<?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: Yalakala-Naveen</title>
    <description>The latest articles on DEV Community by Yalakala-Naveen (@yalakala-naveen).</description>
    <link>https://dev.to/yalakala-naveen</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%2F4246730%2Facc665bf-6d86-4b38-95e4-5c9f5e55aaa1.png</url>
      <title>DEV Community: Yalakala-Naveen</title>
      <link>https://dev.to/yalakala-naveen</link>
    </image>
    <atom:link rel="self" type="application/rss+xml" href="https://dev.to/feed/yalakala-naveen"/>
    <language>en</language>
    <item>
      <title>Every Excel export you receive is broken the same five ways</title>
      <dc:creator>Yalakala-Naveen</dc:creator>
      <pubDate>Tue, 17 Nov 2026 02:38:55 +0000</pubDate>
      <link>https://dev.to/yalakala-naveen/every-excel-export-you-receive-is-broken-the-same-five-ways-5b8o</link>
      <guid>https://dev.to/yalakala-naveen/every-excel-export-you-receive-is-broken-the-same-five-ways-5b8o</guid>
      <description>&lt;h1&gt;
  
  
  Every Excel export you receive is broken the same five ways
&lt;/h1&gt;

&lt;p&gt;The file lands at 4:58 PM. The analysis is due at 6.&lt;/p&gt;

&lt;p&gt;Nothing is wrong with the data. Everything is wrong with the file.&lt;/p&gt;

&lt;p&gt;There is a title row merged across the top. A blank row under it. A header split over two rows, with "Q1" merged across three columns. A totals row baked into the bottom of the data. A footnote sitting where a data row should be.&lt;/p&gt;

&lt;p&gt;&lt;code&gt;pandas.read_excel&lt;/code&gt; reads all of it exactly as instructed — which is the problem. Pandas does not detect any of this structure. You must already know the mess before you can describe it with &lt;code&gt;header=&lt;/code&gt;, &lt;code&gt;skiprows=&lt;/code&gt; and &lt;code&gt;usecols=&lt;/code&gt;. So the columns come out named &lt;code&gt;Unnamed: 0&lt;/code&gt; through &lt;code&gt;Unnamed: 7&lt;/code&gt;, the totals row sits inside your sums quietly doubling them, and the footnote becomes somebody's region.&lt;/p&gt;

&lt;p&gt;Every analyst I know has the same drawer of glue for this. I had it too:&lt;/p&gt;

&lt;ul&gt;
&lt;li&gt;
&lt;code&gt;skiprows=3&lt;/code&gt; for the finance export — until March, when someone inserted a row into the title block and the script silently read the wrong header for a week.&lt;/li&gt;
&lt;li&gt;
&lt;code&gt;header=[0, 1]&lt;/code&gt; plus a hand-written flatten function for the two-row headers, with an &lt;code&gt;Unnamed:&lt;/code&gt; filter because Excel stores a merged cell's value only in its top-left corner.&lt;/li&gt;
&lt;li&gt;An openpyxl pre-pass that walks &lt;code&gt;ws.merged_cells.ranges&lt;/code&gt;, unmerges everything and fills the top-left value across, because &lt;code&gt;ffill&lt;/code&gt; after the fact guesses wrong on horizontal merges.&lt;/li&gt;
&lt;li&gt;A row filter that hunts the word "Total" so the totals row stops inflating the numbers.&lt;/li&gt;
&lt;li&gt;A prayer for the footnote rows.&lt;/li&gt;
&lt;/ul&gt;

&lt;p&gt;Five fixes. Rewritten per file, per sender, forever. pandas recorded 688 million downloads last month and openpyxl 349 million — this hour of glue is being re-lived at industrial scale, and the feature request for pandas to fill merged areas on read (pandas #63684, opened January 2026) is still sitting open, because it does not exist.&lt;/p&gt;

&lt;p&gt;So I stopped rewriting the glue and built the detector.&lt;/p&gt;

&lt;h2&gt;
  
  
  tidyexcel
&lt;/h2&gt;



&lt;div class="highlight js-code-highlight"&gt;
&lt;pre class="highlight python"&gt;&lt;code&gt;&lt;span class="k"&gt;import&lt;/span&gt; &lt;span class="n"&gt;tidy_sheet&lt;/span&gt;

