<?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: Irina Vetlugina</title>
    <description>The latest articles on DEV Community by Irina Vetlugina (@irinavetlugina).</description>
    <link>https://dev.to/irinavetlugina</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%2F4255992%2F3ac93252-a1ea-4488-99d1-93c926df868e.png</url>
      <title>DEV Community: Irina Vetlugina</title>
      <link>https://dev.to/irinavetlugina</link>
    </image>
    <atom:link rel="self" type="application/rss+xml" href="https://dev.to/feed/irinavetlugina"/>
    <language>en</language>
    <item>
      <title>Stop LLMs from Hallucinating: A Portable "Case Filter" Protocol to Separate Facts from Hypotheses</title>
      <dc:creator>Irina Vetlugina</dc:creator>
      <pubDate>Fri, 20 Nov 2026 17:03:35 +0000</pubDate>
      <link>https://dev.to/irinavetlugina/stop-llms-from-hallucinating-a-portable-case-filter-protocol-to-separate-facts-from-hypotheses-24e9</link>
      <guid>https://dev.to/irinavetlugina/stop-llms-from-hallucinating-a-portable-case-filter-protocol-to-separate-facts-from-hypotheses-24e9</guid>
      <description>&lt;p&gt;Anyone who uses Large Language Models (ChatGPT, Claude, Gemini, DeepSeek) for research, analytics, or hypothesis testing knows their biggest flaw: &lt;strong&gt;LLMs describe a proven fact, a bold hypothesis, and a pure fabrication with the exact same smooth confidence.&lt;/strong&gt;&lt;/p&gt;

&lt;p&gt;When you ask an AI &lt;em&gt;"Is this true?"&lt;/em&gt; or ask it to evaluate a controversial claim, it almost always generates a plausible-sounding essay with pros and cons. Inside a standard chat, the model lacks a hard structural brake that forces it to distinguish between:&lt;/p&gt;

&lt;ul&gt;
&lt;li&gt;an event that has actually occurred and been "paid" for with real work or computation;&lt;/li&gt;
&lt;li&gt;a forecast with a specific verification deadline;&lt;/li&gt;
&lt;li&gt;an ungrounded claim with no mechanism or executor.&lt;/li&gt;
&lt;/ul&gt;

&lt;p&gt;To solve this without building complex RAG pipelines or fine-tuning models, I developed &lt;strong&gt;Case Filter v1.2&lt;/strong&gt; — a portable, text-based verification protocol (based on the &lt;strong&gt;Dated Record Method&lt;/strong&gt;) that you paste at the beginning of any chat to turn an LLM from a "plausible text generator" into a strict record auditor.&lt;/p&gt;

&lt;p&gt;Full protocol (English &amp;amp; Russian) and benchmark logs on GitHub: &lt;a href="https://github.com/IrinaVetlugina/case-filter" rel="noopener noreferrer"&gt;https://github.com/IrinaVetlugina/case-filter&lt;/a&gt;&lt;/p&gt;




&lt;h2&gt;
  
  
  Core Idea: Grammatical Cases as Coordinates of a Real Transaction
&lt;/h2&gt;

&lt;p&gt;The Dated Record Method treats grammatical cases not as school syntax rules, but as &lt;strong&gt;seven mandatory audit questions&lt;/strong&gt; for every statement. If key coordinates are missing from a claim, we are not looking at a fact — we are looking at a logical substitution.&lt;/p&gt;

&lt;ol&gt;
&lt;li&gt;
&lt;strong&gt;Nominative ("Who is the executor?"):&lt;/strong&gt; The debtor. Spends ticks (time, compute, resources) and acts rather than predicts.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Accusative-Ê ("What is being created?"):&lt;/strong&gt; The candidate or future target. Always recorded on a separate line from the actual outcome.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Accusative-E ("What was actually created?"):&lt;/strong&gt; The realized result of a completed act.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Dative ("To whom is the record?"):&lt;/strong&gt; The creditor, archive, or ownership after the transaction closes (the past).&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Genitive ("Where is the contribution from?"):&lt;/strong&gt; The baseline record or resource funding a future act; the source of a forecast.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Instrumental ("Paid with what? Thanks to what?"):&lt;/strong&gt; The currency of the transaction, ticks spent, or physical/computational mechanism.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Prepositional ("Where is the transaction?"):&lt;/strong&gt; The institutional position or context &lt;code&gt;σ(O)&lt;/code&gt;. Without a position, there is neither an event nor a legal record.&lt;/li&gt;
&lt;/ol&gt;

