distinction visible makes the receipts useful without turning them into a customer counter.&lt;/p&gt;

&lt;h2&gt;
  
  
  What the activity suggests
&lt;/h2&gt;

&lt;p&gt;Earlier operational review identified &lt;code&gt;0xec2a…d034&lt;/code&gt; repeating the exact published &lt;code&gt;inputExample&lt;/code&gt; for &lt;code&gt;normativa-ar&lt;/code&gt;. That looks consistent with a monitor or probe. This snapshot records 22 sales from that address, including five for &lt;code&gt;normativa-ar&lt;/code&gt;; it does not contain the request inputs needed to reproduce the example comparison. I therefore treat the monitor label as an observation from that earlier review, not a conclusion proved by this JSON.&lt;/p&gt;

&lt;p&gt;The earlier review also identified broad one-off payment patterns across many sellers for &lt;code&gt;0xaa34…1db3&lt;/code&gt;, &lt;code&gt;0x8ec9…9d15&lt;/code&gt; and &lt;code&gt;0x54e1…f4e0&lt;/code&gt;, consistent with sweepers. In this feed, &lt;code&gt;0x54e1…f4e0&lt;/code&gt; paid ten routes within about three and a half hours on the morning of November 1. Their receipts here do not establish repeat product demand.&lt;/p&gt;

&lt;p&gt;The four largest external groups tell a similar story. &lt;code&gt;0xe827…50a0&lt;/code&gt;, &lt;code&gt;0x2465…a1b1&lt;/code&gt;, &lt;code&gt;0xb83b…75cc&lt;/code&gt; and &lt;code&gt;0x38b6…8b9f&lt;/code&gt; account for 143 of the 196 external sales. In this feed, all 143 fall within about three and a half hours on October 25, between 18:33 and 22:11 UTC, at 0.001 or 0.002 USDC each. During that window an earlier review watched them take turns on the cheapest routes, roughly one purchase every 25 seconds. That looks like one automated buyer spread across several addresses, not four customers.&lt;/p&gt;

&lt;p&gt;&lt;code&gt;0xc9c7…1670&lt;/code&gt; is a known payer from earlier operational context. Familiarity alone does not establish a confirmed customer either. The other addresses remain unclassified here. Unknown intent should stay unknown.&lt;/p&gt;

&lt;h2&gt;
  
  
  What an on-chain check can establish
&lt;/h2&gt;

&lt;p&gt;The feed declares Base through &lt;code&gt;eip155:8453&lt;/code&gt; and supplies transaction hashes. Those hashes provide a route to checking transaction receipts and USDC &lt;code&gt;Transfer&lt;/code&gt; logs: token contract, sender, recipient, amount, and transaction success.&lt;/p&gt;

&lt;p&gt;That check can corroborate a token movement. It cannot establish satisfaction, useful output, independent wallet ownership, or purchase intent. The store's receipt record supplies application context; the operator supplies ownership disclosures; customer confirmation requires additional evidence.&lt;/p&gt;

&lt;p&gt;For this article, I fetched and parsed the JSON, grouped payer addresses, and checked that the category counts reconcile. I did not independently inspect Base transaction logs, reproduce request inputs, or verify the earlier payer classifications. I saved the retrieved snapshot with its computed breakdown for review.&lt;/p&gt;

&lt;h2&gt;
  
  
  Payment indexes answer a narrower question
&lt;/h2&gt;

&lt;p&gt;An index counting settlements can report payment activity accurately while saying little about customers. Self-payments, monitoring calls, broad sweeps, and genuine purchases can all produce payment records. Counting those records does not distinguish their purpose.&lt;/p&gt;

&lt;p&gt;My defensible statement is therefore specific: 252 recorded sales, 56 marked own-wallet sales, and 196 external sales across 16 payer addresses, with no genuine customer confirmed at review time. The receipts make the accounting inspectable. They do not supply the missing evidence of demand.&lt;/p&gt;

&lt;p&gt;If you want a pay-per-call API that agents can pay in USDC, with receipts you can audit like these, I offer it as a fixed-price project on Upwork: &lt;a href="https://www.upwork.com/services/product/2114214673238319567" rel="noopener noreferrer"&gt;https://www.upwork.com/services/product/2114214673238319567&lt;/a&gt;&lt;/p&gt;

</description>
      <category>x402</category>
      <category>ai</category>
      <category>web3</category>
      <category>api</category>
    </item>
    <item>
      <title>Your Stripe inbox prevents duplicate updates. What about duplicate emails?</title>
      <dc:creator>Martin Pisacreta</dc:creator>
      <pubDate>Mon, 02 Nov 2026 11:37:56 +0000</pubDate>
      <link>https://dev.to/martinpisacreta/your-stripe-inbox-prevents-duplicate-updates-what-about-duplicate-emails-2738</link>
      <guid>https://dev.to/martinpisacreta/your-stripe-inbox-prevents-duplicate-updates-what-about-duplicate-emails-2738</guid>
      <description>&lt;p&gt;A reader raised a painful case on my earlier Stripe webhook article: a duplicate &lt;code&gt;invoice.payment_failed&lt;/code&gt; delivery that sends two card-declined emails. Preventing a duplicate database update does not automatically prevent that second email. The important question is which effects share the transaction that claims the event.&lt;/p&gt;

&lt;p&gt;My &lt;a href="https://dev.to/martinpisacreta/how-i-make-stripe-webhooks-idempotent-in-aspnet-core-2040"&gt;previous article&lt;/a&gt; describes the inbox approach. This follow-up examines its boundary, using commit &lt;code&gt;f5351e708c802a9f428250002f7fd5b86668834b&lt;/code&gt; of &lt;a href="https://github.com/martinPisacreta/stripe-webhooks-dotnet/tree/f5351e708c802a9f428250002f7fd5b86668834b" rel="noopener noreferrer"&gt;stripe-webhooks-dotnet&lt;/a&gt;.&lt;/p&gt;

&lt;p&gt;First, the scope: this sample does not implement failed-invoice emails. Its payment handler supports successful payment intents and refunded charges. The reader's scenario is an extension to consider, not a feature demonstrated by this repository.&lt;/p&gt;

&lt;h2&gt;
  
  
  The inbox and the worker solve different races
&lt;/h2&gt;

&lt;p&gt;The &lt;a href="https://github.com/martinPisacreta/stripe-webhooks-dotnet/blob/f5351e708c802a9f428250002f7fd5b86668834b/README.md#how-it-works" rel="noopener noreferrer"&gt;repository overview&lt;/a&gt; describes insert-first deduplication at receipt: a unique event identifier prevents repeated deliveries from creating multiple inbox records. The request stores the event before acknowledging it and offers its identifier to a bounded queue. A sweeper recovers pending work.&lt;/p&gt;

&lt;p&gt;That still leaves the worker boundary. A pending identifier can reach processing more than once, so receipt deduplication alone is insufficient.&lt;/p&gt;

&lt;p&gt;In &lt;a href="https://github.com/martinPisacreta/stripe-webhooks-dotnet/blob/f5351e708c802a9f428250002f7fd5b86668834b/src/StripeWebhooks.Api/Webhooks/EventProcessor.cs#L36" rel="noopener noreferrer"&gt;EventProcessor.cs&lt;/a&gt;, processing begins a transaction:&lt;br&gt;
&lt;/p&gt;

&lt;div class="highlight js-code-highlight"&gt;
&lt;pre class="highlight csharp"&gt;&lt;code&gt;&lt;span class="k"&gt;await&lt;/span&gt; &lt;span class="k"&gt;using&lt;/span&gt; &lt;span class="nn"&gt;var&lt;/span&gt; &lt;span class="n"&gt;transaction&lt;/span&gt; &lt;span class="p"&gt;=&lt;/span&gt; &lt;span class="k"&gt;await&lt;/span&gt; &lt;span class="n"&gt;db&lt;/span&gt;&lt;span class="p"&gt;.&lt;/span&gt;&lt;span class="n"&gt;Database&lt;/span&gt;&lt;span class="p"&gt;.&lt;/span&gt;&lt;span class="nf"&gt;BeginTransactionAsync&lt;/span&gt;&lt;span class="p"&gt;(&lt;/span&gt;&lt;span class="n"&gt;cancellationToken&lt;/span&gt;&lt;span class="p"&gt;);&lt;/span&gt;
&lt;/code&gt;&lt;/pre&gt;

&lt;/div&gt;



&lt;p&gt;Inside it, a conditional update changes the matching inbox row from Pending to Processed. The returned affected-row count determines whether this invocation obtained the claim. An invocation that changed no row stops here:&lt;br&gt;
&lt;/p&gt;

&lt;div class="highlight js-code-highlight"&gt;
&lt;pre class="highlight csharp"&gt;&lt;code&gt;&lt;span class="k"&gt;if&lt;/span&gt; &lt;span class="p"&gt;(&lt;/span&gt;&lt;span class="n"&gt;claimed&lt;/span&gt; &lt;span class="p"&gt;==&lt;/span&gt; &lt;span class="m"&gt;0&lt;/span&gt;&lt;span class="p"&gt;)&lt;/span&gt;
&lt;span class="p"&gt;{&lt;/span&gt;
    &lt;span class="n"&gt;logger&lt;/span&gt;&lt;span class="p"&gt;.&lt;/span&gt;&lt;span class="nf"&gt;AlreadyHandled&lt;/span&gt;&lt;span class="p"&gt;(&lt;/span&gt;&lt;span class="n"&gt;eventId&lt;/span&gt;&lt;span class="p"&gt;);&lt;/span&gt;
    &lt;span class="k"&gt;return&lt;/span&gt;&lt;span class="p"&gt;;&lt;/span&gt;
&lt;span class="p"&gt;}&lt;/span&gt;
&lt;/code&gt;&lt;/pre&gt;

&lt;/div&gt;



&lt;p&gt;Marking the row Processed happens before the handler, but inside the transaction. It is not a committed declaration that the handler has already succeeded.&lt;/p&gt;

&lt;h2&gt;
  
  
  The local changes commit with the claim
&lt;/h2&gt;

&lt;p&gt;After claiming the row, the processor parses the stored payload and invokes the handler. These &lt;a href="https://github.com/martinPisacreta/stripe-webhooks-dotnet/blob/f5351e708c802a9f428250002f7fd5b86668834b/src/StripeWebhooks.Api/Webhooks/EventProcessor.cs#L59" rel="noopener noreferrer"&gt;lines&lt;/a&gt; complete the successful path:&lt;br&gt;
&lt;/p&gt;

&lt;div class="highlight js-code-highlight"&gt;
&lt;pre class="highlight csharp"&gt;&lt;code&gt;&lt;span class="n"&gt;row&lt;/span&gt;&lt;span class="p"&gt;.&lt;/span&gt;&lt;span class="n"&gt;Outcome&lt;/span&gt; &lt;span class="p"&gt;=&lt;/span&gt; &lt;span class="k"&gt;await&lt;/span&gt; &lt;span class="n"&gt;handler&lt;/span&gt;&lt;span class="p"&gt;.&lt;/span&gt;&lt;span class="nf"&gt;HandleAsync&lt;/span&gt;&lt;span class="p"&gt;(&lt;/span&gt;&lt;span class="n"&gt;stripeEvent&lt;/span&gt;&lt;span class="p"&gt;,&lt;/span&gt; &lt;span class="n"&gt;cancellationToken&lt;/span&gt;&lt;span class="p"&gt;);&lt;/span&gt;

&lt;span class="k"&gt;await&lt;/span&gt; &lt;span class="n"&gt;db&lt;/span&gt;&lt;span class="p"&gt;.&lt;/span&gt;&lt;span class="nf"&gt;SaveChangesAsync&lt;/span&gt;&lt;span class="p"&gt;(&lt;/span&gt;&lt;span class="n"&gt;cancellationToken&lt;/span&gt;&lt;span class="p"&gt;);&lt;/span&gt;
&lt;span class="k"&gt;await&lt;/span&gt; &lt;span class="n"&gt;transaction&lt;/span&gt;&lt;span class="p"&gt;.&lt;/span&gt;&lt;span class="nf"&gt;CommitAsync&lt;/span&gt;&lt;span class="p"&gt;(&lt;/span&gt;&lt;span class="n"&gt;cancellationToken&lt;/span&gt;&lt;span class="p"&gt;);&lt;/span&gt;
&lt;/code&gt;&lt;/pre&gt;

&lt;/div&gt;



&lt;p&gt;The implemented handler changes payment entities through its database context. If handling throws before commit, transaction disposal rolls back the claim and those local changes. The processor then clears tracked state and records failure information for retry.&lt;/p&gt;

&lt;p&gt;That explains the guarantee's useful boundary: the claim and the payment update land together. It does not mean every possible action added inside &lt;code&gt;HandleAsync&lt;/code&gt; participates in that transaction.&lt;/p&gt;

&lt;h2&gt;
  
  
  Sending an email introduces another system
&lt;/h2&gt;

&lt;p&gt;Suppose we extend the handler to send a card-declined email before committing. The email provider accepts it, then the database commit fails. The inbox claim rolls back, leaving the event eligible for retry. The email provider has no reason to undo its accepted message. A retry can send another email.&lt;/p&gt;

&lt;p&gt;Moving the send after commit creates another gap. The process might stop after committing the payment change but before sending. On retry, the committed inbox claim tells the processor to skip the event. The notification can be lost.&lt;/p&gt;

&lt;p&gt;These are hypothetical failure sequences for an extension. The repository contains no email sender that I exercised, and its transaction cannot demonstrate email-delivery guarantees.&lt;/p&gt;

&lt;h2&gt;
  
  
  An outbox would persist the intent
&lt;/h2&gt;

&lt;p&gt;For that extension, I would save a notification intent in an outbox table within the same transaction as the inbox claim and business update. A separate dispatcher would read committed intents and attempt delivery. This is a proposed design; an outbox and its dispatcher are not implemented in this sample.&lt;/p&gt;

&lt;p&gt;An outbox addresses the gap between saving the business change and remembering that notification work exists. It does not by itself guarantee one external email. A dispatcher can stop after the provider accepts a message but before recording success locally.&lt;/p&gt;

&lt;p&gt;Where the provider supports suitable idempotency, the dispatcher should reuse a stable notification key across retries. Otherwise, the design needs an explicit policy for uncertain outcomes and possible duplicates. I would also define whether two different event IDs should create one notification for the same business occurrence. Event-ID deduplication alone does not answer that product question.&lt;/p&gt;

&lt;h2&gt;
  
  
  What the tests establish
&lt;/h2&gt;