&lt;span class="n"&gt;df&lt;/span&gt;&lt;span class="p"&gt;,&lt;/span&gt; &lt;span class="n"&gt;report&lt;/span&gt; &lt;span class="o"&gt;=&lt;/span&gt; &lt;span class="n"&gt;tidy_sheet&lt;/span&gt;&lt;span class="p"&gt;.&lt;/span&gt;&lt;span class="nf"&gt;read_excel&lt;/span&gt;&lt;span class="p"&gt;(&lt;/span&gt;&lt;span class="s2"&gt;"q3-report.xlsx"&lt;/span&gt;&lt;span class="p"&gt;)&lt;/span&gt;   &lt;span class="c1"&gt;# or read_csv(...)&lt;/span&gt;

&lt;span class="n"&gt;df&lt;/span&gt;&lt;span class="p"&gt;.&lt;/span&gt;&lt;span class="nf"&gt;head&lt;/span&gt;&lt;span class="p"&gt;(&lt;/span&gt;&lt;span class="p"&gt;)&lt;/span&gt;        &lt;span class="c1"&gt;# one header row, typed columns, no totals row&lt;/span&gt;
&lt;span class="k"&gt;print&lt;/span&gt;&lt;span class="p"&gt;(&lt;/span&gt;&lt;span class="n"&gt;report&lt;/span&gt;&lt;span class="p"&gt;)&lt;/span&gt;    &lt;span class="c1"&gt;# every fix, listed — nothing is changed silently&lt;/span&gt;
&lt;/code&gt;&lt;/pre&gt;

&lt;/div&gt;





&lt;div class="highlight js-code-highlight"&gt;
&lt;pre class="highlight console"&gt;&lt;code&gt;tidyexcel report &lt;span class="k"&gt;for&lt;/span&gt; q3-report.xlsx
  rows: 9 &lt;span class="k"&gt;in&lt;/span&gt; -&lt;span class="o"&gt;&amp;gt;&lt;/span&gt; 4 out &lt;span class="o"&gt;(&lt;/span&gt;4 dropped&lt;span class="o"&gt;)&lt;/span&gt;
  header: row&lt;span class="o"&gt;(&lt;/span&gt;s&lt;span class="o"&gt;)&lt;/span&gt; 3
  fixes:
    - title rows skipped: 1
    - blank rows skipped above the header: 1
    - totals row dropped &lt;span class="o"&gt;(&lt;/span&gt;row 8&lt;span class="o"&gt;)&lt;/span&gt;: &lt;span class="s1"&gt;'Total'&lt;/span&gt;
    - footnote row dropped &lt;span class="o"&gt;(&lt;/span&gt;row 9&lt;span class="o"&gt;)&lt;/span&gt;: &lt;span class="s1"&gt;'Note: Revenue excludes tax.'&lt;/span&gt;
&lt;/code&gt;&lt;/pre&gt;

&lt;/div&gt;



&lt;p&gt;What it detects in v0.1.0:&lt;/p&gt;

&lt;ul&gt;
&lt;li&gt;
&lt;strong&gt;Title / preamble rows&lt;/strong&gt; above the header — it finds the real header row instead of being told where it is.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Merged header cells and multi-row headers&lt;/strong&gt; — merged labels filled across their columns, then flattened to single names like &lt;code&gt;Q1 / Sales&lt;/code&gt;.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Totals and subtotal rows&lt;/strong&gt; inside the data — dropped, and named in the report.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Footnote / footer rows&lt;/strong&gt; — &lt;code&gt;Note:&lt;/code&gt;, &lt;code&gt;Source:&lt;/code&gt;, sparse trailing rows below the data.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Merged data cells&lt;/strong&gt; — the value filled across the merged range.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Numbers stored as text&lt;/strong&gt; — converted to numeric, except values with leading zeros (zip codes stay zip codes).&lt;/li&gt;
&lt;/ul&gt;

&lt;p&gt;The part I care about most is the report. Every fix is listed with the original row numbers. If a file was already clean, &lt;code&gt;report.fixes&lt;/code&gt; is empty and the frame comes through untouched — there is a control-file test in the repo that fails the build if a detector ever misfires on a clean file. Deterministic heuristics over the raw cell grid: no LLM, no model calls, no network.&lt;/p&gt;

&lt;p&gt;&lt;strong&gt;Try it before you install anything:&lt;/strong&gt; the &lt;a href="https://yalakala-naveen.github.io/tidysheet/" rel="noopener noreferrer"&gt;live demo&lt;/a&gt; runs the same detection in your browser on sample files or your own — nothing is uploaded anywhere.&lt;br&gt;
&lt;/p&gt;