&lt;h3&gt;
  
  
  Two Core State Transitions
&lt;/h3&gt;



&lt;div class="highlight js-code-highlight"&gt;
&lt;pre class="highlight plaintext"&gt;&lt;code&gt;Completed Act:        Ω(T) --[act σ(O)]--&amp;gt; E ∈ σ(O)
Forecast (Pre-Act):   Ω(T) --[act σ(O)]--&amp;gt; Ê ∈ σ(O) --[verification]--&amp;gt; E
&lt;/code&gt;&lt;/pre&gt;

&lt;/div&gt;



&lt;p&gt;&lt;strong&gt;Strict Protocol Rules for the AI:&lt;/strong&gt;&lt;/p&gt;

&lt;ul&gt;
&lt;li&gt;
&lt;strong&gt;Banned certainty words:&lt;/strong&gt; The LLM is strictly prohibited from using words like &lt;em&gt;"confirmed"&lt;/em&gt;, &lt;em&gt;"proven"&lt;/em&gt;, or &lt;em&gt;"truth"&lt;/em&gt;. It may only assign strict protocol statuses: &lt;code&gt;legal record&lt;/code&gt;, &lt;code&gt;[R] candidate&lt;/code&gt;, &lt;code&gt;left edge&lt;/code&gt;, &lt;code&gt;cell 1&lt;/code&gt; (record without act), or &lt;code&gt;false removed&lt;/code&gt;.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Dated verification windows:&lt;/strong&gt; A candidate &lt;code&gt;Ê&lt;/code&gt; is valid &lt;strong&gt;only prior to the act&lt;/strong&gt; and &lt;strong&gt;only with a dated verification window&lt;/strong&gt;.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Legitimizing "I don't know":&lt;/strong&gt; When external verification criteria are missing, the AI is forbidden from guessing and must assign &lt;strong&gt;&lt;code&gt;unidentified remainder&lt;/code&gt;&lt;/strong&gt;.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Role separation:&lt;/strong&gt; The AI acts strictly as the &lt;code&gt;carrier: AI&lt;/code&gt; (and must label every card as such), while the human user remains the &lt;strong&gt;"key holder"&lt;/strong&gt; responsible for time-based verification.&lt;/li&gt;
&lt;/ul&gt;




&lt;h2&gt;
  
  
  Benchmark Experiment: Running 10 Major Science Problems
&lt;/h2&gt;

&lt;p&gt;To test the filter's discipline on complex, ambiguous claims, I ran 10 fundamental problems across different fields of science through the protocol — &lt;strong&gt;2 solved and 8 unsolved&lt;/strong&gt; (full log available in &lt;code&gt;run-science-problems-en.md&lt;/code&gt; in the repository).&lt;/p&gt;

&lt;p&gt;Crucial rule: the tool does not attempt to "solve" equations for scientists — it audits the &lt;strong&gt;status of records&lt;/strong&gt; about them.&lt;/p&gt;

&lt;h3&gt;
  
  
  1. Solved Problems → &lt;code&gt;legal record&lt;/code&gt;
&lt;/h3&gt;

&lt;p&gt;When auditing the &lt;strong&gt;Poincaré Conjecture&lt;/strong&gt; and &lt;strong&gt;Protein Folding (AlphaFold2)&lt;/strong&gt;, the model moved both records into the Dative case as established facts, because both presented paid computational/mathematical ticks and closed external verification windows:&lt;br&gt;
&lt;/p&gt;

&lt;div class="highlight js-code-highlight"&gt;
&lt;pre class="highlight plaintext"&gt;&lt;code&gt;RECORD: Poincaré Conjecture solved (every simply connected, closed 3-manifold is homeomorphic to the 3-sphere).
FRAMEWORK Ω(T): 3-manifold topology; parameter T absent (timeless mathematical framework)
ACT: completed; paid with mathematical ticks of Ricci flow with surgery (2002–2003) and independent verification (2006); crossed the creator's boundary
POSITION σ(O): corpus of mathematical knowledge (preprint archives, peer-reviewed publications)
CLASS: legal record
VERIFICATION WINDOW: closed (verified in 2003–2006, Ê = E)
FORECAST Ê: removed (returned to Dative case as fact)
ERRORS: none detected
SUBJECT IN PROTOCOL: omitted (Rule 4 of Chair Laws)
carrier: AI
&lt;/code&gt;&lt;/pre&gt;