&lt;p&gt;The &lt;a href="https://github.com/martinPisacreta/stripe-webhooks-dotnet/blob/f5351e708c802a9f428250002f7fd5b86668834b/tests/StripeWebhooks.Tests/IdempotencyTests.cs#L10" rel="noopener noreferrer"&gt;duplicate-delivery test&lt;/a&gt; posts the same payload again after the first delivery is handled. Its visible assertions check accepted and duplicate acknowledgments and one stored event row.&lt;/p&gt;

&lt;p&gt;The README also documents concurrency tests: ten rounds of twenty-five parallel deliveries, repeated worker processing, and concurrent claims of one pending event. It describes handler-call counting and injected failures to distinguish repeated application from an unchanged final state. Those are local-processing checks, not email tests.&lt;/p&gt;

&lt;p&gt;For this article, I fetched public files with curl at the fixed commit and inspected the processor, payment handler, visible idempotency-test excerpts, and README. I ran no build, &lt;code&gt;dotnet test&lt;/code&gt;, Stripe CLI session, email-provider request, or crash experiment. SQLite serializes writers; the documented tests do not establish another database server's locking behavior.&lt;/p&gt;

&lt;p&gt;The extension needs its own evidence: durable notification creation, dispatcher recovery, and behavior when provider acceptance has an uncertain result. Those checks would support claims about notifications that the existing payment tests cannot establish.&lt;/p&gt;

&lt;p&gt;If you need Stripe webhooks handled this way in your ASP.NET Core app, I offer it as a fixed-price project on Upwork: &lt;a href="https://www.upwork.com/services/product/2114777474333655338" rel="noopener noreferrer"&gt;https://www.upwork.com/services/product/2114777474333655338&lt;/a&gt;&lt;/p&gt;

</description>
      <category>dotnet</category>
      <category>csharp</category>
      <category>stripe</category>
      <category>webdev</category>
    </item>
    <item>
      <title>When a document assistant refuses: quota, context, and the missing model call</title>
      <dc:creator>Martin Pisacreta</dc:creator>
      <pubDate>Mon, 02 Nov 2026 11:18:56 +0000</pubDate>
      <link>https://dev.to/martinpisacreta/when-a-document-assistant-refuses-quota-context-and-the-missing-model-call-2o78</link>
      <guid>https://dev.to/martinpisacreta/when-a-document-assistant-refuses-quota-context-and-the-missing-model-call-2o78</guid>
      <description>&lt;p&gt;An assistant over your documents needs a useful answer when the documents cannot help. In my ASP.NET Core sample, that means returning a refusal instead of asking the model to improvise. But the refusal has consequences beyond the response text: it consumes an admitted question and, when the request completes normally, becomes part of the conversation history.&lt;/p&gt;

&lt;p&gt;My &lt;a href="https://dev.to/martinpisacreta/how-i-validate-cited-answers-in-an-aspnet-core-8-assistant-48f5"&gt;previous article&lt;/a&gt; explains retrieval and citation checks. Here I follow the state changes around an unsuccessful answer. The sample contains three synthetic cooking lessons, so its behavior is small enough to inspect directly.&lt;/p&gt;

&lt;p&gt;All implementation links below point to commit &lt;code&gt;a540d6fbf57c0df4be35eafac076b2688592cdb4&lt;/code&gt; of &lt;a href="https://github.com/martinPisacreta/cited-assistant-dotnet/tree/a540d6fbf57c0df4be35eafac076b2688592cdb4" rel="noopener noreferrer"&gt;cited-assistant-dotnet&lt;/a&gt;.&lt;/p&gt;

&lt;h2&gt;
  
  
  Admission happens before evidence lookup
&lt;/h2&gt;

&lt;p&gt;In &lt;a href="https://github.com/martinPisacreta/cited-assistant-dotnet/blob/a540d6fbf57c0df4be35eafac076b2688592cdb4/ChefService.cs#L43" rel="noopener noreferrer"&gt;ChefService.cs&lt;/a&gt;, the service reserves a question before retrieving a passage:&lt;br&gt;
&lt;/p&gt;

&lt;div class="highlight js-code-highlight"&gt;
&lt;pre class="highlight csharp"&gt;&lt;code&gt;&lt;span class="kt"&gt;var&lt;/span&gt; &lt;span class="n"&gt;access&lt;/span&gt; &lt;span class="p"&gt;=&lt;/span&gt; &lt;span class="n"&gt;subscriptions&lt;/span&gt;&lt;span class="p"&gt;.&lt;/span&gt;&lt;span class="nf"&gt;TryReserveQuestion&lt;/span&gt;&lt;span class="p"&gt;(&lt;/span&gt;&lt;span class="n"&gt;memberId&lt;/span&gt;&lt;span class="p"&gt;);&lt;/span&gt;
&lt;span class="k"&gt;if&lt;/span&gt; &lt;span class="p"&gt;(!&lt;/span&gt;&lt;span class="n"&gt;access&lt;/span&gt;&lt;span class="p"&gt;.&lt;/span&gt;&lt;span class="n"&gt;Allowed&lt;/span&gt;&lt;span class="p"&gt;)&lt;/span&gt; &lt;span class="k"&gt;throw&lt;/span&gt; &lt;span class="k"&gt;new&lt;/span&gt; &lt;span class="nf"&gt;ChefAccessException&lt;/span&gt;&lt;span class="p"&gt;(&lt;/span&gt;&lt;span class="n"&gt;access&lt;/span&gt;&lt;span class="p"&gt;.&lt;/span&gt;&lt;span class="n"&gt;Reason&lt;/span&gt;&lt;span class="p"&gt;);&lt;/span&gt;
&lt;/code&gt;&lt;/pre&gt;

&lt;/div&gt;



&lt;p&gt;Only after admission does the service find or create the conversation history and ask the lesson library for evidence. That ordering matters: a question can use an allowance even when no model call follows.&lt;/p&gt;

&lt;p&gt;The allowance measures admitted questions, not successful answers or provider requests. Someone integrating this sample should make that policy visible to users. Otherwise, a refusal that saved an inference call could still produce a confusing allowance decrease.&lt;/p&gt;

&lt;p&gt;The repository documents the distinction in its &lt;a href="https://github.com/martinPisacreta/cited-assistant-dotnet/blob/a540d6fbf57c0df4be35eafac076b2688592cdb4/README.md#limits" rel="noopener noreferrer"&gt;limits&lt;/a&gt;: refusals and model failures consume a reservation; rejected admission does not.&lt;/p&gt;

&lt;h2&gt;
  
  
  Missing evidence keeps the model out
&lt;/h2&gt;

&lt;p&gt;The service initializes the result as a refusal. This &lt;a href="https://github.com/martinPisacreta/cited-assistant-dotnet/blob/a540d6fbf57c0df4be35eafac076b2688592cdb4/ChefService.cs#L53" rel="noopener noreferrer"&gt;excerpt&lt;/a&gt; comes before its conditional model branch:&lt;br&gt;
&lt;/p&gt;

&lt;div class="highlight js-code-highlight"&gt;
&lt;pre class="highlight csharp"&gt;&lt;code&gt;&lt;span class="kt"&gt;var&lt;/span&gt; &lt;span class="n"&gt;evidence&lt;/span&gt; &lt;span class="p"&gt;=&lt;/span&gt; &lt;span class="n"&gt;library&lt;/span&gt;&lt;span class="p"&gt;.&lt;/span&gt;&lt;span class="nf"&gt;Find&lt;/span&gt;&lt;span class="p"&gt;(&lt;/span&gt;&lt;span class="n"&gt;question&lt;/span&gt;&lt;span class="p"&gt;,&lt;/span&gt; &lt;span class="n"&gt;history&lt;/span&gt;&lt;span class="p"&gt;);&lt;/span&gt;
&lt;span class="kt"&gt;var&lt;/span&gt; &lt;span class="n"&gt;answer&lt;/span&gt; &lt;span class="p"&gt;=&lt;/span&gt; &lt;span class="n"&gt;OutsideMaterial&lt;/span&gt;&lt;span class="p"&gt;;&lt;/span&gt;
&lt;span class="kt"&gt;string&lt;/span&gt;&lt;span class="p"&gt;?&lt;/span&gt; &lt;span class="n"&gt;source&lt;/span&gt; &lt;span class="p"&gt;=&lt;/span&gt; &lt;span class="k"&gt;null&lt;/span&gt;&lt;span class="p"&gt;;&lt;/span&gt;
&lt;span class="kt"&gt;var&lt;/span&gt; &lt;span class="n"&gt;refused&lt;/span&gt; &lt;span class="p"&gt;=&lt;/span&gt; &lt;span class="k"&gt;true&lt;/span&gt;&lt;span class="p"&gt;;&lt;/span&gt;
&lt;/code&gt;&lt;/pre&gt;

&lt;/div&gt;



&lt;p&gt;When evidence is null, the model branch is skipped. The response explains that no passage answers the question and suggests the sample's supported topics. It carries no source reference.&lt;/p&gt;

&lt;p&gt;This is an application decision with an observable boundary. It does not rely on a model choosing to admit uncertainty. It also does not prove that retrieval catches every unsupported question: keyword retrieval can select an irrelevant passage. This branch handles the narrower condition that retrieval returned nothing.&lt;/p&gt;

&lt;p&gt;The &lt;a href="https://github.com/martinPisacreta/cited-assistant-dotnet/blob/a540d6fbf57c0df4be35eafac076b2688592cdb4/tests/ChefServiceTests.cs#L43" rel="noopener noreferrer"&gt;outside-material test&lt;/a&gt; asks who won an election and checks three properties:&lt;br&gt;
&lt;/p&gt;

&lt;div class="highlight js-code-highlight"&gt;
&lt;pre class="highlight csharp"&gt;&lt;code&gt;&lt;span class="n"&gt;Assert&lt;/span&gt;&lt;span class="p"&gt;.&lt;/span&gt;&lt;span class="nf"&gt;True&lt;/span&gt;&lt;span class="p"&gt;(&lt;/span&gt;&lt;span class="n"&gt;reply&lt;/span&gt;&lt;span class="p"&gt;.&lt;/span&gt;&lt;span class="n"&gt;Refused&lt;/span&gt;&lt;span class="p"&gt;);&lt;/span&gt;
&lt;span class="n"&gt;Assert&lt;/span&gt;&lt;span class="p"&gt;.&lt;/span&gt;&lt;span class="nf"&gt;Null&lt;/span&gt;&lt;span class="p"&gt;(&lt;/span&gt;&lt;span class="n"&gt;reply&lt;/span&gt;&lt;span class="p"&gt;.&lt;/span&gt;&lt;span class="n"&gt;Source&lt;/span&gt;&lt;span class="p"&gt;);&lt;/span&gt;
&lt;span class="n"&gt;Assert&lt;/span&gt;&lt;span class="p"&gt;.&lt;/span&gt;&lt;span class="nf"&gt;Equal&lt;/span&gt;&lt;span class="p"&gt;(&lt;/span&gt;&lt;span class="m"&gt;0&lt;/span&gt;&lt;span class="p"&gt;,&lt;/span&gt; &lt;span class="n"&gt;model&lt;/span&gt;&lt;span class="p"&gt;.&lt;/span&gt;&lt;span class="n"&gt;Calls&lt;/span&gt;&lt;span class="p"&gt;);&lt;/span&gt;
&lt;/code&gt;&lt;/pre&gt;

&lt;/div&gt;



&lt;p&gt;Checking the call counter matters. A refusal string alone would not establish that generation was skipped.&lt;/p&gt;

&lt;h2&gt;
  
  
  A failed model call takes a different path
&lt;/h2&gt;

&lt;p&gt;If retrieval finds evidence, the service calls the configured model. Ordinary exceptions from that call are caught without exposing dependency error details. Cancellation is explicitly rethrown.&lt;/p&gt;

&lt;p&gt;The service then validates the candidate. For a cited model, it delegates to &lt;code&gt;IsValidAnswer&lt;/code&gt;. For another model implementation, it requires the candidate to equal the retrieved passage exactly. If validation fails, the service returns its unverified-answer refusal instead of the candidate.&lt;/p&gt;

&lt;p&gt;These paths distinguish two situations: no retrieved passage, and no accepted answer from the selected passage. Neither establishes that the underlying documents contain no possible answer. They describe what this implementation could retrieve and accept.&lt;/p&gt;

&lt;p&gt;Citation validation also has a limit: a matching source identifier does not prove that every generated claim follows from the passage. A refusal policy should not be described as a factual correctness guarantee.&lt;/p&gt;

&lt;h2&gt;
  
  
  A completed refusal remains in context
&lt;/h2&gt;

&lt;p&gt;After either answer path, &lt;a href="https://github.com/martinPisacreta/cited-assistant-dotnet/blob/a540d6fbf57c0df4be35eafac076b2688592cdb4/ChefService.cs#L88" rel="noopener noreferrer"&gt;the service records a turn&lt;/a&gt;:&lt;br&gt;
&lt;/p&gt;

&lt;div class="highlight js-code-highlight"&gt;
&lt;pre class="highlight csharp"&gt;&lt;code&gt;&lt;span class="n"&gt;cancellationToken&lt;/span&gt;&lt;span class="p"&gt;.&lt;/span&gt;&lt;span class="nf"&gt;ThrowIfCancellationRequested&lt;/span&gt;&lt;span class="p"&gt;();&lt;/span&gt;
&lt;span class="n"&gt;history&lt;/span&gt;&lt;span class="p"&gt;.&lt;/span&gt;&lt;span class="nf"&gt;Add&lt;/span&gt;&lt;span class="p"&gt;(&lt;/span&gt;&lt;span class="k"&gt;new&lt;/span&gt; &lt;span class="nf"&gt;ChefTurn&lt;/span&gt;&lt;span class="p"&gt;(&lt;/span&gt;&lt;span class="n"&gt;question&lt;/span&gt;&lt;span class="p"&gt;,&lt;/span&gt; &lt;span class="n"&gt;answer&lt;/span&gt;&lt;span class="p"&gt;,&lt;/span&gt; &lt;span class="n"&gt;source&lt;/span&gt;&lt;span class="p"&gt;,&lt;/span&gt; &lt;span class="n"&gt;time&lt;/span&gt;&lt;span class="p"&gt;.&lt;/span&gt;&lt;span class="nf"&gt;GetUtcNow&lt;/span&gt;&lt;span class="p"&gt;()));&lt;/span&gt;
&lt;span class="k"&gt;if&lt;/span&gt; &lt;span class="p"&gt;(&lt;/span&gt;&lt;span class="n"&gt;history&lt;/span&gt;&lt;span class="p"&gt;.&lt;/span&gt;&lt;span class="n"&gt;Count&lt;/span&gt; &lt;span class="p"&gt;&amp;gt;&lt;/span&gt; &lt;span class="n"&gt;HistoryLimit&lt;/span&gt;&lt;span class="p"&gt;)&lt;/span&gt;
    &lt;span class="n"&gt;history&lt;/span&gt;&lt;span class="p"&gt;.&lt;/span&gt;&lt;span class="nf"&gt;RemoveRange&lt;/span&gt;&lt;span class="p"&gt;(&lt;/span&gt;&lt;span class="m"&gt;0&lt;/span&gt;&lt;span class="p"&gt;,&lt;/span&gt; &lt;span class="n"&gt;history&lt;/span&gt;&lt;span class="p"&gt;.&lt;/span&gt;&lt;span class="n"&gt;Count&lt;/span&gt; &lt;span class="p"&gt;-&lt;/span&gt; &lt;span class="n"&gt;HistoryLimit&lt;/span&gt;&lt;span class="p"&gt;);&lt;/span&gt;
&lt;/code&gt;&lt;/pre&gt;

