AI content policy

Small Optic uses language-model pipelines to draft and evaluate the ideas you read here. This page is a plain-language description of what is automated, what isn’t, and what we do when we get something wrong.

What is automated

• The candidate draft. A language-model producer drafts the initial structured candidate — title, market, pain evidence, leverage angle — from a short seed description.
• The bull case. A second model writes the advocate’s framing of why the idea could work.
• The bear case. A third model writes the skeptic’s framing of why the idea could fail.
• The verdict. A fourth model reads both cases and assigns a structured verdict on a rubric.
• The category assignments and topical taxonomy. The industry vertical, leverage angle, and market-size bucket are assigned by a rule-based classifier on top of the structured verdict.

What is not automated

• The editorial review before publish. Every Dig that reaches a publish-eligible verdict is reviewed by the appropriate desk on Small Optic Desk before it goes live.
• The decision to publish or not publish. The automated process produces a verdict; the desk decides whether the verdict and supporting case survive editorial standards.
• The methodology. The methodology itself is human-authored and human-maintained.
• Corrections. When we get something wrong, a desk lead reads the original page, the source claim, and the correction submission, and decides how to handle it.

Author identities

Pages on Small Optic carry a byline from one of three named editors — M. Vance (data desk), R. Tate (consumer + indie desk), or J. Calder (cases desk). These names are persistent editorial identities for the three editorial voices. They are not the legal names of individual biographical persons. We use them because consistent editorial identity matters more for readers and for structured-data signals than the alternative of either (a) inventing biographical detail we’d have to fabricate or (b) attributing pieces to “the desk” with no consistent voice signal at all.

Every published page on Small Optic represents a desk’s review. No piece is published without a human review pass.

Sources and claims

Every numerical or external-factual claim in a Dig is either (a) sourced to an external URL we link inline, or (b) marked [unverified] so readers know the desk wasn’t able to independently confirm the figure. Claims with no source and no unverified-marker get removed in editorial review.

External links resolve to real, publicly-accessible URLs at publish time. If a source link goes dead later, we don’t silently update — we add a note that the link was once live and is no longer reachable.

Corrections and retractions

If you find an error on Small Optic — a wrong number, a misattributed quote, a factual claim that doesn’t hold up — please tell us. We treat corrections as first-priority desk work. Outcomes:

• Correction. We fix the page, mark it as corrected at the bottom, and log what changed and when.
• Retraction. If the underlying premise of a Dig doesn’t hold, we retract the Dig. The URL stays live but the content is replaced with a retraction note explaining what was wrong and why. We don’t silently delete pages.

We don’t edit live pages silently to make them look like they were always right.

Things we don’t do

• We don’t fabricate quotes from real people. If a Dig references a quote, it’s linked to the public source where the quote appeared.
• We don’t invent statistics. Numerical claims are sourced or marked unverified.
• We don’t pose as biographical individuals. The desk identities above are editorial pseudonyms.
• We don’t take payment to publish, suppress, or favor a Dig.