&lt;div class="highlight js-code-highlight"&gt;
&lt;pre class="highlight console"&gt;&lt;code&gt;pip install tidyexcel
&lt;/code&gt;&lt;/pre&gt;

&lt;/div&gt;



&lt;p&gt;Repo: &lt;a href="https://github.com/Yalakala-Naveen/tidysheet" rel="noopener noreferrer"&gt;https://github.com/Yalakala-Naveen/tidysheet&lt;/a&gt; (MIT, Python ≥ 3.10, pandas + openpyxl).&lt;/p&gt;

&lt;h2&gt;
  
  
  What it deliberately does not do
&lt;/h2&gt;

&lt;ul&gt;
&lt;li&gt;No type inference beyond numbers-stored-as-text. Dates stay as the values openpyxl/pandas give you.&lt;/li&gt;
&lt;li&gt;One sheet per read (the active sheet by default; pass &lt;code&gt;sheet="Name"&lt;/code&gt; or an index).&lt;/li&gt;
&lt;li&gt;
&lt;code&gt;.xlsm&lt;/code&gt; and &lt;code&gt;.csv&lt;/code&gt; are supported; legacy &lt;code&gt;.xls&lt;/code&gt; is not.&lt;/li&gt;
&lt;li&gt;The header search covers the first 50 rows and at most a 3-row header block. A file whose "data" is all text with repeated values can fool the multi-row-header rule — and when it does, the report shows the decision it made, so a wrong call is visible, never silent.&lt;/li&gt;
&lt;/ul&gt;

&lt;p&gt;If you feed it a file shape it gets wrong, open an issue with the file's structure (not the data) — the detectors only get better on real shapes. The merged-cell fill pandas users have been asking for since January is the first thing it does; tell me what your sender's files do that it misses.&lt;/p&gt;

</description>
      <category>python</category>
      <category>pandas</category>
      <category>datascience</category>
      <category>opensource</category>
    </item>
    <item>
      <title>My Agent Said "Done." The Work Wasn't. So I Built a Done-Check.</title>
      <dc:creator>Yalakala-Naveen</dc:creator>
      <pubDate>Mon, 16 Nov 2026 01:56:15 +0000</pubDate>
      <link>https://dev.to/yalakala-naveen/my-agent-said-done-the-work-wasnt-so-i-built-a-done-check-1o9</link>
      <guid>https://dev.to/yalakala-naveen/my-agent-said-done-the-work-wasnt-so-i-built-a-done-check-1o9</guid>
      <description>&lt;p&gt;The launch was marked complete. The page was live. And the body of the page still held placeholder copy — the agent had shipped its own scaffolding and reported success, because nothing in its pipeline ever checked the output against the claim.&lt;/p&gt;

&lt;p&gt;Another run reported a batch as sent. A bounce sat unrecorded. "Sent" had been asserted by the same component that did the sending, and nobody — human or machine — asked it for a message id.&lt;/p&gt;

&lt;p&gt;I'm a Forward Deployed Engineer. I run AI agents in production, and these weren't model failures. The models did fine. The failures were silent: outputs that said &lt;em&gt;done&lt;/em&gt; while the work was placeholder, unbacked, stale, miscounted, or missing pieces — and every dashboard stayed green because the agent graded its own homework.&lt;/p&gt;

&lt;p&gt;So I wrote down the five ways I'd actually seen agent output lie, and built a checker for exactly those:&lt;/p&gt;

&lt;div class="table-wrapper-paragraph"&gt;&lt;table&gt;
&lt;thead&gt;
&lt;tr&gt;
&lt;th&gt;Check&lt;/th&gt;
&lt;th&gt;Name&lt;/th&gt;
&lt;th&gt;What it catches&lt;/th&gt;
&lt;/tr&gt;
&lt;/thead&gt;
&lt;tbody&gt;
&lt;tr&gt;
&lt;td&gt;FL001&lt;/td&gt;
&lt;td&gt;&lt;code&gt;placeholder_shipped&lt;/code&gt;&lt;/td&gt;
&lt;td&gt;"Done" with an empty payload, lorem ipsum, or unrendered &lt;code&gt;{{template}}&lt;/code&gt; braces&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;FL002&lt;/td&gt;
&lt;td&gt;&lt;code&gt;unbacked_claim&lt;/code&gt;&lt;/td&gt;
&lt;td&gt;"Sent / posted / published" with no message id, URL, receipt, or artifact behind it&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;FL003&lt;/td&gt;
&lt;td&gt;&lt;code&gt;stale_state&lt;/code&gt;&lt;/td&gt;
&lt;td&gt;Work based on state older than allowed, untimestamped, or version-skewed&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;FL004&lt;/td&gt;
&lt;td&gt;&lt;code&gt;count_mismatch&lt;/code&gt;&lt;/td&gt;
&lt;td&gt;The report says 12, the itemized record lists 9 — only the list was ever true&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;FL005&lt;/td&gt;
&lt;td&gt;&lt;code&gt;missing_components&lt;/code&gt;&lt;/td&gt;
&lt;td&gt;Fields present but silently empty; dead stretches inside a deliverable&lt;/td&gt;
&lt;/tr&gt;
&lt;/tbody&gt;
&lt;/table&gt;&lt;/div&gt;