&lt;/div&gt;



&lt;h3&gt;
  
  
  2. Unsolved Problems with a Theoretical Contribution → &lt;code&gt;left edge&lt;/code&gt;
&lt;/h3&gt;

&lt;p&gt;For the &lt;strong&gt;Riemann Hypothesis&lt;/strong&gt;, &lt;strong&gt;Navier–Stokes Existence and Smoothness&lt;/strong&gt;, &lt;strong&gt;P vs NP&lt;/strong&gt;, &lt;strong&gt;Abiogenesis&lt;/strong&gt;, and &lt;strong&gt;Linear A Decipherment&lt;/strong&gt;, the protocol assigned &lt;code&gt;left edge&lt;/code&gt; (or &lt;code&gt;[R] candidate&lt;/code&gt; for &lt;strong&gt;Quantum Gravity&lt;/strong&gt;). A foundational contribution exists, but the final act has not been paid with ticks and no dated verification window exists:&lt;br&gt;
&lt;/p&gt;

&lt;div class="highlight js-code-highlight"&gt;
&lt;pre class="highlight plaintext"&gt;&lt;code&gt;RECORD: Riemann Hypothesis (all non-trivial zeros of the Riemann zeta function have real part 1/2).
FRAMEWORK Ω(T): analytic number theory; parameter T absent (timeless frame)
ACT: not completed (no general proof; partial numerical checks of billions of zeros do not close the general framework)
POSITION σ(O): corpus of mathematics (Clay Mathematics Institute, Millennium Prize Problems)
CLASS: left edge
VERIFICATION WINDOW: absent (no dated window for the general case)
FORECAST Ê: issued without window (contribution since 1859 without final act)
ERRORS: none detected
SUBJECT IN PROTOCOL: omitted (Rule 4 of Chair Laws)
carrier: AI
&lt;/code&gt;&lt;/pre&gt;

&lt;/div&gt;



&lt;h3&gt;
  
  
  3. Problems Lacking External Criteria → &lt;code&gt;unidentified remainder&lt;/code&gt;
&lt;/h3&gt;

&lt;p&gt;The most revealing result came on the &lt;strong&gt;Nature of Dark Matter&lt;/strong&gt; and the &lt;strong&gt;Hard Problem of Consciousness&lt;/strong&gt; (how physical brain processes generate subjective experience / qualia). Instead of summarizing popular theories as an answer, the AI legally refused to classify them:&lt;br&gt;
&lt;/p&gt;

&lt;div class="highlight js-code-highlight"&gt;
&lt;pre class="highlight plaintext"&gt;&lt;code&gt;RECORD: Hard Problem of Consciousness (how and why physical processes in the brain give rise to subjective experience / qualia).
FRAMEWORK Ω(T): philosophy of mind and neurobiology; parameter T — timescale of neural correlates
ACT: not completed (objective neural ticks are measured, but the transition from physical process to subjective experience is neither described nor measured)
POSITION σ(O): corpus of neuroscience and philosophy of mind (conflicting positions)
CLASS: unidentified remainder
VERIFICATION WINDOW: absent (no operational criterion for external verification of subjective experience)
FORECAST Ê: absent (cannot issue a legal forecast without verification criterion E)
ERRORS: none detected
SUBJECT IN PROTOCOL: omitted (Rule 4 of Chair Laws)
carrier: AI
&lt;/code&gt;&lt;/pre&gt;

&lt;/div&gt;






&lt;h2&gt;
  
  
  Quick Start (30 Seconds)
&lt;/h2&gt;

&lt;p&gt;No libraries, plugins, or API keys required:&lt;/p&gt;

&lt;ol&gt;
&lt;li&gt;&lt;p&gt;Open the repository: &lt;a href="https://github.com/IrinaVetlugina/case-filter" rel="noopener noreferrer"&gt;https://github.com/IrinaVetlugina/case-filter&lt;/a&gt;&lt;/p&gt;&lt;/li&gt;
&lt;li&gt;&lt;p&gt;Copy the entire contents of &lt;code&gt;case-filter-v1.2-en.md&lt;/code&gt; (or &lt;code&gt;case-filter-v1.2.md&lt;/code&gt; for Russian).&lt;/p&gt;&lt;/li&gt;
&lt;li&gt;&lt;p&gt;Paste it at the very beginning of a new chat with any modern LLM (ChatGPT, Claude, Gemini, DeepSeek, etc.).&lt;/p&gt;&lt;/li&gt;
&lt;li&gt;&lt;p&gt;Submit any claim, news item, or working hypothesis using:&lt;br&gt;
&lt;/p&gt;&lt;/li&gt;
&lt;/ol&gt;