&lt;/div&gt;



&lt;p&gt;There is no success-only condition around that append. A normally completed refusal enters history with its refusal text and null source. History retains at most twenty turns per conversation, and its key includes both member and conversation identifiers.&lt;/p&gt;

&lt;p&gt;Cancellation is different. If cancellation propagates before the append, that request does not add a completed turn through this code path. The earlier reservation has already happened; the service shows no refund operation.&lt;/p&gt;

&lt;p&gt;Keeping a refusal in history does not turn it into evidence. The &lt;a href="https://github.com/martinPisacreta/cited-assistant-dotnet/blob/a540d6fbf57c0df4be35eafac076b2688592cdb4/ChefContracts.cs#L16" rel="noopener noreferrer"&gt;model-input contract&lt;/a&gt; explicitly separates conversational history from the selected evidence passage.&lt;/p&gt;

&lt;h2&gt;
  
  
  What I checked, and what I did not run
&lt;/h2&gt;

&lt;p&gt;For this article, I retrieved the public repository files with curl at the fixed commit and inspected the service, contracts, README, and visible test cases. I inspected the outside-material assertions above and the unknown-member test, which checks that rejected admission makes no model call.&lt;/p&gt;

&lt;p&gt;I ran no build, automated test suite, API session, or real-provider integration test for this article. The README records a previous 49-test pass; that is recorded validation, not a fresh result from this review.&lt;/p&gt;

&lt;p&gt;Subscription, quota, and conversation state are in memory. Restarting resets them, and multiple instances do not share them. These are prototype limits to address before relying on the policy across a deployed service.&lt;/p&gt;

&lt;p&gt;If you want an assistant like this over your own documents, I offer it as a fixed-price project on Upwork: &lt;a href="https://www.upwork.com/services/product/2115562756244627175" rel="noopener noreferrer"&gt;https://www.upwork.com/services/product/2115562756244627175&lt;/a&gt;&lt;/p&gt;

</description>
      <category>dotnet</category>
      <category>csharp</category>
      <category>ai</category>
      <category>rag</category>
    </item>
    <item>
      <title>Claude Code hooks on Windows: when a Bash matcher sees nothing</title>
      <dc:creator>Martin Pisacreta</dc:creator>
      <pubDate>Mon, 02 Nov 2026 09:19:01 +0000</pubDate>
      <link>https://dev.to/martinpisacreta/claude-code-hooks-on-windows-when-a-bash-matcher-sees-nothing-30l4</link>
      <guid>https://dev.to/martinpisacreta/claude-code-hooks-on-windows-when-a-bash-matcher-sees-nothing-30l4</guid>
      <description>&lt;p&gt;A reader spotted a Windows gap in my earlier article about blocking destructive shell commands in Claude Code. The hook matched &lt;code&gt;Bash&lt;/code&gt;. On a machine where Claude uses its PowerShell tool, that hook can miss the command entirely.&lt;/p&gt;

&lt;p&gt;&lt;a href="https://dev.to/martinpisacreta/how-i-stop-claude-code-from-running-destructive-shell-commands-5c98"&gt;The previous article now includes the correction&lt;/a&gt;. This follow-up explains the two configuration issues involved: which tool triggers the hook, and what happens if the hook cannot run. There is also a separate limitation: the original &lt;code&gt;guard.py&lt;/code&gt; understands POSIX-style commands, so PowerShell commands remain outside its coverage.&lt;/p&gt;

&lt;h2&gt;
  
  
  A Bash matcher can receive no events
&lt;/h2&gt;

&lt;p&gt;Claude Code has separate Bash and PowerShell tools. Matching one tool does not automatically match the other.&lt;/p&gt;

&lt;p&gt;The &lt;a href="https://code.claude.com/docs/en/hooks.md#powershell" rel="noopener noreferrer"&gt;PowerShell section of the hooks documentation&lt;/a&gt; states: “A hook that matches only &lt;code&gt;Bash&lt;/code&gt; never fires there.” The surrounding explanation concerns Windows without Git Bash, where PowerShell is enabled automatically and Bash is not registered.&lt;/p&gt;

&lt;p&gt;The issue also matters wherever the PowerShell tool is enabled on Windows: Claude treats it as the primary shell and routes shell commands through it. The &lt;a href="https://code.claude.com/docs/en/tools-reference.md#powershell-tool" rel="noopener noreferrer"&gt;tools reference&lt;/a&gt; explains availability by platform and account. With Git Bash installed, Bash can remain available for POSIX scripts.&lt;/p&gt;

&lt;p&gt;The practical correction is to match both tool names with &lt;code&gt;Bash|PowerShell&lt;/code&gt;. Both supply the command string through &lt;code&gt;tool_input.command&lt;/code&gt;, according to the tools reference.&lt;/p&gt;

&lt;p&gt;That fixes event routing. It does not establish that a script understands both command languages.&lt;/p&gt;

&lt;h2&gt;
  
  
  A hook that cannot start needs an explicit policy
&lt;/h2&gt;

&lt;p&gt;A second gap appears when the hook is selected but cannot execute. Perhaps its Python executable is missing, its script path is wrong, or the process crashes.&lt;/p&gt;

&lt;p&gt;The &lt;a href="https://code.claude.com/docs/en/hooks.md#block-the-action-when-a-hook-fails" rel="noopener noreferrer"&gt;failure-handling documentation&lt;/a&gt; says actions generally continue when hooks fail or time out. An unavailable guard can therefore leave an action proceeding through the normal permission flow.&lt;/p&gt;

&lt;p&gt;For a command hook, setting &lt;code&gt;onFailure&lt;/code&gt; to &lt;code&gt;block&lt;/code&gt; makes a failure block the triggering action according to that event's blocking behavior. For &lt;code&gt;PreToolUse&lt;/code&gt;, that means blocking the tool call. The documentation specifies: “Requires Claude Code v2.1.295 or later.”&lt;/p&gt;

&lt;p&gt;There is a Windows qualification: without Git Bash, a shell-form &lt;code&gt;PreToolUse&lt;/code&gt; or &lt;code&gt;PermissionRequest&lt;/code&gt; hook can block on failure when &lt;code&gt;onFailure&lt;/code&gt; is unset. That documented exception is why describing every missing-interpreter failure as non-blocking would be inaccurate. Set the failure policy explicitly instead of depending on that default.&lt;/p&gt;

&lt;h2&gt;
  
  
  Settings that address routing and startup failure
&lt;/h2&gt;

&lt;p&gt;For an existing guard saved at &lt;code&gt;.claude/hooks/guard.py&lt;/code&gt;, this illustrates the configuration in &lt;code&gt;.claude/settings.json&lt;/code&gt;:&lt;br&gt;
&lt;/p&gt;

&lt;div class="highlight js-code-highlight"&gt;
&lt;pre class="highlight json"&gt;&lt;code&gt;&lt;span class="p"&gt;{&lt;/span&gt;&lt;span class="w"&gt;
  &lt;/span&gt;&lt;span class="nl"&gt;"hooks"&lt;/span&gt;&lt;span class="p"&gt;:&lt;/span&gt;&lt;span class="w"&gt; &lt;/span&gt;&lt;span class="p"&gt;{&lt;/span&gt;&lt;span class="w"&gt;
    &lt;/span&gt;&lt;span class="nl"&gt;"PreToolUse"&lt;/span&gt;&lt;span class="p"&gt;:&lt;/span&gt;&lt;span class="w"&gt; &lt;/span&gt;&lt;span class="p"&gt;[&lt;/span&gt;&lt;span class="w"&gt;
      &lt;/span&gt;&lt;span class="p"&gt;{&lt;/span&gt;&lt;span class="w"&gt;
        &lt;/span&gt;&lt;span class="nl"&gt;"matcher"&lt;/span&gt;&lt;span class="p"&gt;:&lt;/span&gt;&lt;span class="w"&gt; &lt;/span&gt;&lt;span class="s2"&gt;"Bash|PowerShell"&lt;/span&gt;&lt;span class="p"&gt;,&lt;/span&gt;&lt;span class="w"&gt;
        &lt;/span&gt;&lt;span class="nl"&gt;"hooks"&lt;/span&gt;&lt;span class="p"&gt;:&lt;/span&gt;&lt;span class="w"&gt; &lt;/span&gt;&lt;span class="p"&gt;[&lt;/span&gt;&lt;span class="w"&gt;
          &lt;/span&gt;&lt;span class="p"&gt;{&lt;/span&gt;&lt;span class="w"&gt;
            &lt;/span&gt;&lt;span class="nl"&gt;"type"&lt;/span&gt;&lt;span class="p"&gt;:&lt;/span&gt;&lt;span class="w"&gt; &lt;/span&gt;&lt;span class="s2"&gt;"command"&lt;/span&gt;&lt;span class="p"&gt;,&lt;/span&gt;&lt;span class="w"&gt;
            &lt;/span&gt;&lt;span class="nl"&gt;"command"&lt;/span&gt;&lt;span class="p"&gt;:&lt;/span&gt;&lt;span class="w"&gt; &lt;/span&gt;&lt;span class="s2"&gt;"python.exe"&lt;/span&gt;&lt;span class="p"&gt;,&lt;/span&gt;&lt;span class="w"&gt;
            &lt;/span&gt;&lt;span class="nl"&gt;"args"&lt;/span&gt;&lt;span class="p"&gt;:&lt;/span&gt;&lt;span class="w"&gt; &lt;/span&gt;&lt;span class="p"&gt;[&lt;/span&gt;&lt;span class="w"&gt;
              &lt;/span&gt;&lt;span class="s2"&gt;"${CLAUDE_PROJECT_DIR}/.claude/hooks/guard.py"&lt;/span&gt;&lt;span class="w"&gt;
            &lt;/span&gt;&lt;span class="p"&gt;],&lt;/span&gt;&lt;span class="w"&gt;
            &lt;/span&gt;&lt;span class="nl"&gt;"onFailure"&lt;/span&gt;&lt;span class="p"&gt;:&lt;/span&gt;&lt;span class="w"&gt; &lt;/span&gt;&lt;span class="s2"&gt;"block"&lt;/span&gt;&lt;span class="w"&gt;
          &lt;/span&gt;&lt;span class="p"&gt;}&lt;/span&gt;&lt;span class="w"&gt;
        &lt;/span&gt;&lt;span class="p"&gt;]&lt;/span&gt;&lt;span class="w"&gt;
      &lt;/span&gt;&lt;span class="p"&gt;}&lt;/span&gt;&lt;span class="w"&gt;
    &lt;/span&gt;&lt;span class="p"&gt;]&lt;/span&gt;&lt;span class="w"&gt;
  &lt;/span&gt;&lt;span class="p"&gt;}&lt;/span&gt;&lt;span class="w"&gt;
&lt;/span&gt;&lt;span class="p"&gt;}&lt;/span&gt;&lt;span class="w"&gt;
&lt;/span&gt;&lt;/code&gt;&lt;/pre&gt;

&lt;/div&gt;



&lt;p&gt;Use the executable actually installed on your machine. &lt;code&gt;python.exe&lt;/code&gt; here is an example, not a claim that Python is installed or discoverable on your PATH.&lt;/p&gt;

&lt;p&gt;The &lt;code&gt;args&lt;/code&gt; array selects exec form. As described in the &lt;a href="https://code.claude.com/docs/en/hooks.md#exec-form-and-shell-form" rel="noopener noreferrer"&gt;command-hook documentation&lt;/a&gt;, Claude Code starts the executable directly and passes each argument separately. The script path therefore stays one argument even when the project directory contains spaces. On Windows, exec form requires a real executable.&lt;/p&gt;

&lt;p&gt;This example is a configuration illustration checked against the documentation; I have not tested it on your Windows installation.&lt;/p&gt;

&lt;h2&gt;
  
  
  PowerShell commands are still uncovered
&lt;/h2&gt;

&lt;p&gt;The original &lt;code&gt;guard.py&lt;/code&gt; only understands POSIX-style command syntax. Adding PowerShell to the matcher does not teach it PowerShell grammar, aliases, parameter binding, or command composition.&lt;/p&gt;

&lt;p&gt;The wider matcher delivers PowerShell events to the script, but its existing analysis still leaves those commands uncovered. A successful exit from that script is not evidence that it evaluated a PowerShell command correctly.&lt;/p&gt;

&lt;p&gt;Likewise, &lt;code&gt;onFailure: block&lt;/code&gt; covers hook failures. It cannot repair a parser that runs successfully but misunderstands its input or returns no blocking decision. PowerShell coverage requires an appropriate implementation and separate verification. This settings change does not provide it.&lt;/p&gt;

&lt;p&gt;The shell launching a hook and the tool being inspected are also separate choices. Running the hook under PowerShell does not change the language of the command received in the event.&lt;/p&gt;

&lt;h2&gt;
  
  
  Verify each boundary with harmless commands
&lt;/h2&gt;

&lt;p&gt;Before relying on this setup, check that a harmless call through each available shell tool actually reaches the hook. Confirm the received tool name and command field rather than inferring coverage from the settings file.&lt;/p&gt;

&lt;p&gt;Then exercise failure handling safely: temporarily point the hook at a missing script and request a harmless listing. On a supported version with &lt;code&gt;onFailure: block&lt;/code&gt;, the call should be blocked. Restore the path afterward.&lt;/p&gt;

&lt;p&gt;Finally, test the guard's actual decisions separately for each supported language. Until PowerShell analysis exists and passes those checks, describe this as a POSIX guard with corrected routing and failure handling.&lt;/p&gt;

&lt;p&gt;If your team wants this set up and tested on your own machines, I offer it as a fixed-price package on Upwork: &lt;a href="https://www.upwork.com/services/product/2115796250869866384" rel="noopener noreferrer"&gt;https://www.upwork.com/services/product/2115796250869866384&lt;/a&gt;&lt;/p&gt;