&lt;p&gt;FailLoud never asks a model whether an output "looks good." Every verdict is deterministic, reproducible, and explainable line by line. No LLM sits in the verdict path. It runs three ways: a CLI for CI (exit 0 pass, 1 failed output, 2 tool error), a Python API, and an MCP server so an orchestrator can verify a sub-agent's output &lt;em&gt;inside the loop&lt;/em&gt;, one tool call before acting on it.&lt;/p&gt;

&lt;p&gt;The repo ships a realistic agent output that commits all five failures. Here's what it looks like:&lt;br&gt;
&lt;/p&gt;

&lt;div class="highlight js-code-highlight"&gt;
&lt;pre class="highlight shell"&gt;&lt;code&gt;$ pip install failloud
$ failloud check examples/failing_output.json

FAILLOUD — agent output verification
Verdict: FAIL &lt;span class="o"&gt;(&lt;/span&gt;8 error&lt;span class="o"&gt;(&lt;/span&gt;s&lt;span class="o"&gt;)&lt;/span&gt;, 1 warning&lt;span class="o"&gt;(&lt;/span&gt;s&lt;span class="o"&gt;)&lt;/span&gt;&lt;span class="o"&gt;)&lt;/span&gt;

&lt;span class="o"&gt;[&lt;/span&gt;FL001&lt;span class="o"&gt;]&lt;/span&gt; placeholder_shipped — FAIL
  ✗ $.output.body — Placeholder content shipped as final output:
    matched &lt;span class="s1"&gt;'coming soon'&lt;/span&gt;, &lt;span class="s1"&gt;'lorem ipsum'&lt;/span&gt;, &lt;span class="s1"&gt;'{{'&lt;/span&gt;, &lt;span class="s1"&gt;'}}'&lt;/span&gt;.

&lt;span class="o"&gt;[&lt;/span&gt;FL002&lt;span class="o"&gt;]&lt;/span&gt; unbacked_claim — FAIL
  ✗ $.claims&lt;span class="o"&gt;[&lt;/span&gt;1&lt;span class="o"&gt;]&lt;/span&gt; — Claimed action &lt;span class="s1"&gt;'card_published'&lt;/span&gt; has no evidence —
    no id, URL, receipt, or artifact backs that it happened.

&lt;span class="o"&gt;[&lt;/span&gt;FL004&lt;span class="o"&gt;]&lt;/span&gt; count_mismatch — FAIL
  ✗ $.counts.sent — Count &lt;span class="k"&gt;for&lt;/span&gt; &lt;span class="s1"&gt;'sent'&lt;/span&gt; is reported as 12 but the
    itemized record contains 9 item&lt;span class="o"&gt;(&lt;/span&gt;s&lt;span class="o"&gt;)&lt;/span&gt;.
&lt;/code&gt;&lt;/pre&gt;

&lt;/div&gt;



&lt;p&gt;Or in Python:&lt;br&gt;
&lt;/p&gt;

&lt;div class="highlight js-code-highlight"&gt;
&lt;pre class="highlight python"&gt;&lt;code&gt;&lt;span class="k"&gt;import&lt;/span&gt; &lt;span class="n"&gt;json&lt;/span&gt;
&lt;span class="k"&gt;from&lt;/span&gt; &lt;span class="n"&gt;failloud&lt;/span&gt; &lt;span class="k"&gt;import&lt;/span&gt; &lt;span class="n"&gt;check_output&lt;/span&gt;