&lt;div class="highlight js-code-highlight"&gt;
&lt;pre class="highlight plaintext"&gt;&lt;code&gt;   RUN: [your claim here]
&lt;/code&gt;&lt;/pre&gt;

&lt;/div&gt;



&lt;h3&gt;
  
  
  Useful Session Commands
&lt;/h3&gt;

&lt;ul&gt;
&lt;li&gt;
&lt;code&gt;RUN: [text]&lt;/code&gt; — audit a record and output a strict protocol card + brief explanation.&lt;/li&gt;
&lt;li&gt;
&lt;code&gt;FORECAST: [Ê] window [date]&lt;/code&gt; — register a hypothesis prior to its verification deadline.&lt;/li&gt;
&lt;li&gt;
&lt;code&gt;VERIFY: [Ê against fact E]&lt;/code&gt; — verify a prediction once the window arrives.&lt;/li&gt;
&lt;li&gt;
&lt;code&gt;SUMMARY&lt;/code&gt; — self-audit the AI carrier for classification/reading errors (&lt;code&gt;O1–O3&lt;/code&gt;) during the session.&lt;/li&gt;
&lt;/ul&gt;




&lt;h2&gt;
  
  
  Limitations &amp;amp; Honest Accounting
&lt;/h2&gt;

&lt;p&gt;To get real value from the protocol, keep three boundaries in mind:&lt;/p&gt;

&lt;ul&gt;
&lt;li&gt;
&lt;strong&gt;It audits records, not primary reality.&lt;/strong&gt; The filter exposes hidden passive voice (&lt;em&gt;"it happened on its own"&lt;/em&gt;), missing mechanisms, and unverified forecasts, but it does not run lab experiments or fact-check raw numbers for you.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Input sensitivity.&lt;/strong&gt; At boundary zones (e.g., between &lt;code&gt;[R] candidate&lt;/code&gt; and &lt;code&gt;left edge&lt;/code&gt;), missing explicit timelines in the prompt can cause classification slips (&lt;code&gt;O1 error&lt;/code&gt;), which is why the human always remains the "key holder."&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Context drift in long chats.&lt;/strong&gt; Like any system prompt, over dozens of messages an LLM may loosen its card formatting — simply run &lt;code&gt;SUMMARY&lt;/code&gt; or start a fresh chat with the carrier file.&lt;/li&gt;
&lt;/ul&gt;




&lt;h2&gt;
  
  
  License &amp;amp; Repository
&lt;/h2&gt;

&lt;ul&gt;
&lt;li&gt;&lt;p&gt;**GitHub Repository: &lt;a href="https://github.com/IrinaVetlugina/case-filter" rel="noopener noreferrer"&gt;https://github.com/IrinaVetlugina/case-filter&lt;/a&gt;&lt;/p&gt;&lt;/li&gt;
&lt;li&gt;&lt;p&gt;&lt;strong&gt;License:&lt;/strong&gt; Deposited with n'RIS (Reg. No. &lt;code&gt;852-137-831&lt;/code&gt; dated &lt;code&gt;20.11.2026&lt;/code&gt;). Free for individuals for personal, educational, and non-commercial research use; commercial license agreement required for organizations and businesses.&lt;/p&gt;&lt;/li&gt;
&lt;/ul&gt;

&lt;p&gt;I'd love to hear your feedback — try running a controversial claim, startup pitch, or research hypothesis through the filter and share what status your LLM assigns!&lt;/p&gt;

</description>
      <category>ai</category>
      <category>llm</category>
      <category>opensource</category>
      <category>showdev</category>
    </item>
    <item>
      <title>Stop LLMs from Hallucinating: A Portable "Case Filter" Protocol to Separate Facts from Hypotheses</title>
      <dc:creator>Irina Vetlugina</dc:creator>
      <pubDate>Fri, 20 Nov 2026 16:28:03 +0000</pubDate>
      <link>https://dev.to/irinavetlugina/stop-llms-from-hallucinating-a-portable-case-filter-protocol-to-separate-facts-from-hypotheses-379a</link>
      <guid>https://dev.to/irinavetlugina/stop-llms-from-hallucinating-a-portable-case-filter-protocol-to-separate-facts-from-hypotheses-379a</guid>
      <description></description>
    </item>
  </channel>
</rss>