</description>
      <category>claudecode</category>
      <category>windows</category>
      <category>powershell</category>
      <category>ai</category>
    </item>
    <item>
      <title>How I cap OpenAI spending per user in ASP.NET Core</title>
      <dc:creator>Martin Pisacreta</dc:creator>
      <pubDate>Mon, 02 Nov 2026 04:39:22 +0000</pubDate>
      <link>https://dev.to/martinpisacreta/how-i-cap-openai-spending-per-user-in-aspnet-core-27bb</link>
      <guid>https://dev.to/martinpisacreta/how-i-cap-openai-spending-per-user-in-aspnet-core-27bb</guid>
      <description>&lt;p&gt;A per-user spending limit needs to affect whether an application sends a request. Recording token usage afterward helps explain costs, but it cannot prevent several concurrent requests from starting against the same remaining balance.&lt;/p&gt;

&lt;p&gt;My &lt;a href="https://github.com/martinPisacreta/openai-spend-caps-dotnet/tree/990e9ec" rel="noopener noreferrer"&gt;ASP.NET Core sample&lt;/a&gt; addresses that admission decision with reservations. It authenticates a user, reserves estimated exposure before contacting the provider, and records known or uncertain costs afterward. This walkthrough refers to commit &lt;code&gt;990e9ec&lt;/code&gt;; it does not assume current model prices or claim live OpenAI validation.&lt;/p&gt;

&lt;h2&gt;
  
  
  Take identity from authentication
&lt;/h2&gt;

&lt;p&gt;The caller submits a prompt. The server chooses the user account from a configured bearer token, rather than accepting a user ID in request JSON. That distinction keeps a caller from selecting someone else's budget.&lt;/p&gt;

&lt;p&gt;The request contract is deliberately small:&lt;br&gt;
&lt;/p&gt;

&lt;div class="highlight js-code-highlight"&gt;
&lt;pre class="highlight csharp"&gt;&lt;code&gt;&lt;span class="k"&gt;public&lt;/span&gt; &lt;span class="k"&gt;sealed&lt;/span&gt; &lt;span class="k"&gt;record&lt;/span&gt; &lt;span class="nc"&gt;ChatRequest&lt;/span&gt;&lt;span class="p"&gt;(&lt;/span&gt;&lt;span class="kt"&gt;string&lt;/span&gt; &lt;span class="n"&gt;Prompt&lt;/span&gt;&lt;span class="p"&gt;);&lt;/span&gt;
&lt;/code&gt;&lt;/pre&gt;

&lt;/div&gt;



&lt;p&gt;After authentication, the endpoint passes the resolved identity separately:&lt;br&gt;
&lt;/p&gt;

&lt;div class="highlight js-code-highlight"&gt;
&lt;pre class="highlight csharp"&gt;&lt;code&gt;&lt;span class="kt"&gt;var&lt;/span&gt; &lt;span class="n"&gt;reply&lt;/span&gt; &lt;span class="p"&gt;=&lt;/span&gt; &lt;span class="k"&gt;await&lt;/span&gt; &lt;span class="n"&gt;gateway&lt;/span&gt;&lt;span class="p"&gt;.&lt;/span&gt;&lt;span class="nf"&gt;SendAsync&lt;/span&gt;&lt;span class="p"&gt;(&lt;/span&gt;
    &lt;span class="n"&gt;userId&lt;/span&gt;&lt;span class="p"&gt;,&lt;/span&gt; &lt;span class="n"&gt;request&lt;/span&gt;&lt;span class="p"&gt;.&lt;/span&gt;&lt;span class="n"&gt;Prompt&lt;/span&gt;&lt;span class="p"&gt;,&lt;/span&gt; &lt;span class="n"&gt;context&lt;/span&gt;&lt;span class="p"&gt;.&lt;/span&gt;&lt;span class="n"&gt;RequestAborted&lt;/span&gt;&lt;span class="p"&gt;);&lt;/span&gt;
&lt;/code&gt;&lt;/pre&gt;

&lt;/div&gt;



&lt;p&gt;&lt;a href="https://github.com/martinPisacreta/openai-spend-caps-dotnet/blob/990e9ec/Program.cs" rel="noopener noreferrer"&gt;Source: Program.cs&lt;/a&gt;.&lt;/p&gt;

&lt;p&gt;The server maps configured token hashes to identities. The endpoint's JSON options reject unmapped members, so a request that adds a &lt;code&gt;user&lt;/code&gt; field gets a 400 instead of choosing a budget. Model selection, prices, credentials, and budgets also come from server configuration. The provider credential and the user's authentication token serve different purposes.&lt;/p&gt;

&lt;h2&gt;
  
  
  Estimate one attempt, then reserve every permitted attempt
&lt;/h2&gt;

&lt;p&gt;The policy calculates exposure from the maximum prompt size, an input overhead allowance, maximum output tokens, and explicitly configured prices:&lt;br&gt;
&lt;/p&gt;

&lt;div class="highlight js-code-highlight"&gt;
&lt;pre class="highlight csharp"&gt;&lt;code&gt;&lt;span class="k"&gt;return&lt;/span&gt; &lt;span class="k"&gt;checked&lt;/span&gt;&lt;span class="p"&gt;(&lt;/span&gt;
    &lt;span class="p"&gt;((&lt;/span&gt;&lt;span class="n"&gt;MaxPromptUtf8Bytes&lt;/span&gt; &lt;span class="p"&gt;+&lt;/span&gt; &lt;span class="n"&gt;InputOverheadTokenAllowance&lt;/span&gt;&lt;span class="p"&gt;)&lt;/span&gt; &lt;span class="p"&gt;*&lt;/span&gt;
        &lt;span class="n"&gt;InputUsdPerMillionTokens&lt;/span&gt; &lt;span class="p"&gt;+&lt;/span&gt;
     &lt;span class="n"&gt;MaxOutputTokens&lt;/span&gt; &lt;span class="p"&gt;*&lt;/span&gt; &lt;span class="n"&gt;OutputUsdPerMillionTokens&lt;/span&gt;&lt;span class="p"&gt;)&lt;/span&gt; &lt;span class="p"&gt;/&lt;/span&gt; &lt;span class="m"&gt;1000000m&lt;/span&gt;&lt;span class="p"&gt;);&lt;/span&gt;
&lt;/code&gt;&lt;/pre&gt;

&lt;/div&gt;



&lt;p&gt;&lt;a href="https://github.com/martinPisacreta/openai-spend-caps-dotnet/blob/990e9ec/ChatPolicy.cs#L40-L50" rel="noopener noreferrer"&gt;Source: ChatPolicy.cs&lt;/a&gt;.&lt;/p&gt;

&lt;p&gt;The input calculation allows one token per UTF-8 byte plus configured overhead. It is an admission estimate for this text-only request shape, rather than a tokenizer. The application must maintain suitable allowances and prices for its selected model.&lt;/p&gt;

&lt;p&gt;Retries also create exposure. Reserving enough for one attempt would leave later attempts outside the initial budget decision. The request estimate therefore includes every permitted attempt:&lt;br&gt;
&lt;/p&gt;

&lt;div class="highlight js-code-highlight"&gt;
&lt;pre class="highlight csharp"&gt;&lt;code&gt;&lt;span class="k"&gt;public&lt;/span&gt; &lt;span class="kt"&gt;decimal&lt;/span&gt; &lt;span class="nf"&gt;MaximumEstimatedRequestUsd&lt;/span&gt;&lt;span class="p"&gt;()&lt;/span&gt; &lt;span class="p"&gt;=&amp;gt;&lt;/span&gt;
    &lt;span class="k"&gt;checked&lt;/span&gt;&lt;span class="p"&gt;(&lt;/span&gt;&lt;span class="nf"&gt;MaximumEstimatedAttemptUsd&lt;/span&gt;&lt;span class="p"&gt;()&lt;/span&gt; &lt;span class="p"&gt;*&lt;/span&gt; &lt;span class="n"&gt;MaxAttempts&lt;/span&gt;&lt;span class="p"&gt;);&lt;/span&gt;
&lt;/code&gt;&lt;/pre&gt;

&lt;/div&gt;



&lt;p&gt;&lt;a href="https://github.com/martinPisacreta/openai-spend-caps-dotnet/blob/990e9ec/ChatPolicy.cs#L52-L54" rel="noopener noreferrer"&gt;Source: ChatPolicy.cs&lt;/a&gt;.&lt;/p&gt;

&lt;p&gt;This approach can reject a request whose eventual successful response would have been cheaper. That is a consequence of reserving the allowed exposure before knowing how many attempts will happen.&lt;/p&gt;

&lt;h2&gt;
  
  
  Reserve before the provider call
&lt;/h2&gt;

&lt;p&gt;The gateway makes its admission decision before entering the sending loop:&lt;br&gt;
&lt;/p&gt;

&lt;div class="highlight js-code-highlight"&gt;
&lt;pre class="highlight csharp"&gt;&lt;code&gt;&lt;span class="k"&gt;if&lt;/span&gt; &lt;span class="p"&gt;(!&lt;/span&gt;&lt;span class="n"&gt;ledger&lt;/span&gt;&lt;span class="p"&gt;.&lt;/span&gt;&lt;span class="nf"&gt;TryReserve&lt;/span&gt;&lt;span class="p"&gt;(&lt;/span&gt;&lt;span class="n"&gt;authenticatedUserId&lt;/span&gt;&lt;span class="p"&gt;,&lt;/span&gt; &lt;span class="n"&gt;requestId&lt;/span&gt;&lt;span class="p"&gt;,&lt;/span&gt;
        &lt;span class="n"&gt;policy&lt;/span&gt;&lt;span class="p"&gt;.&lt;/span&gt;&lt;span class="nf"&gt;MaximumEstimatedRequestUsd&lt;/span&gt;&lt;span class="p"&gt;()))&lt;/span&gt;
    &lt;span class="k"&gt;throw&lt;/span&gt; &lt;span class="k"&gt;new&lt;/span&gt; &lt;span class="nf"&gt;BudgetExceededException&lt;/span&gt;&lt;span class="p"&gt;();&lt;/span&gt;
&lt;/code&gt;&lt;/pre&gt;

&lt;/div&gt;



&lt;p&gt;&lt;a href="https://github.com/martinPisacreta/openai-spend-caps-dotnet/blob/990e9ec/ChatGateway.cs#L77-L80" rel="noopener noreferrer"&gt;Source: ChatGateway.cs&lt;/a&gt;.&lt;/p&gt;

&lt;p&gt;The default ledger coordinates reservations with a lock. Available budget accounts for both recorded costs and held exposure, so concurrent callers cannot independently reserve the same balance. The HTTP endpoint translates insufficient budget into a 429 response.&lt;/p&gt;

&lt;p&gt;The lock protects this process's ledger. It does not coordinate separate application instances, and restarting loses the in-memory records. The &lt;a href="https://github.com/martinPisacreta/openai-spend-caps-dotnet/blob/990e9ec/SpendingLedger.cs" rel="noopener noreferrer"&gt;ledger implementation&lt;/a&gt; explicitly documents those limits. A deployment using multiple instances needs shared, durable, atomic reservations.&lt;/p&gt;

&lt;h2&gt;
  
  
  Track uncertainty when an attempt starts
&lt;/h2&gt;

&lt;p&gt;A failed response does not supply reliable token accounting. An interrupted connection can also leave the application without a usable answer about what happened remotely.&lt;/p&gt;

&lt;p&gt;Immediately before sending, the gateway adds one attempt's estimated exposure to its uncertain total:&lt;br&gt;
&lt;/p&gt;

&lt;div class="highlight js-code-highlight"&gt;
&lt;pre class="highlight csharp"&gt;&lt;code&gt;&lt;span class="n"&gt;attempts&lt;/span&gt;&lt;span class="p"&gt;++;&lt;/span&gt;
&lt;span class="n"&gt;uncertain&lt;/span&gt; &lt;span class="p"&gt;+=&lt;/span&gt; &lt;span class="n"&gt;attemptExposure&lt;/span&gt;&lt;span class="p"&gt;;&lt;/span&gt;
&lt;span class="k"&gt;using&lt;/span&gt; &lt;span class="nn"&gt;var&lt;/span&gt; &lt;span class="n"&gt;response&lt;/span&gt; &lt;span class="p"&gt;=&lt;/span&gt; &lt;span class="k"&gt;await&lt;/span&gt; &lt;span class="n"&gt;client&lt;/span&gt;&lt;span class="p"&gt;.&lt;/span&gt;&lt;span class="nf"&gt;SendAsync&lt;/span&gt;&lt;span class="p"&gt;(&lt;/span&gt;
    &lt;span class="n"&gt;request&lt;/span&gt;&lt;span class="p"&gt;,&lt;/span&gt; &lt;span class="n"&gt;HttpCompletionOption&lt;/span&gt;&lt;span class="p"&gt;.&lt;/span&gt;&lt;span class="n"&gt;ResponseHeadersRead&lt;/span&gt;&lt;span class="p"&gt;,&lt;/span&gt; &lt;span class="n"&gt;token&lt;/span&gt;&lt;span class="p"&gt;);&lt;/span&gt;
&lt;/code&gt;&lt;/pre&gt;

&lt;/div&gt;



&lt;p&gt;&lt;a href="https://github.com/martinPisacreta/openai-spend-caps-dotnet/blob/990e9ec/ChatGateway.cs#L106-L111" rel="noopener noreferrer"&gt;Source: ChatGateway.cs&lt;/a&gt;.&lt;/p&gt;

&lt;p&gt;The retry loop bounds the attempt count, retries 429 and server errors, and applies policy delays within an overall timeout. Earlier attempts retain their uncertain exposure even when a later attempt succeeds. A successful retry cannot establish the earlier attempts' billing outcome.&lt;/p&gt;

&lt;p&gt;This conservative treatment can hold more budget than the eventual provider charge. The sample preserves that uncertainty instead of assuming that every unsuccessful attempt was free.&lt;/p&gt;

&lt;h2&gt;
  
  
  Release unused exposure and account for returned usage
&lt;/h2&gt;

&lt;p&gt;When a response contains valid nonnegative token counts, the gateway calculates their cost using configured prices and removes that attempt's estimated uncertain exposure:&lt;br&gt;
&lt;/p&gt;