&lt;span class="n"&gt;result&lt;/span&gt; &lt;span class="o"&gt;=&lt;/span&gt; &lt;span class="nf"&gt;check_output&lt;/span&gt;&lt;span class="p"&gt;(&lt;/span&gt;&lt;span class="n"&gt;json&lt;/span&gt;&lt;span class="p"&gt;.&lt;/span&gt;&lt;span class="nf"&gt;load&lt;/span&gt;&lt;span class="p"&gt;(&lt;/span&gt;&lt;span class="nb"&gt;open&lt;/span&gt;&lt;span class="p"&gt;(&lt;/span&gt;&lt;span class="s2"&gt;"agent_output.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="ow"&gt;not&lt;/span&gt; &lt;span class="n"&gt;result&lt;/span&gt;&lt;span class="p"&gt;.&lt;/span&gt;&lt;span class="n"&gt;passed&lt;/span&gt;&lt;span class="p"&gt;:&lt;/span&gt;
    &lt;span class="k"&gt;for&lt;/span&gt; &lt;span class="n"&gt;finding&lt;/span&gt; &lt;span class="k"&gt;in&lt;/span&gt; &lt;span class="n"&gt;result&lt;/span&gt;&lt;span class="p"&gt;.&lt;/span&gt;&lt;span class="n"&gt;errors&lt;/span&gt;&lt;span class="p"&gt;:&lt;/span&gt;
        &lt;span class="k"&gt;print&lt;/span&gt;&lt;span class="p"&gt;(&lt;/span&gt;&lt;span class="n"&gt;finding&lt;/span&gt;&lt;span class="p"&gt;.&lt;/span&gt;&lt;span class="n"&gt;check_id&lt;/span&gt;&lt;span class="p"&gt;,&lt;/span&gt; &lt;span class="n"&gt;finding&lt;/span&gt;&lt;span class="p"&gt;.&lt;/span&gt;&lt;span class="n"&gt;path&lt;/span&gt;&lt;span class="p"&gt;,&lt;/span&gt; &lt;span class="n"&gt;finding&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;/code&gt;&lt;/pre&gt;

&lt;/div&gt;



&lt;p&gt;&lt;strong&gt;The critique that made it better&lt;/strong&gt;&lt;/p&gt;

&lt;p&gt;When I posted the launch, a reader pushed on exactly the right weak point: a claim like "9 sent" can pass every check at run time — and one of those sends can still bounce later. A green verdict at minute one says nothing about minute sixty.&lt;/p&gt;

&lt;p&gt;He was right, and it changed the design. In 0.5.0, a PASS is now labelled &lt;strong&gt;provisional&lt;/strong&gt;. A new command, &lt;code&gt;failloud recheck&lt;/code&gt;, takes the original claims and re-judges them against a later snapshot of the system of record. Every claim comes back &lt;strong&gt;confirmed&lt;/strong&gt;, &lt;strong&gt;contradicted&lt;/strong&gt;, or &lt;strong&gt;unverifiable&lt;/strong&gt;. Verification isn't a moment anymore; it's a second pass, against the record, after reality has had time to answer back. (Current release: 0.5.1 — same checks, now also published in the official MCP Registry.)&lt;/p&gt;

&lt;p&gt;&lt;strong&gt;Try it before you install anything&lt;/strong&gt;&lt;/p&gt;

&lt;p&gt;The Playground runs the checks in your browser — paste an agent output, get the failure report, nothing uploaded:&lt;br&gt;
&lt;a href="https://yalakala-naveen.github.io/failloud/" rel="noopener noreferrer"&gt;https://yalakala-naveen.github.io/failloud/&lt;/a&gt;&lt;/p&gt;

&lt;p&gt;If it's useful: &lt;code&gt;pip install failloud&lt;/code&gt; (add &lt;code&gt;[mcp]&lt;/code&gt; for the server). Code, examples, and the full check reference: &lt;a href="https://github.com/Yalakala-Naveen/failloud" rel="noopener noreferrer"&gt;https://github.com/Yalakala-Naveen/failloud&lt;/a&gt; — MIT licensed, zero-dependency core, 147 tests.&lt;/p&gt;

&lt;p&gt;Honest closing: this just shipped and has no users yet. If you run agents in production, I want the failure mode my five checks miss. That's the most useful thing you can hand me — it's already happened once, and it shipped as 0.5.0.&lt;/p&gt;

</description>
      <category>ai</category>
      <category>agents</category>
      <category>mcp</category>
      <category>python</category>
    </item>
  </channel>
</rss>
