About Trigpoint

1. What Trigpoint does

Trigpoint is an independently published knowledge base. The work is to turn verified research from across disciplines into checkable knowledge nodes. Those we have field-tested in our own workflow carry the Lab-tested mark.

Take human factors: since 1948, the field has established a great deal about how people and automation work together. As AI enters everyone's workflow, that body of findings is being forgotten wholesale and the same mistakes are being repeated wholesale. Trigpoint's position: read those who came before, stand on the shoulders of giants, then talk about innovation.

Source: Bernard of Chartres, as reported in John of Salisbury, Metalogicon III.4 (1159)

Mission

Trigpoint is the public face of a cross-disciplinary theory library. Every entry comes from recognized literature, passes the same verification process, and is compressed into a form you can use directly. We take reliable knowledge scattered across the literatures and lay it out as paths you can walk. New shelves open in batches on the monthly update day, and the theory layer opens to all readers that day. Subscribers can read each new shelf in full from seven days before it opens, and unlock the verification details, failure boundaries and case readings of complete nodes. What we deliver is a collection you can check, and the judgment to put it to use.

Vision

Trigpoint's vision is a checkable infrastructure for judgment in the age of AI. The reading layer of every topic becomes open to all readers; reprints must credit and link back. Every full node goes through the same verification: a figure that has not been checked against a primary source does not appear in a chart. Errors are corrected in public, with thanks to the readers who report them.

2. Why it can be trusted

Most of the time, opening an AI chat yourself is enough. It is free, it is immediate, and on most questions it answers well enough. Trigpoint does not exist for those moments. It exists for the other kind, where the model is wrong fluently and confidently and you have no way to notice.

What a language model knows converges on the mode of its training corpus. Wherever consensus departs from fact, in textbook drift, in misattributed quotations, in statistics passed along uncorrected, the model will be wrong persuasively. A knowledge base verified line by line stores exactly the residual between corpus consensus and established fact. That residual is hard for a model to regenerate from its own weights, because it takes checking rather than recall.

The residual may also be widening. Fabricated content keeps entering the open web. A “25% success rate” statistic that appears nowhere in the original report is spreading between third-party sites, and that is a case we verified ourselves. Unless someone checks, future training corpora may sit further from the facts, not closer. Our working assumption is that the more fluent the models get, the more a checked residual is worth. The assumption can be refuted: if a model learns to tell corpus consensus from established fact on its own, the value of this library should be re-assessed.

We know you cannot verify our quality. A correct citation and an incorrect one read identically, unless you already know the answer. So we do not promise that every answer is right. What we commit to is the process: every load-bearing statement goes back to a primary source, verification status is marked on every page, and errors are corrected in public. You do not need to trust our conclusions. You can inspect our procedure.

Verification costs something, so the effort is graded by the cost of being wrong. High-risk claims are cross-checked across models. Ordinary citations are verified against the primary source. Definitional material is marked honestly as unverified. A citation being present does not by itself make a claim sound. The risk of a citation being misused is something we manage too, and it is one of the theories in our own library.

Give every claim a reference point to stand on.

3. What is here

4. Governance and support

Production is bound by five ground rules, including a public corrections regime: once an error is reported and confirmed, the correction record stays on the page, with thanks to the reader who reported it.

The theory layer is open for reprint; credit it and link back to the original page. The case interpretation library is Trigpoint Pass subscription content, and subscriptions support the independent operation of Trigpoint. Objections to any of the content, and experience with it, are welcome through the form at the bottom of the relevant node page, with no registration required. Once the editor has handled them, they are listed in the discussion and response record. For corrections, withdrawing a submission or privacy questions, email hello@trigpoint.pub.

5. Human and AI division of labour

Production at Trigpoint is human and AI working together. The division follows the four-stage allocation principle in Levels of automation. The full settings and the reasons for each are published in the allocation table lab case.

AI carries information gathering and first-pass analysis: source scanning, assembling literature leads, drafting, and pre-filling data. Topic selection, interpretation, the final confirmation of every source trace, and the decision to publish each page rest with the editor.

The monitoring limits, attention mechanisms and allocation discipline you read about on the shelves are used, broken and repaired inside our own production process first. That is what separates Trigpoint from content that only restates.

6. The editor

Ansel

Editor · topic selection · field testing · source tracing · publication gate

The day job is turning AI from one-off prompts into workflows that can be governed and reused. What Trigpoint records is what that looks like in practice: which theories hold up, where they fail in application, and how the failures get repaired.

This page lists no titles and no institutions. Under the ground rules, credibility rests on verification, not on credentials.