&lt;div class="highlight js-code-highlight"&gt;
&lt;pre class="highlight csharp"&gt;&lt;code&gt;&lt;span class="n"&gt;confirmed&lt;/span&gt; &lt;span class="p"&gt;+=&lt;/span&gt; &lt;span class="n"&gt;policy&lt;/span&gt;&lt;span class="p"&gt;.&lt;/span&gt;&lt;span class="nf"&gt;EstimateUsageUsd&lt;/span&gt;&lt;span class="p"&gt;(&lt;/span&gt;&lt;span class="n"&gt;inputTokens&lt;/span&gt;&lt;span class="p"&gt;,&lt;/span&gt; &lt;span class="n"&gt;outputTokens&lt;/span&gt;&lt;span class="p"&gt;);&lt;/span&gt;
&lt;span class="n"&gt;uncertain&lt;/span&gt; &lt;span class="p"&gt;-=&lt;/span&gt; &lt;span class="n"&gt;attemptExposure&lt;/span&gt;&lt;span class="p"&gt;;&lt;/span&gt;
&lt;span class="n"&gt;knownUsage&lt;/span&gt; &lt;span class="p"&gt;=&lt;/span&gt; &lt;span class="k"&gt;true&lt;/span&gt;&lt;span class="p"&gt;;&lt;/span&gt;
&lt;/code&gt;&lt;/pre&gt;

&lt;/div&gt;



&lt;p&gt;&lt;a href="https://github.com/martinPisacreta/openai-spend-caps-dotnet/blob/990e9ec/ChatGateway.cs#L142-L155" rel="noopener noreferrer"&gt;Source: ChatGateway.cs&lt;/a&gt;.&lt;/p&gt;

&lt;p&gt;Completion runs in a &lt;code&gt;finally&lt;/code&gt; block:&lt;br&gt;
&lt;/p&gt;

&lt;div class="highlight js-code-highlight"&gt;
&lt;pre class="highlight csharp"&gt;&lt;code&gt;&lt;span class="n"&gt;completed&lt;/span&gt; &lt;span class="p"&gt;=&lt;/span&gt; &lt;span class="n"&gt;ledger&lt;/span&gt;&lt;span class="p"&gt;.&lt;/span&gt;&lt;span class="nf"&gt;Complete&lt;/span&gt;&lt;span class="p"&gt;(&lt;/span&gt;&lt;span class="n"&gt;requestId&lt;/span&gt;&lt;span class="p"&gt;,&lt;/span&gt; &lt;span class="n"&gt;confirmed&lt;/span&gt;&lt;span class="p"&gt;,&lt;/span&gt; &lt;span class="n"&gt;uncertain&lt;/span&gt;&lt;span class="p"&gt;,&lt;/span&gt; &lt;span class="n"&gt;outcome&lt;/span&gt;&lt;span class="p"&gt;);&lt;/span&gt;
&lt;/code&gt;&lt;/pre&gt;

&lt;/div&gt;



&lt;p&gt;&lt;a href="https://github.com/martinPisacreta/openai-spend-caps-dotnet/blob/990e9ec/ChatGateway.cs#L199-L204" rel="noopener noreferrer"&gt;Source: ChatGateway.cs&lt;/a&gt;.&lt;/p&gt;

&lt;p&gt;The ledger releases the original reservation while retaining the calculated confirmed cost and any uncertain exposure. Consequently, unattempted retries release their reserved share. Known usage replaces its attempt estimate. A successful answer without usage keeps that attempt's estimated exposure held.&lt;/p&gt;

&lt;p&gt;For example, suppose the configured estimate is $0.01 per attempt and three attempts are permitted. Admission reserves $0.03. If the first attempt succeeds with $0.002 of calculated usage, completion retains $0.002 and releases $0.028. These are illustrative figures, not OpenAI prices.&lt;/p&gt;

&lt;h2&gt;
  
  
  Understand what the cap measures
&lt;/h2&gt;

&lt;p&gt;Here, confirmed means calculated from returned usage and configured prices. It does not mean reconciled with a provider invoice. The ledger records reported usage even when its calculated cost exceeds the admission estimate.&lt;/p&gt;

&lt;p&gt;The mechanism controls admission against application accounting. Its effectiveness depends on configuration, supported request shapes, persistent accounting where required, and treatment of uncertain outcomes. It cannot promise an exact provider bill.&lt;/p&gt;

&lt;h2&gt;
  
  
  What was verified
&lt;/h2&gt;

&lt;p&gt;For commit &lt;code&gt;990e9ec&lt;/code&gt;, the sample builds on Windows with &lt;strong&gt;0 warnings&lt;/strong&gt;, the self-test passes &lt;strong&gt;9 of 9&lt;/strong&gt;, and a local HTTP check on 127.0.0.1 returned &lt;strong&gt;401 without a token&lt;/strong&gt;, &lt;strong&gt;400 for malformed JSON or a &lt;code&gt;user&lt;/code&gt; field&lt;/strong&gt;, and &lt;strong&gt;429 when the budget was exhausted, without calling the provider&lt;/strong&gt;. Those checks were run with Claude, the AI assistant I use for this kind of work.&lt;/p&gt;

&lt;p&gt;The sample was &lt;strong&gt;not tested against real OpenAI&lt;/strong&gt;. Its &lt;a href="https://github.com/martinPisacreta/openai-spend-caps-dotnet/blob/990e9ec/SelfTests.cs" rel="noopener noreferrer"&gt;self-tests&lt;/a&gt; use a scripted local provider. To run that path without an API key:&lt;br&gt;
&lt;/p&gt;

&lt;div class="highlight js-code-highlight"&gt;
&lt;pre class="highlight shell"&gt;&lt;code&gt;dotnet run &lt;span class="nt"&gt;--project&lt;/span&gt; LlmIntegration.csproj &lt;span class="nt"&gt;--&lt;/span&gt; &lt;span class="nt"&gt;--self-test&lt;/span&gt;
&lt;/code&gt;&lt;/pre&gt;

&lt;/div&gt;



&lt;p&gt;The &lt;a href="https://github.com/martinPisacreta/openai-spend-caps-dotnet/blob/990e9ec/README.md" rel="noopener noreferrer"&gt;README&lt;/a&gt; explains real-provider environment configuration and the memory ledger's limits. Live compatibility, current pricing, and actual billing remain outside the reported validation.&lt;/p&gt;

&lt;p&gt;If you want this in your ASP.NET Core app, I offer it as a fixed-price project on Upwork: &lt;a href="https://www.upwork.com/services/product/2114839106032222506" rel="noopener noreferrer"&gt;https://www.upwork.com/services/product/2114839106032222506&lt;/a&gt;&lt;/p&gt;

</description>
      <category>dotnet</category>
      <category>csharp</category>
      <category>openai</category>
      <category>webdev</category>
    </item>
    <item>
      <title>What I check when I review an MCP server</title>
      <dc:creator>Martin Pisacreta</dc:creator>
      <pubDate>Mon, 02 Nov 2026 02:52:13 +0000</pubDate>
      <link>https://dev.to/martinpisacreta/what-i-check-when-i-review-an-mcp-server-fne</link>
      <guid>https://dev.to/martinpisacreta/what-i-check-when-i-review-an-mcp-server-fne</guid>
      <description>&lt;p&gt;An MCP server gives an AI client a callable interface. Reviewing it means following the path from a tool description to the code that reads data, changes state, or authorizes a payment. I want the interface to explain that behavior accurately, including what happens when a request fails.&lt;/p&gt;

&lt;p&gt;These examples come from two public projects: textokit-mcp and dotnet-mcp-orders-server. The pinned textokit revision uses JavaScript modules, not TypeScript; the orders server uses C#. This article is a source-based walkthrough. I inspected the referenced code and test definitions; I did not execute their tests for this article.&lt;/p&gt;

&lt;h2&gt;
  
  
  1. Names and input schemas must match the implementation
&lt;/h2&gt;

&lt;p&gt;I start with the interface exposed to the client. Can a caller distinguish a lookup from a cancellation? Which arguments are required? Does the implementation enforce the constraints advertised by the schema?&lt;/p&gt;

&lt;p&gt;In textokit-mcp, paid tool names and schemas are derived from the upstream OpenAPI document:&lt;br&gt;
&lt;/p&gt;

&lt;div class="highlight js-code-highlight"&gt;
&lt;pre class="highlight javascript"&gt;&lt;code&gt;&lt;span class="nx"&gt;name&lt;/span&gt;&lt;span class="p"&gt;:&lt;/span&gt; &lt;span class="dl"&gt;'&lt;/span&gt;&lt;span class="s1"&gt;textokit_&lt;/span&gt;&lt;span class="dl"&gt;'&lt;/span&gt; &lt;span class="o"&gt;+&lt;/span&gt; &lt;span class="nx"&gt;route&lt;/span&gt;&lt;span class="p"&gt;.&lt;/span&gt;&lt;span class="nf"&gt;replace&lt;/span&gt;&lt;span class="p"&gt;(&lt;/span&gt;&lt;span class="sr"&gt;/-/g&lt;/span&gt;&lt;span class="p"&gt;,&lt;/span&gt; &lt;span class="dl"&gt;'&lt;/span&gt;&lt;span class="s1"&gt;_&lt;/span&gt;&lt;span class="dl"&gt;'&lt;/span&gt;&lt;span class="p"&gt;),&lt;/span&gt;
&lt;/code&gt;&lt;/pre&gt;

&lt;/div&gt;





&lt;div class="highlight js-code-highlight"&gt;
&lt;pre class="highlight javascript"&gt;&lt;code&gt;&lt;span class="nx"&gt;inputSchema&lt;/span&gt;&lt;span class="p"&gt;:&lt;/span&gt; &lt;span class="nf"&gt;leanSchema&lt;/span&gt;&lt;span class="p"&gt;(&lt;/span&gt;&lt;span class="nx"&gt;op&lt;/span&gt;&lt;span class="p"&gt;.&lt;/span&gt;&lt;span class="nx"&gt;requestBody&lt;/span&gt;&lt;span class="p"&gt;?.&lt;/span&gt;&lt;span class="nx"&gt;content&lt;/span&gt;&lt;span class="p"&gt;?.[&lt;/span&gt;&lt;span class="dl"&gt;'&lt;/span&gt;&lt;span class="s1"&gt;application/json&lt;/span&gt;&lt;span class="dl"&gt;'&lt;/span&gt;&lt;span class="p"&gt;]?.&lt;/span&gt;&lt;span class="nx"&gt;schema&lt;/span&gt;&lt;span class="p"&gt;),&lt;/span&gt;
&lt;/code&gt;&lt;/pre&gt;

&lt;/div&gt;



&lt;p&gt;&lt;a href="https://github.com/martinPisacreta/textokit-mcp/blob/fa04fb03e39ed3ac3da8f3a04e605749eeaaf3e6/index.mjs#L104-L120" rel="noopener noreferrer"&gt;Source: index.mjs, buildTools&lt;/a&gt;.&lt;/p&gt;

&lt;p&gt;That transformation deserves attention. The same file removes top-level composition and conditional keywords in &lt;code&gt;leanSchema&lt;/code&gt;. Its comment says cross-field rules remain upstream. I would therefore check the upstream rejection behavior before claiming that the advertised schema captures every rule. Publishing a schema and validating a request are separate responsibilities.&lt;/p&gt;

&lt;h2&gt;
  
  
  2. Descriptions should state the actual boundary
&lt;/h2&gt;

&lt;p&gt;A useful description tells the client what a tool can do and what must already be true. It should avoid implying access, persistence, or authority that the implementation does not provide.&lt;/p&gt;

&lt;p&gt;The orders server makes its example data explicit:&lt;br&gt;
&lt;/p&gt;

&lt;div class="highlight js-code-highlight"&gt;
&lt;pre class="highlight csharp"&gt;&lt;code&gt;&lt;span class="p"&gt;[&lt;/span&gt;&lt;span class="nf"&gt;McpServerTool&lt;/span&gt;&lt;span class="p"&gt;(&lt;/span&gt;&lt;span class="n"&gt;Name&lt;/span&gt; &lt;span class="p"&gt;=&lt;/span&gt; &lt;span class="s"&gt;"list_orders"&lt;/span&gt;&lt;span class="p"&gt;)]&lt;/span&gt;
&lt;span class="p"&gt;[&lt;/span&gt;&lt;span class="nf"&gt;Description&lt;/span&gt;&lt;span class="p"&gt;(&lt;/span&gt;&lt;span class="s"&gt;"List the fictional orders in this server's memory."&lt;/span&gt;&lt;span class="p"&gt;)]&lt;/span&gt;
&lt;span class="k"&gt;public&lt;/span&gt; &lt;span class="n"&gt;Order&lt;/span&gt;&lt;span class="p"&gt;[]&lt;/span&gt; &lt;span class="nf"&gt;ListOrders&lt;/span&gt;&lt;span class="p"&gt;()&lt;/span&gt; &lt;span class="p"&gt;=&amp;gt;&lt;/span&gt; &lt;span class="n"&gt;orders&lt;/span&gt;&lt;span class="p"&gt;.&lt;/span&gt;&lt;span class="nf"&gt;List&lt;/span&gt;&lt;span class="p"&gt;();&lt;/span&gt;
&lt;/code&gt;&lt;/pre&gt;

&lt;/div&gt;



&lt;p&gt;&lt;a href="https://github.com/martinPisacreta/dotnet-mcp-orders-server/blob/85ee9c52a5197e1108ffd48f8a61103fa51943c6/src/OrdersMcp/OrderTools.cs#L10-L12" rel="noopener noreferrer"&gt;Source: OrderTools.cs&lt;/a&gt;.&lt;/p&gt;

&lt;p&gt;“Fictional” and “in this server's memory” are meaningful limits. A reader should not mistake this demonstration for a connection to a production order system. I also compare descriptions with failure paths: wording that promises cancellation would be misleading if the operation still needs an independently recorded approval.&lt;/p&gt;

&lt;h2&gt;
  
  
  3. Failures and timeouts need an understandable contract
&lt;/h2&gt;

&lt;p&gt;I look for explicit failure responses and bounded network waits. A caller needs to distinguish an unsuccessful request from a successful result containing an error-shaped string.&lt;/p&gt;

&lt;p&gt;Textokit builds MCP error results with:&lt;br&gt;
&lt;/p&gt;

