Compiled with MiniMax M3 (latest). Source-linked; no auto-publish. Model version history.
This page is the framework — the design patterns, source-criticism rules, and anti-fallacy methodology — that every Factual Foundations topic is built on. It does not cover any specific topic. For example topics, see the Volhynia report and the Portal proposal.
The Factual Foundations reference library has three kinds of pages:
If you are trying to understand a specific issue, read a topic page. If you want to know why the reference library works this way, read a proposal page. This page is for readers who want to evaluate the methodology itself before reading any specific topic.
Five design patterns. Every Factual Foundations topic page applies all five.
"Killed 100,000 people" conceals the severity. Include specific methods, not just aggregate figures. A reader who only sees the count has no way to evaluate the moral weight of the event.
Admit every source has editorial constraints. Show readers how to triangulate across sources rather than picking one. Cross-edition comparison (PL/EN/UK Wikipedia on the same topic) is the strongest example.
For every outlet referenced, disclose known biases with citations to the bias studies themselves. Manhattan Institute 2024, AER 2012, Wikipedia's own systemic bias page are all citable sources.
A page that avoids uncomfortable specifics is sanitized, not neutral. Present the facts that explain the reaction, even when they are graphic. Use content warnings, not omission.
"Poland calls it genocide, Ukraine calls it tragedy" can imply equal validity. Better: present evidence for each position, let readers weigh it. Don't artificially balance a well-documented conclusion against a political position.
Every outlet referenced in a Factual Foundations topic has its known editorial bias documented. The framework has three layers.
When a topic is contested, cite real bias research rather than guesswork. Examples:
For any topic with multiple Wikipedia editions (most history topics, current events, people), read all three major editions. The differences in coverage are themselves evidence of editorial intent. On a topic like the Volhynia massacres:
Always start with the most detailed language edition. Treat the least-detailed as the official-narrative sample and weight accordingly.
Every outlet referenced on a Factual Foundations topic has its known limitations documented. The standard table:
| Source type | Best for | Missing |
|---|---|---|
| Wire services (AP / Reuters) | Dependable "what happened" with named attribution | Almost no historical or causal context |
| Public broadcasters (BBC) | Balanced framing, open methodology | May omit uncomfortable specifics for tonal reasons |
| Independent journalism (Notes from Poland) | Detailed chronology, embedded primary sources | Country-centric framing; less balanced toward opponents |
| Government institutes (IPN / UINM) | Forensic primary-source documentation | Government-aligned framing; treat as the institutional voice, not neutral |
| Wikipedia (any language) | Sourced synthesis across multiple language editions | Systemic left-bias (EN), language-nationalist framing (PL/UK/UK/RU/UA) |
The 19 fallacies below are the canonical abstract patterns of the dismissal moves FF counters. The example quotes in the table are taken from FF reports - currently the Volhynia page is the canonical worked example, but each fallacy can be applied to any topic with a similar pattern.
How to use this handbook on a new topic:
The volhynia page is the canonical worked example. It has 18 fallacy cards (1-18; #19 is methodology-only as the meta-fallacy) using the Volhynia-specific examples. Those cards stay on the volhynia page. Other topic pages will have their own fallacy cards with their own examples, sharing the same 19 underlying patterns.
The fallacy-handbook table below is the single source of truth for the 19 patterns. It should only be edited when adding/removing a fallacy, not when adding a new topic. The full table below lists all 19 with their canonical examples; the topic pages rephrase the examples for their topic.
Any factual report on a contested political topic will be dismissed. The dismissals are not based on engaging with the evidence — they are rhetorical moves designed to skip the engagement. Every Factual Foundations topic page includes a fallacy section anticipating the most common moves. The standard handbook:
| Fallacy | Pattern | Standard counter | Share |
|---|---|---|---|
| Genetic | ""It was generated by [AI / hostile actor], so it can't be trusted"" | Attack a specific claim or accept the cited source. The producer of the prose does not change the underlying fact. | 𝕏 |
| Poisoning the well | ""That's just [foreign propaganda / opposition talking points]"" | Who benefits from a claim is independent of whether it is true. State the documented evidence. | 𝕏 |
| Ad hominem by association | ""That's what [party X] says, so it's partisan"" | Test the claim against the cited primary sources, not which group repeats it. Truth is not determined by who repeats it. | 𝕏 |
| Whataboutism | ""What about [unrelated action by the other side]?"" | Widens the lens to other events; rebuttal narrows it back to the cited evidence. Even if the other action is real, it does not negate the documented actions here. | 𝕏 |
| Sanitisation by reduction | ""What about [actor's positive trait]? They were a [heroic role]"" | Honoring a complex figure requires honoring all of their record. A single positive trait is not a counter to documented criticism. | 𝕏 |