&lt;div class="highlight js-code-highlight"&gt;
&lt;pre class="highlight javascript"&gt;&lt;code&gt;&lt;span class="kd"&gt;const&lt;/span&gt; &lt;span class="nx"&gt;fail&lt;/span&gt; &lt;span class="o"&gt;=&lt;/span&gt; &lt;span class="p"&gt;(&lt;/span&gt;&lt;span class="nx"&gt;s&lt;/span&gt;&lt;span class="p"&gt;)&lt;/span&gt; &lt;span class="o"&gt;=&amp;gt;&lt;/span&gt; &lt;span class="p"&gt;({&lt;/span&gt; &lt;span class="p"&gt;...&lt;/span&gt;&lt;span class="nf"&gt;text&lt;/span&gt;&lt;span class="p"&gt;(&lt;/span&gt;&lt;span class="nx"&gt;s&lt;/span&gt;&lt;span class="p"&gt;),&lt;/span&gt; &lt;span class="na"&gt;isError&lt;/span&gt;&lt;span class="p"&gt;:&lt;/span&gt; &lt;span class="kc"&gt;true&lt;/span&gt; &lt;span class="p"&gt;});&lt;/span&gt;
&lt;/code&gt;&lt;/pre&gt;

&lt;/div&gt;



&lt;p&gt;Its payment requests include:&lt;br&gt;
&lt;/p&gt;

&lt;div class="highlight js-code-highlight"&gt;
&lt;pre class="highlight javascript"&gt;&lt;code&gt;&lt;span class="nx"&gt;signal&lt;/span&gt;&lt;span class="p"&gt;:&lt;/span&gt; &lt;span class="nx"&gt;AbortSignal&lt;/span&gt;&lt;span class="p"&gt;.&lt;/span&gt;&lt;span class="nf"&gt;timeout&lt;/span&gt;&lt;span class="p"&gt;(&lt;/span&gt;&lt;span class="nx"&gt;TIMEOUT_MS&lt;/span&gt;&lt;span class="p"&gt;),&lt;/span&gt;
&lt;/code&gt;&lt;/pre&gt;

&lt;/div&gt;



&lt;p&gt;&lt;a href="https://github.com/martinPisacreta/textokit-mcp/blob/fa04fb03e39ed3ac3da8f3a04e605749eeaaf3e6/index.mjs" rel="noopener noreferrer"&gt;Source: index.mjs, requests and handler&lt;/a&gt;.&lt;/p&gt;

&lt;p&gt;The timeout applies to individual fetch calls. It does not establish a single deadline for the entire payment flow, which can make an initial request and a signed retry. I would also inspect configuration validation and error sanitization: the handler forwards caught exception messages. A timeout response alone cannot establish whether a remote side effect occurred, so retry behavior needs its own review.&lt;/p&gt;

&lt;h2&gt;
  
  
  4. Secrets must stay out of results and logs
&lt;/h2&gt;

&lt;p&gt;I trace credentials from configuration into signing or authentication, then inspect output paths. That includes startup failures and exception messages, not only successful responses.&lt;/p&gt;

&lt;p&gt;Textokit reads its wallet credential from the environment:&lt;br&gt;
&lt;/p&gt;

&lt;div class="highlight js-code-highlight"&gt;
&lt;pre class="highlight javascript"&gt;&lt;code&gt;&lt;span class="kd"&gt;const&lt;/span&gt; &lt;span class="nx"&gt;rawKey&lt;/span&gt; &lt;span class="o"&gt;=&lt;/span&gt; &lt;span class="nx"&gt;process&lt;/span&gt;&lt;span class="p"&gt;.&lt;/span&gt;&lt;span class="nx"&gt;env&lt;/span&gt;&lt;span class="p"&gt;.&lt;/span&gt;&lt;span class="nx"&gt;TEXTOKIT_PRIVATE_KEY&lt;/span&gt;&lt;span class="p"&gt;;&lt;/span&gt;
&lt;/code&gt;&lt;/pre&gt;

&lt;/div&gt;



&lt;p&gt;Its startup log is:&lt;br&gt;
&lt;/p&gt;

&lt;div class="highlight js-code-highlight"&gt;
&lt;pre class="highlight javascript"&gt;&lt;code&gt;&lt;span class="nx"&gt;console&lt;/span&gt;&lt;span class="p"&gt;.&lt;/span&gt;&lt;span class="nf"&gt;error&lt;/span&gt;&lt;span class="p"&gt;(&lt;/span&gt;&lt;span class="s2"&gt;`textokit-mcp ready: &lt;/span&gt;&lt;span class="p"&gt;${&lt;/span&gt;&lt;span class="nx"&gt;paid&lt;/span&gt;&lt;span class="p"&gt;.&lt;/span&gt;&lt;span class="nx"&gt;length&lt;/span&gt;&lt;span class="p"&gt;}&lt;/span&gt;&lt;span class="s2"&gt; paid tools + &lt;/span&gt;&lt;span class="p"&gt;${&lt;/span&gt;&lt;span class="nx"&gt;free&lt;/span&gt;&lt;span class="p"&gt;.&lt;/span&gt;&lt;span class="nx"&gt;length&lt;/span&gt;&lt;span class="p"&gt;}&lt;/span&gt;&lt;span class="s2"&gt; free; wallet &lt;/span&gt;&lt;span class="p"&gt;${&lt;/span&gt;&lt;span class="nx"&gt;account&lt;/span&gt; &lt;span class="p"&gt;?&lt;/span&gt; &lt;span class="nx"&gt;account&lt;/span&gt;&lt;span class="p"&gt;.&lt;/span&gt;&lt;span class="nx"&gt;address&lt;/span&gt; &lt;span class="p"&gt;:&lt;/span&gt; &lt;span class="dl"&gt;'&lt;/span&gt;&lt;span class="s1"&gt;none (free tools only)&lt;/span&gt;&lt;span class="dl"&gt;'&lt;/span&gt;&lt;span class="p"&gt;}&lt;/span&gt;&lt;span class="s2"&gt;; cap &lt;/span&gt;&lt;span class="p"&gt;${&lt;/span&gt;&lt;span class="nx"&gt;MAX_CENTS&lt;/span&gt;&lt;span class="p"&gt;}&lt;/span&gt;&lt;span class="s2"&gt; cents per call`&lt;/span&gt;&lt;span class="p"&gt;);&lt;/span&gt;
&lt;/code&gt;&lt;/pre&gt;

&lt;/div&gt;



&lt;p&gt;&lt;a href="https://github.com/martinPisacreta/textokit-mcp/blob/fa04fb03e39ed3ac3da8f3a04e605749eeaaf3e6/index.mjs" rel="noopener noreferrer"&gt;Source: index.mjs&lt;/a&gt;.&lt;/p&gt;

&lt;p&gt;That log references the public address rather than the private key. This observation does not prove that every dependency or failure path protects secrets. I would still check returned upstream bodies, signing errors, and captured logs with synthetic sentinel credentials. No such execution is claimed here.&lt;/p&gt;

&lt;h2&gt;
  
  
  5. Writes and spending need authority outside model arguments
&lt;/h2&gt;

&lt;p&gt;For consequential tools, I ask where permission originates and what limits it. An argument saying “approved” is insufficient when the caller can supply that argument itself.&lt;/p&gt;

&lt;p&gt;The orders approval store checks an action, an exact order, and an expiration:&lt;br&gt;
&lt;/p&gt;

&lt;div class="highlight js-code-highlight"&gt;
&lt;pre class="highlight csharp"&gt;&lt;code&gt;&lt;span class="k"&gt;return&lt;/span&gt; &lt;span class="n"&gt;approval&lt;/span&gt; &lt;span class="k"&gt;is&lt;/span&gt; &lt;span class="n"&gt;not&lt;/span&gt; &lt;span class="k"&gt;null&lt;/span&gt; &lt;span class="p"&gt;&amp;amp;&amp;amp;&lt;/span&gt;
    &lt;span class="n"&gt;approval&lt;/span&gt;&lt;span class="p"&gt;.&lt;/span&gt;&lt;span class="n"&gt;Action&lt;/span&gt; &lt;span class="p"&gt;==&lt;/span&gt; &lt;span class="s"&gt;"cancel_order"&lt;/span&gt; &lt;span class="p"&gt;&amp;amp;&amp;amp;&lt;/span&gt;
    &lt;span class="n"&gt;approval&lt;/span&gt;&lt;span class="p"&gt;.&lt;/span&gt;&lt;span class="n"&gt;OrderId&lt;/span&gt; &lt;span class="p"&gt;==&lt;/span&gt; &lt;span class="n"&gt;orderId&lt;/span&gt; &lt;span class="p"&gt;&amp;amp;&amp;amp;&lt;/span&gt;
    &lt;span class="n"&gt;approval&lt;/span&gt;&lt;span class="p"&gt;.&lt;/span&gt;&lt;span class="n"&gt;ExpiresAt&lt;/span&gt; &lt;span class="p"&gt;&amp;gt;&lt;/span&gt; &lt;span class="n"&gt;DateTimeOffset&lt;/span&gt;&lt;span class="p"&gt;.&lt;/span&gt;&lt;span class="n"&gt;UtcNow&lt;/span&gt;&lt;span class="p"&gt;;&lt;/span&gt;
&lt;/code&gt;&lt;/pre&gt;

&lt;/div&gt;



&lt;p&gt;&lt;a href="https://github.com/martinPisacreta/dotnet-mcp-orders-server/blob/85ee9c52a5197e1108ffd48f8a61103fa51943c6/src/OrdersMcp/OrderService.cs#L17-L46" rel="noopener noreferrer"&gt;Source: OrderService.cs&lt;/a&gt;.&lt;/p&gt;

&lt;p&gt;The implementation claims the approval file through a rename and deletes the claimed file afterward. The tool description explicitly says the tool cannot create approvals. Deployment permissions must preserve that separation.&lt;/p&gt;

&lt;p&gt;Textokit's spending boundary is different: an externally configured wallet and per-call cap authorize automatic payment. It does not provide the orders server's independent, single-use approval workflow. I would document that distinction and review recipient restrictions and cumulative spending separately.&lt;/p&gt;

&lt;h2&gt;
  
  
  6. Untrusted inputs remain data
&lt;/h2&gt;

&lt;p&gt;Tool arguments, retrieved text, and upstream responses should not acquire authority merely because they appear in a conversation. I follow each value into paths, requests, and state changes.&lt;/p&gt;

&lt;p&gt;Before using an order ID in an approval filename, the orders implementation applies:&lt;br&gt;
&lt;/p&gt;

&lt;div class="highlight js-code-highlight"&gt;
&lt;pre class="highlight csharp"&gt;&lt;code&gt;&lt;span class="k"&gt;public&lt;/span&gt; &lt;span class="k"&gt;static&lt;/span&gt; &lt;span class="kt"&gt;bool&lt;/span&gt; &lt;span class="nf"&gt;IsSafeId&lt;/span&gt;&lt;span class="p"&gt;(&lt;/span&gt;&lt;span class="kt"&gt;string&lt;/span&gt; &lt;span class="n"&gt;id&lt;/span&gt;&lt;span class="p"&gt;)&lt;/span&gt; &lt;span class="p"&gt;=&amp;gt;&lt;/span&gt;
    &lt;span class="p"&gt;!&lt;/span&gt;&lt;span class="kt"&gt;string&lt;/span&gt;&lt;span class="p"&gt;.&lt;/span&gt;&lt;span class="nf"&gt;IsNullOrEmpty&lt;/span&gt;&lt;span class="p"&gt;(&lt;/span&gt;&lt;span class="n"&gt;id&lt;/span&gt;&lt;span class="p"&gt;)&lt;/span&gt; &lt;span class="p"&gt;&amp;amp;&amp;amp;&lt;/span&gt; &lt;span class="n"&gt;id&lt;/span&gt;&lt;span class="p"&gt;.&lt;/span&gt;&lt;span class="n"&gt;Length&lt;/span&gt; &lt;span class="p"&gt;&amp;lt;=&lt;/span&gt; &lt;span class="m"&gt;64&lt;/span&gt; &lt;span class="p"&gt;&amp;amp;&amp;amp;&lt;/span&gt;
    &lt;span class="n"&gt;id&lt;/span&gt;&lt;span class="p"&gt;.&lt;/span&gt;&lt;span class="nf"&gt;All&lt;/span&gt;&lt;span class="p"&gt;(&lt;/span&gt;&lt;span class="n"&gt;c&lt;/span&gt; &lt;span class="p"&gt;=&amp;gt;&lt;/span&gt; &lt;span class="kt"&gt;char&lt;/span&gt;&lt;span class="p"&gt;.&lt;/span&gt;&lt;span class="nf"&gt;IsAsciiLetterOrDigit&lt;/span&gt;&lt;span class="p"&gt;(&lt;/span&gt;&lt;span class="n"&gt;c&lt;/span&gt;&lt;span class="p"&gt;)&lt;/span&gt; &lt;span class="p"&gt;||&lt;/span&gt; &lt;span class="n"&gt;c&lt;/span&gt; &lt;span class="p"&gt;==&lt;/span&gt; &lt;span class="sc"&gt;'-'&lt;/span&gt;&lt;span class="p"&gt;);&lt;/span&gt;
&lt;/code&gt;&lt;/pre&gt;

&lt;/div&gt;



&lt;p&gt;&lt;a href="https://github.com/martinPisacreta/dotnet-mcp-orders-server/blob/85ee9c52a5197e1108ffd48f8a61103fa51943c6/src/OrdersMcp/OrderService.cs#L13-L20" rel="noopener noreferrer"&gt;Source: OrderService.cs&lt;/a&gt;.&lt;/p&gt;

&lt;p&gt;This restricts the filename component; it is not a general prompt-injection defense. The broader review asks whether external content can bypass approval, influence destinations, or become executable instructions. An allowlist at one boundary provides evidence about that boundary only.&lt;/p&gt;

&lt;h2&gt;
  
  
  7. Tests should check denied actions and unchanged state
&lt;/h2&gt;

&lt;p&gt;I read assertions before summarizing test coverage. A test name is weaker evidence than a check that an unauthorized operation leaves the original state intact.&lt;/p&gt;

&lt;p&gt;The orders tests include:&lt;br&gt;
&lt;/p&gt;

&lt;div class="highlight js-code-highlight"&gt;
&lt;pre class="highlight csharp"&gt;&lt;code&gt;&lt;span class="kt"&gt;var&lt;/span&gt; &lt;span class="n"&gt;before&lt;/span&gt; &lt;span class="p"&gt;=&lt;/span&gt; &lt;span class="n"&gt;tools&lt;/span&gt;&lt;span class="p"&gt;.&lt;/span&gt;&lt;span class="nf"&gt;GetOrder&lt;/span&gt;&lt;span class="p"&gt;(&lt;/span&gt;&lt;span class="s"&gt;"ORD-1001"&lt;/span&gt;&lt;span class="p"&gt;);&lt;/span&gt;
&lt;span class="kt"&gt;var&lt;/span&gt; &lt;span class="n"&gt;error&lt;/span&gt; &lt;span class="p"&gt;=&lt;/span&gt; &lt;span class="n"&gt;Assert&lt;/span&gt;&lt;span class="p"&gt;.&lt;/span&gt;&lt;span class="n"&gt;Throws&lt;/span&gt;&lt;span class="p"&gt;&amp;lt;&lt;/span&gt;&lt;span class="n"&gt;McpException&lt;/span&gt;&lt;span class="p"&gt;&amp;gt;(()&lt;/span&gt; &lt;span class="p"&gt;=&amp;gt;&lt;/span&gt; &lt;span class="n"&gt;tools&lt;/span&gt;&lt;span class="p"&gt;.&lt;/span&gt;&lt;span class="nf"&gt;CancelOrder&lt;/span&gt;&lt;span class="p"&gt;(&lt;/span&gt;&lt;span class="s"&gt;"ORD-1001"&lt;/span&gt;&lt;span class="p"&gt;));&lt;/span&gt;
&lt;span class="n"&gt;Assert&lt;/span&gt;&lt;span class="p"&gt;.&lt;/span&gt;&lt;span class="nf"&gt;Contains&lt;/span&gt;&lt;span class="p"&gt;(&lt;/span&gt;&lt;span class="s"&gt;"No order was changed"&lt;/span&gt;&lt;span class="p"&gt;,&lt;/span&gt; &lt;span class="n"&gt;error&lt;/span&gt;&lt;span class="p"&gt;.&lt;/span&gt;&lt;span class="n"&gt;Message&lt;/span&gt;&lt;span class="p"&gt;);&lt;/span&gt;
&lt;span class="n"&gt;Assert&lt;/span&gt;&lt;span class="p"&gt;.&lt;/span&gt;&lt;span class="nf"&gt;Equal&lt;/span&gt;&lt;span class="p"&gt;(&lt;/span&gt;&lt;span class="n"&gt;before&lt;/span&gt;&lt;span class="p"&gt;,&lt;/span&gt; &lt;span class="n"&gt;tools&lt;/span&gt;&lt;span class="p"&gt;.&lt;/span&gt;&lt;span class="nf"&gt;GetOrder&lt;/span&gt;&lt;span class="p"&gt;(&lt;/span&gt;&lt;span class="s"&gt;"ORD-1001"&lt;/span&gt;&lt;span class="p"&gt;));&lt;/span&gt;
&lt;/code&gt;&lt;/pre&gt;

&lt;/div&gt;



&lt;p&gt;&lt;a href="https://github.com/martinPisacreta/dotnet-mcp-orders-server/blob/85ee9c52a5197e1108ffd48f8a61103fa51943c6/tests/OrdersMcp.Tests/UnitTest1.cs" rel="noopener noreferrer"&gt;Source: UnitTest1.cs&lt;/a&gt;.&lt;/p&gt;

&lt;p&gt;That file also defines cases for expired, mismatched, malformed, and consumed approvals. Textokit's &lt;a href="https://github.com/martinPisacreta/textokit-mcp/blob/fa04fb03e39ed3ac3da8f3a04e605749eeaaf3e6/test/smoke.mjs" rel="noopener noreferrer"&gt;smoke test&lt;/a&gt; exercises initialization and tool discovery; it does not establish payment correctness or secret protection.&lt;/p&gt;

&lt;p&gt;A review report should distinguish source observations, executable test definitions, and observed execution results. Each finding needs a concrete consequence, supporting evidence, and a next action that fits the server's actual scope.&lt;/p&gt;

&lt;p&gt;If you want this checklist applied to your server, I offer a fixed-price review in 48 hours on Upwork: &lt;a href="https://www.upwork.com/services/product/2116921806127265220" rel="noopener noreferrer"&gt;https://www.upwork.com/services/product/2116921806127265220&lt;/a&gt;&lt;/p&gt;

</description>
      <category>mcp</category>
      <category>ai</category>
      <category>dotnet</category>
      <category>security</category>
    </item>
    <item>
      <title>Inside My x402 Store: Payment Requirements, Settlement, and Receipts</title>
      <dc:creator>Martin Pisacreta</dc:creator>
      <pubDate>Thu, 29 Oct 2026 13:15:24 +0000</pubDate>
      <link>https://dev.to/martinpisacreta/inside-my-x402-store-payment-requirements-settlement-and-receipts-3gdd</link>
      <guid>https://dev.to/martinpisacreta/inside-my-x402-store-payment-requirements-settlement-and-receipts-3gdd</guid>
      <description>&lt;p&gt;I run Textokit with Automata, my autonomous agent. The store exposes services over HTTP and accepts per-call payments in USDC on Base. Its Node.js implementation gives me a concrete sequence to inspect: the payment challenge, verification, service execution, settlement, and the response returned to the caller.&lt;/p&gt;

&lt;p&gt;The commercial result needs to come first. Our assessment as of October 29, 2026 is zero real buyers, with traffic dominated by indexers. Internal payments, probes, and automated sweeps do not establish customer demand. The payment infrastructure exists, while the business still needs evidence.&lt;/p&gt;

&lt;p&gt;This article follows the existing &lt;code&gt;public/server.js&lt;/code&gt; implementation. I did not make a purchase or run a new settlement to write it. The excerpts are from that source, with surrounding branches omitted.&lt;/p&gt;

&lt;h2&gt;
  
  
  A request starts with a service route
&lt;/h2&gt;

&lt;p&gt;The store matches &lt;code&gt;/api/&amp;lt;service&amp;gt;&lt;/code&gt; against its loaded service definitions. Unknown services return 404. The execution path accepts POST requests, reads a bounded body, and passes the selected service to a shared payment handler.&lt;/p&gt;

&lt;p&gt;For example, the catalog documents &lt;code&gt;POST https://tienda.textokit.store/api/translate-es&lt;/code&gt;. A separate sample endpoint describes the service without purchasing or executing it. That gives callers a way to inspect the expected input and output before paying.&lt;/p&gt;

&lt;p&gt;Inside &lt;code&gt;paidCall&lt;/code&gt;, the incoming headers select the payment format:&lt;br&gt;
&lt;/p&gt;

&lt;div class="highlight js-code-highlight"&gt;
&lt;pre class="highlight javascript"&gt;&lt;code&gt;&lt;span class="kd"&gt;const&lt;/span&gt; &lt;span class="nx"&gt;v2Header&lt;/span&gt; &lt;span class="o"&gt;=&lt;/span&gt; &lt;span class="nx"&gt;req&lt;/span&gt;&lt;span class="p"&gt;.&lt;/span&gt;&lt;span class="nx"&gt;headers&lt;/span&gt;&lt;span class="p"&gt;[&lt;/span&gt;&lt;span class="dl"&gt;'&lt;/span&gt;&lt;span class="s1"&gt;payment-signature&lt;/span&gt;&lt;span class="dl"&gt;'&lt;/span&gt;&lt;span class="p"&gt;];&lt;/span&gt;
&lt;span class="kd"&gt;const&lt;/span&gt; &lt;span class="nx"&gt;version&lt;/span&gt; &lt;span class="o"&gt;=&lt;/span&gt; &lt;span class="nx"&gt;v2Header&lt;/span&gt; &lt;span class="p"&gt;?&lt;/span&gt; &lt;span class="mi"&gt;2&lt;/span&gt; &lt;span class="p"&gt;:&lt;/span&gt; &lt;span class="mi"&gt;1&lt;/span&gt;&lt;span class="p"&gt;;&lt;/span&gt;
&lt;span class="kd"&gt;const&lt;/span&gt; &lt;span class="nx"&gt;requirements&lt;/span&gt; &lt;span class="o"&gt;=&lt;/span&gt; &lt;span class="nx"&gt;version&lt;/span&gt; &lt;span class="o"&gt;===&lt;/span&gt; &lt;span class="mi"&gt;2&lt;/span&gt; &lt;span class="p"&gt;?&lt;/span&gt; &lt;span class="nf"&gt;requirementsV2&lt;/span&gt;&lt;span class="p"&gt;(&lt;/span&gt;&lt;span class="nx"&gt;svc&lt;/span&gt;&lt;span class="p"&gt;)&lt;/span&gt; &lt;span class="p"&gt;:&lt;/span&gt; &lt;span class="nf"&gt;requirementsV1&lt;/span&gt;&lt;span class="p"&gt;(&lt;/span&gt;&lt;span class="nx"&gt;svc&lt;/span&gt;&lt;span class="p"&gt;);&lt;/span&gt;
&lt;span class="kd"&gt;const&lt;/span&gt; &lt;span class="nx"&gt;header&lt;/span&gt; &lt;span class="o"&gt;=&lt;/span&gt; &lt;span class="nx"&gt;v2Header&lt;/span&gt; &lt;span class="o"&gt;||&lt;/span&gt; &lt;span class="nx"&gt;req&lt;/span&gt;&lt;span class="p"&gt;.&lt;/span&gt;&lt;span class="nx"&gt;headers&lt;/span&gt;&lt;span class="p"&gt;[&lt;/span&gt;&lt;span class="dl"&gt;'&lt;/span&gt;&lt;span class="s1"&gt;x-payment&lt;/span&gt;&lt;span class="dl"&gt;'&lt;/span&gt;&lt;span class="p"&gt;];&lt;/span&gt;
&lt;/code&gt;&lt;/pre&gt;

&lt;/div&gt;



&lt;p&gt;Without either payment header, the handler returns HTTP 402. This store preserves a compatibility detail: the JSON body contains v1 requirements, while the &lt;code&gt;PAYMENT-REQUIRED&lt;/code&gt; header carries a base64-encoded v2 challenge. Clients need to consume the format they support.&lt;/p&gt;

&lt;p&gt;The v2 requirements specify the &lt;code&gt;exact&lt;/code&gt; scheme, Base as &lt;code&gt;eip155:8453&lt;/code&gt;, the amount in atomic units, the USDC asset, the recipient, and a maximum timeout of 300 seconds. The challenge also includes resource metadata and a Bazaar extension:&lt;br&gt;
&lt;/p&gt;

&lt;div class="highlight js-code-highlight"&gt;
&lt;pre class="highlight javascript"&gt;&lt;code&gt;&lt;span class="kd"&gt;const&lt;/span&gt; &lt;span class="nx"&gt;paymentRequired&lt;/span&gt; &lt;span class="o"&gt;=&lt;/span&gt; &lt;span class="p"&gt;(&lt;/span&gt;&lt;span class="nx"&gt;svc&lt;/span&gt;&lt;span class="p"&gt;,&lt;/span&gt; &lt;span class="nx"&gt;error&lt;/span&gt;&lt;span class="p"&gt;)&lt;/span&gt; &lt;span class="o"&gt;=&amp;gt;&lt;/span&gt; &lt;span class="p"&gt;({&lt;/span&gt; &lt;span class="na"&gt;x402Version&lt;/span&gt;&lt;span class="p"&gt;:&lt;/span&gt; &lt;span class="mi"&gt;2&lt;/span&gt;&lt;span class="p"&gt;,&lt;/span&gt; &lt;span class="nx"&gt;error&lt;/span&gt;&lt;span class="p"&gt;,&lt;/span&gt; &lt;span class="na"&gt;resource&lt;/span&gt;&lt;span class="p"&gt;:&lt;/span&gt; &lt;span class="nf"&gt;resourceV2&lt;/span&gt;&lt;span class="p"&gt;(&lt;/span&gt;&lt;span class="nx"&gt;svc&lt;/span&gt;&lt;span class="p"&gt;),&lt;/span&gt; &lt;span class="na"&gt;accepts&lt;/span&gt;&lt;span class="p"&gt;:&lt;/span&gt; &lt;span class="p"&gt;[&lt;/span&gt;&lt;span class="nf"&gt;requirementsV2&lt;/span&gt;&lt;span class="p"&gt;(&lt;/span&gt;&lt;span class="nx"&gt;svc&lt;/span&gt;&lt;span class="p"&gt;)],&lt;/span&gt; &lt;span class="na"&gt;extensions&lt;/span&gt;&lt;span class="p"&gt;:&lt;/span&gt; &lt;span class="p"&gt;{&lt;/span&gt; &lt;span class="na"&gt;bazaar&lt;/span&gt;&lt;span class="p"&gt;:&lt;/span&gt; &lt;span class="nf"&gt;bazaar&lt;/span&gt;&lt;span class="p"&gt;(&lt;/span&gt;&lt;span class="nx"&gt;svc&lt;/span&gt;&lt;span class="p"&gt;)&lt;/span&gt; &lt;span class="p"&gt;}&lt;/span&gt; &lt;span class="p"&gt;});&lt;/span&gt;
&lt;/code&gt;&lt;/pre&gt;

&lt;/div&gt;



&lt;p&gt;These are details of this implementation. They are useful when debugging this store, but they do not establish how every x402 server behaves.&lt;/p&gt;

&lt;h2&gt;
  
  
  Verification precedes execution
&lt;/h2&gt;

&lt;p&gt;The caller repeats the request with a signed payment payload. The handler validates the service input, checks daily capacity, decodes the payment header, and prepares the facilitator request. Invalid input and malformed payment headers do not reach settlement.&lt;/p&gt;

&lt;p&gt;There are compatibility branches too. The store normalizes a v1 network label and can forward a v1 authorization in a v2 envelope when its configured facilitator is Coinbase. Those branches belong to this integration and deserve separate review before reuse.&lt;/p&gt;

&lt;p&gt;After that preparation, verification is explicit:&lt;br&gt;
&lt;/p&gt;

&lt;div class="highlight js-code-highlight"&gt;
&lt;pre class="highlight javascript"&gt;&lt;code&gt;&lt;span class="kd"&gt;const&lt;/span&gt; &lt;span class="nx"&gt;verified&lt;/span&gt; &lt;span class="o"&gt;=&lt;/span&gt; &lt;span class="k"&gt;await&lt;/span&gt; &lt;span class="nf"&gt;facilitator&lt;/span&gt;&lt;span class="p"&gt;(&lt;/span&gt;&lt;span class="dl"&gt;'&lt;/span&gt;&lt;span class="s1"&gt;verify&lt;/span&gt;&lt;span class="dl"&gt;'&lt;/span&gt;&lt;span class="p"&gt;,&lt;/span&gt; &lt;span class="nx"&gt;facPayment&lt;/span&gt;&lt;span class="p"&gt;,&lt;/span&gt; &lt;span class="nx"&gt;facRequirements&lt;/span&gt;&lt;span class="p"&gt;,&lt;/span&gt; &lt;span class="nx"&gt;facVersion&lt;/span&gt;&lt;span class="p"&gt;);&lt;/span&gt;
&lt;/code&gt;&lt;/pre&gt;

&lt;/div&gt;



&lt;p&gt;The handler checks &lt;code&gt;verified.isValid&lt;/code&gt;. Failure creates a &lt;code&gt;verify_failed_not_settled&lt;/code&gt; log entry and another payment challenge. Successful verification allows the next stage. It does not establish that settlement or response delivery has happened.&lt;/p&gt;

&lt;h2&gt;
  
  
  The store does the work before settling
&lt;/h2&gt;

&lt;p&gt;Once verification passes, the handler executes the service:&lt;br&gt;
&lt;/p&gt;

&lt;div class="highlight js-code-highlight"&gt;
&lt;pre class="highlight javascript"&gt;&lt;code&gt;&lt;span class="kd"&gt;const&lt;/span&gt; &lt;span class="nx"&gt;result&lt;/span&gt; &lt;span class="o"&gt;=&lt;/span&gt; &lt;span class="k"&gt;await&lt;/span&gt; &lt;span class="nf"&gt;execute&lt;/span&gt;&lt;span class="p"&gt;(&lt;/span&gt;&lt;span class="nx"&gt;svc&lt;/span&gt;&lt;span class="p"&gt;,&lt;/span&gt; &lt;span class="nx"&gt;input&lt;/span&gt;&lt;span class="p"&gt;);&lt;/span&gt;
&lt;/code&gt;&lt;/pre&gt;

&lt;/div&gt;



&lt;p&gt;If execution reports failure, the handler records &lt;code&gt;work_failed_not_settled&lt;/code&gt; and returns 503. That branch does not call settlement.&lt;/p&gt;

&lt;p&gt;This ordering has a cost for the operator: work can consume resources without producing a settled sale. It also means retries need careful treatment, especially if a service has side effects. The excerpts here do not demonstrate an exactly-once execution guarantee.&lt;/p&gt;

&lt;p&gt;Successful execution reaches the next facilitator call:&lt;br&gt;
&lt;/p&gt;

&lt;div class="highlight js-code-highlight"&gt;
&lt;pre class="highlight javascript"&gt;&lt;code&gt;&lt;span class="kd"&gt;const&lt;/span&gt; &lt;span class="nx"&gt;settled&lt;/span&gt; &lt;span class="o"&gt;=&lt;/span&gt; &lt;span class="k"&gt;await&lt;/span&gt; &lt;span class="nf"&gt;facilitator&lt;/span&gt;&lt;span class="p"&gt;(&lt;/span&gt;&lt;span class="dl"&gt;'&lt;/span&gt;&lt;span class="s1"&gt;settle&lt;/span&gt;&lt;span class="dl"&gt;'&lt;/span&gt;&lt;span class="p"&gt;,&lt;/span&gt; &lt;span class="nx"&gt;facPayment&lt;/span&gt;&lt;span class="p"&gt;,&lt;/span&gt; &lt;span class="nx"&gt;facRequirements&lt;/span&gt;&lt;span class="p"&gt;,&lt;/span&gt; &lt;span class="nx"&gt;facVersion&lt;/span&gt;&lt;span class="p"&gt;);&lt;/span&gt;
&lt;/code&gt;&lt;/pre&gt;

&lt;/div&gt;



&lt;p&gt;The handler checks &lt;code&gt;settled.success&lt;/code&gt;. Failure returns a payment challenge. Success creates a paid log entry containing the service, amount, payer, transaction, facilitator information, and protocol version.&lt;/p&gt;

&lt;p&gt;Verification, execution, settlement, and delivery remain separate events. A failure between them can leave a caller uncertain about what happened. That matters when deciding whether to retry.&lt;/p&gt;

&lt;h2&gt;
  
  
  The response and the public receipt
&lt;/h2&gt;

&lt;p&gt;After successful settlement, the response carries the service result in its JSON body and the settlement response in a header:&lt;br&gt;
&lt;/p&gt;

&lt;div class="highlight js-code-highlight"&gt;
&lt;pre class="highlight javascript"&gt;&lt;code&gt;&lt;span class="k"&gt;return&lt;/span&gt; &lt;span class="nf"&gt;json&lt;/span&gt;&lt;span class="p"&gt;(&lt;/span&gt;&lt;span class="nx"&gt;res&lt;/span&gt;&lt;span class="p"&gt;,&lt;/span&gt; &lt;span class="mi"&gt;200&lt;/span&gt;&lt;span class="p"&gt;,&lt;/span&gt; &lt;span class="nx"&gt;result&lt;/span&gt;&lt;span class="p"&gt;.&lt;/span&gt;&lt;span class="nx"&gt;value&lt;/span&gt;&lt;span class="p"&gt;,&lt;/span&gt; &lt;span class="p"&gt;{&lt;/span&gt; &lt;span class="p"&gt;[&lt;/span&gt;&lt;span class="nx"&gt;version&lt;/span&gt; &lt;span class="o"&gt;===&lt;/span&gt; &lt;span class="mi"&gt;2&lt;/span&gt; &lt;span class="p"&gt;?&lt;/span&gt; &lt;span class="dl"&gt;'&lt;/span&gt;&lt;span class="s1"&gt;PAYMENT-RESPONSE&lt;/span&gt;&lt;span class="dl"&gt;'&lt;/span&gt; &lt;span class="p"&gt;:&lt;/span&gt; &lt;span class="dl"&gt;'&lt;/span&gt;&lt;span class="s1"&gt;X-PAYMENT-RESPONSE&lt;/span&gt;&lt;span class="dl"&gt;'&lt;/span&gt;&lt;span class="p"&gt;]:&lt;/span&gt; &lt;span class="nf"&gt;b64&lt;/span&gt;&lt;span class="p"&gt;(&lt;/span&gt;&lt;span class="nx"&gt;settled&lt;/span&gt;&lt;span class="p"&gt;)&lt;/span&gt; &lt;span class="p"&gt;});&lt;/span&gt;
&lt;/code&gt;&lt;/pre&gt;

&lt;/div&gt;



&lt;p&gt;The header contains the facilitator's settlement response. The application result stays in the body. A paid log entry alone does not prove that the caller received and consumed either one.&lt;/p&gt;

&lt;p&gt;The store also exposes a &lt;a href="https://tienda.textokit.store/receipts.json" rel="noopener noreferrer"&gt;public receipt feed&lt;/a&gt;. Its implementation reads the local sales log and selects records whose outcome is &lt;code&gt;paid&lt;/code&gt;. The projection includes the service, atomic amount, payer, transaction, facilitator, and an &lt;code&gt;ownWallet&lt;/code&gt; flag. Agent purchases appear separately.&lt;/p&gt;

&lt;p&gt;That ownership flag prevents known internal settlements from being counted as customers. It has limits: an external wallet could still be a probe or a sweeper. The feed records settled activity according to the store's logs. It does not independently establish customer satisfaction, recurring demand, or successful delivery.&lt;/p&gt;

&lt;h2&gt;
  
  
  Where MCP connects
&lt;/h2&gt;

&lt;p&gt;The storefront contains an integration with a remote MCP Worker. After reading an API request body, it derives a context from the Worker's signed request and chooses the execution path:&lt;br&gt;
&lt;/p&gt;

&lt;div class="highlight js-code-highlight"&gt;
&lt;pre class="highlight javascript"&gt;&lt;code&gt;&lt;span class="kd"&gt;const&lt;/span&gt; &lt;span class="nx"&gt;mcpCtx&lt;/span&gt; &lt;span class="o"&gt;=&lt;/span&gt; &lt;span class="nf"&gt;freeTierContext&lt;/span&gt;&lt;span class="p"&gt;(&lt;/span&gt;&lt;span class="nx"&gt;req&lt;/span&gt;&lt;span class="p"&gt;,&lt;/span&gt; &lt;span class="dl"&gt;'&lt;/span&gt;&lt;span class="s1"&gt;POST&lt;/span&gt;&lt;span class="dl"&gt;'&lt;/span&gt;&lt;span class="p"&gt;,&lt;/span&gt; &lt;span class="nx"&gt;p&lt;/span&gt;&lt;span class="p"&gt;,&lt;/span&gt; &lt;span class="nx"&gt;body&lt;/span&gt;&lt;span class="p"&gt;);&lt;/span&gt;
&lt;span class="k"&gt;return&lt;/span&gt; &lt;span class="k"&gt;await &lt;/span&gt;&lt;span class="p"&gt;(&lt;/span&gt;&lt;span class="nx"&gt;mcpCtx&lt;/span&gt; &lt;span class="p"&gt;?&lt;/span&gt; &lt;span class="nf"&gt;mcpPaidCall&lt;/span&gt;&lt;span class="p"&gt;(&lt;/span&gt;&lt;span class="nx"&gt;svc&lt;/span&gt;&lt;span class="p"&gt;,&lt;/span&gt; &lt;span class="nx"&gt;req&lt;/span&gt;&lt;span class="p"&gt;,&lt;/span&gt; &lt;span class="nx"&gt;res&lt;/span&gt;&lt;span class="p"&gt;,&lt;/span&gt; &lt;span class="nx"&gt;body&lt;/span&gt;&lt;span class="p"&gt;,&lt;/span&gt; &lt;span class="nx"&gt;mcpCtx&lt;/span&gt;&lt;span class="p"&gt;)&lt;/span&gt; &lt;span class="p"&gt;:&lt;/span&gt; &lt;span class="nf"&gt;paidCall&lt;/span&gt;&lt;span class="p"&gt;(&lt;/span&gt;&lt;span class="nx"&gt;svc&lt;/span&gt;&lt;span class="p"&gt;,&lt;/span&gt; &lt;span class="nx"&gt;req&lt;/span&gt;&lt;span class="p"&gt;,&lt;/span&gt; &lt;span class="nx"&gt;res&lt;/span&gt;&lt;span class="p"&gt;,&lt;/span&gt; &lt;span class="nx"&gt;body&lt;/span&gt;&lt;span class="p"&gt;));&lt;/span&gt;
&lt;/code&gt;&lt;/pre&gt;

&lt;/div&gt;



&lt;p&gt;Within that integration, requests carrying payment headers continue through the paid handler. Eligible requests without payment can use a limited free allowance. When the allowance is unavailable, they can reach the payment challenge.&lt;/p&gt;

&lt;p&gt;Free executions use a separate usage log and are excluded from the settled-sales log that feeds public receipts. The storefront also accepts authenticated Worker events for initialization, tool discovery, and tool activity. Those events measure access and usage, rather than revenue.&lt;/p&gt;

&lt;p&gt;The public MCP endpoint is &lt;code&gt;https://tienda.textokit.store/mcp&lt;/code&gt;. On October 29, 2026, I checked it using only &lt;code&gt;initialize&lt;/code&gt;, the initialization notification, and &lt;code&gt;tools/list&lt;/code&gt;. Discovery required no authentication and returned 23 tools, including &lt;code&gt;textokit_catalog&lt;/code&gt;, &lt;code&gt;textokit_sample&lt;/code&gt;, &lt;code&gt;textokit_how_to_pay&lt;/code&gt;, &lt;code&gt;textokit_translate_es&lt;/code&gt;, and &lt;code&gt;textokit_base_usdc_receipts&lt;/code&gt;. These names were returned by discovery. I did not invoke any tools or make a payment.&lt;/p&gt;

&lt;p&gt;For a remote MCP client that supports HTTP entries under &lt;code&gt;mcpServers&lt;/code&gt;, the connection configuration can look like this:&lt;br&gt;
&lt;/p&gt;

&lt;div class="highlight js-code-highlight"&gt;
&lt;pre class="highlight json"&gt;&lt;code&gt;&lt;span class="p"&gt;{&lt;/span&gt;&lt;span class="w"&gt;
  &lt;/span&gt;&lt;span class="nl"&gt;"mcpServers"&lt;/span&gt;&lt;span class="p"&gt;:&lt;/span&gt;&lt;span class="w"&gt; &lt;/span&gt;&lt;span class="p"&gt;{&lt;/span&gt;&lt;span class="w"&gt;
    &lt;/span&gt;&lt;span class="nl"&gt;"textokit"&lt;/span&gt;&lt;span class="p"&gt;:&lt;/span&gt;&lt;span class="w"&gt; &lt;/span&gt;&lt;span class="p"&gt;{&lt;/span&gt;&lt;span class="w"&gt;
      &lt;/span&gt;&lt;span class="nl"&gt;"type"&lt;/span&gt;&lt;span class="p"&gt;:&lt;/span&gt;&lt;span class="w"&gt; &lt;/span&gt;&lt;span class="s2"&gt;"http"&lt;/span&gt;&lt;span class="p"&gt;,&lt;/span&gt;&lt;span class="w"&gt;
      &lt;/span&gt;&lt;span class="nl"&gt;"url"&lt;/span&gt;&lt;span class="p"&gt;:&lt;/span&gt;&lt;span class="w"&gt; &lt;/span&gt;&lt;span class="s2"&gt;"https://tienda.textokit.store/mcp"&lt;/span&gt;&lt;span class="w"&gt;
    &lt;/span&gt;&lt;span class="p"&gt;}&lt;/span&gt;&lt;span class="w"&gt;
  &lt;/span&gt;&lt;span class="p"&gt;}&lt;/span&gt;&lt;span class="w"&gt;
&lt;/span&gt;&lt;span class="p"&gt;}&lt;/span&gt;&lt;span class="w"&gt;
&lt;/span&gt;&lt;/code&gt;&lt;/pre&gt;

&lt;/div&gt;



&lt;p&gt;Client configuration formats vary, so adapt the wrapper to your client's settings. This example supplies the verified endpoint, but I have not tested a client connection with this configuration. Unauthenticated discovery does not imply that every tool executes for free.&lt;/p&gt;

&lt;h2&gt;
  
  
  What the implementation establishes
&lt;/h2&gt;

&lt;p&gt;The source shows a concrete path from payment requirements through verification, execution, settlement, and a response. It also separates free usage, failed work, and paid log entries. This review verified the remote MCP tool catalog through discovery. It did not add a new end-to-end payment test, invoke MCP tools, or test the example client configuration.&lt;/p&gt;

&lt;p&gt;The remaining commercial question is whether someone has a recurring problem worth paying this API to solve. So far, our assessment remains zero real buyers. Keeping that result beside the code makes the technical progress and its practical limits easier to judge.&lt;/p&gt;

&lt;p&gt;If you want a pay-per-call endpoint like this for your own API, I take fixed-price work on Upwork: &lt;a href="https://www.upwork.com/services/product/2114214673238319567" rel="noopener noreferrer"&gt;a pay-per-call x402 API that AI agents pay in USDC&lt;/a&gt;.&lt;/p&gt;

</description>
      <category>webdev</category>
      <category>javascript</category>
      <category>ai</category>
      <category>blockchain</category>
    </item>
  </channel>
</rss>