| Loaded question | ""Anyone who criticises [actor] must be [discredited group], right?"" | A claim's validity is not determined by the alleged politics of the speaker. Independent sources agree; the pattern is evidence, not conspiracy. | 𝕏 |
| Ad hominem circumstantial | ""Why now? / bad timing / [motivated by political calendar]"" | The timing of when a fact is cited is not the fact itself. The historical record is what it is regardless of who cites it and when. | 𝕏 |
| Tu quoque | ""The critic has [tainted past / hypocrisy], so they have no standing"" | The standing of the speaker is not the truth of the claim. Even granting the speaker has a tainted past, the documented facts do not change. | 𝕏 |
| False equivalence | ""It was [context: wartime, both sides, etc.]"" | Equivalence needs evidence of parity. Spontaneous chaos is not pre-planned, ordered, named operations with documented methods. | 𝕏 |
| Ad populum inverse | ""Polling shows [X]% disagree, so the claim is just opinion"" | A poll measures sentiment, not historical facts. Sentiment can be wrong about facts. The cited evidence stands independent of any poll. | 𝕏 |
| Context inflation | ""You have to see it from [actor's] perspective! / Context!"" | Context adds understanding; it does not retroactively justify. The events are not erased by appeals to context, especially when used to soften documented actions. | 𝕏 |
| Appeal to consequences | ""Saying this helps [bad actor] / undermines [good cause]"" | Whether citing evidence is politically inconvenient does not change whether it is true. Naming a consequence is not a refutation of the underlying fact. | 𝕏 |
| Appeal to model bias | ""[AI model X] is biased - don't trust what an AI says"" | Genetic fallacy at one remove. The AI confirmed what cited primary sources already say. Engage the cited sources, not the model that confirmed them. | 𝕏 |
| Motte and Bailey | "First [extreme claim]; when challenged, retreat to "we only meant [modest claim]"" | State the original claim plainly. The arguer cannot stake out an extreme claim, then retreat to a defensible one after critique. | 𝕏 |
| Straw Man | ""So you're saying [exaggerated / simplified version of the claim]"" | Restate the original claim exactly. The straw man is a simpler or more extreme version. Attacking the simplified version does not rebut the original. | 𝕏 |
| Tone Policing | ""Even if true, the way [you/they] say it is too aggressive"" | Tone is not evidence. The truth of a documented action does not depend on the politeness of the document. Engage with the cited content. | 𝕏 |
| Bulverism | ""Of course [you/they] say [claim]; you are [position], so biased"" | Speculating about motive is not rebuttal. Motive-inquiry (with evidence) is legitimate; motive-stacking (as substitute) is dismissal. | 𝕏 |
| No True Scotsman | ""Real [members of group] wouldn't do [action]; just fringe elements"" | Reject the purity gate. If the action was committed by people claiming the identity, the identity-membership claim does not erase the action. | 𝕏 |
| Argument from Fallacy (the fallacy fallacy) | ""Your argument uses a fallacy, so the conclusion is false"" | A fallacious argument does not make the conclusion false. Evaluate on its own merits and by independent evidence. | 𝕏 |
Every Factual Foundations report is compiled by an LLM. The specific model varies — see the version history at the end of this section. The predictable dismissal is: "It was written by AI, therefore it's unreliable / hallucinated / propaganda." This is a genetic fallacy (judging a claim by its origin rather than its content) and it is easily dismantled with facts.
A more recent variant — common on X — is: "Grok is biased." Used as a counter when someone else has used Grok (or another AI) to fact-check a Factual Foundations claim, and the AI confirms the claim. Dismissing the AI's confirmation because "the AI is biased" is genetic fallacy at one remove. The model returned the same conclusion the report did — that is not bias, that is independent confirmation of the external, human-produced sources cited. The right next move is to engage those sources, not to dismiss the tool. See fallacy #13 in the handbook above for the full counter.
The LLM is a research assistant and compositor. A human researcher with the same tools, same articles, and same methodology would produce the same factual content — the LLM merely automated retrieval, reading, and formatting. Every report stands or falls on its citations — not on whether a human or an LLM typed the words.
The specific LLM used to compile each Factual Foundations report is documented here. Reports cite the model in their meta-bar at the top of the page; the version history records changes over time.
| Model | Period used | Notes |
|---|---|---|
| DeepSeek V4 Flash | Jun–Jul 2026 (initial pages) | First model used to compile the FF portal and three reference pages. Model name was shown explicitly in early LLM-disclosure sections. |
| MiniMax M3 | From Jul 2026 (current) | Switched to this model in July 2026. All subsequent updates, corrections, and re-deploys use this model. The fact-checking methodology is identical: every claim is cited, the LLM only retrieves and formats. |
Use this decision tree:
Read a topic page. The Volhynia report is the canonical example — full factual record on a specific contested issue.
Read the Portal proposal. It argues for self-updating, source-linked, bias-disclosed, fallacy-audited reference pages as an alternative to mainline news.
You're reading it. This page exists so readers can assess the framework before applying it to any specific topic. It documents the patterns, the source-criticism rules, and the fallacy handbook.
Studies and sources cited in this methodology page: