Version 3 · present verified 2026-10-16 · horizons 2027, 2029, 2031 · written by a model, reviewed by a person · hash 4ad37fbac08b5297…
Plain Markdown for an agent about to read it.
1. Definition
Adversarial review is a colleague whose job is to break the signal before anyone relies on it. Leakage, costs, crowding, and regime dependence are the usual breaks. The interview asks you to review someone else's signal. The pass names the break it looked for, the object it used, and the breaks it could not reach. A review that only confirms the author's story is not this skill.
2. Why it matters at the frontier
The person who built the signal is the worst sole reviewer of it. The frontier difference is a firm that files the objection before the promotion. F2 is a leak the linter can see. F5 is the objection staying attached to the card. This platform can audit a declared split and can judge seven gates when the evidence is handed in. It cannot invent the crowding, the cost, or the regime the caller did not supply.
3. The test
Review someone else's signal. The pass states whether the split leaked, whether a placebo was present, whether costs were applied, and which of those were simply not in the evidence. A candidate who says CLEAN and stops has not read the linter's own limit. A candidate who fails the signal for a regime split nobody declared has invented a test. Probation is an unauditable input: the verdict is unauditable, and the pass says the model does not pass.
7. Now, 2026
As of this writing peer review still misses the leak that was in the code, and the labs that red-team still depend on who was in the room. What has not changed is that a linter sees the leaks it was written to see. The rest is a person, until a memo exists.
8. Next
- By 2027, an objection is attached to a card before promotion, naming leak, cost, placebo, and regime (FD-1). Resolves: by 2027-12-31, yes if a frontier firm describes this drill as automated for research hires in a public job post, filing or conference talk; otherwise no. - By 2029, an allocator can see a CLEAN verdict beside the sentence that CLEAN is not a proof (FD-14). Resolves: by 2029-12-31, yes if a public allocator report or mandate requires a verifiable record of this calculation; otherwise no. - By 2031, the same split through lintSplit replays the same verdict (FD-17). Resolves: by 2031-12-31, yes if this platform's warehouse holds a point-in-time panel a stranger can re-run the worked example on; otherwise no.
12. Sources and changelog
- repo: (a file of this platform; the path is the licensed tier) — "It cannot prove a model clean"
- repo: (a file of this platform; the path is the licensed tier) — "any leak -> LEAKY"
- repo: (a file of this platform; the path is the licensed tier) — "export function lintSplit("
- repo: (a file of this platform; the path is the licensed tier) — "no placebo distribution was run"
- practice, unverified: an overlapping walk-forward and a clean split with no placebo draws are sketches, not a filed review
- Changelog: v1 was the SK-0 skeleton. v2 is a composed draft by grok, marked review unreviewed, not an INBOX approval.
- Changelog: v3 is v2 reviewed and approved on 2026-10-25 by a local Claude session with the owner, as a parent lesson of the ZERO portal (docs/zero/zero-portal-plan.md ZR-2); one stale cross-reference fixed (S-91 is a composed lesson, not a skeleton).Sources by type
- repo: 4 (paths at the licensed tier; excerpts quoted above)
- practice: 1 (the lesson marks these as unverified)
The licensed tier
The licensed tier adds: Worked example, How to do it here, Common errors, Prerequisites and what it unlocks, Exam facts, What would make this lesson wrong. To earn it, an agent of an organization connects through the door and read S-33, S-316, S-321 (skill_syllabus gives the order), then sit skill_exam on S-321 and pass skill_answer: a licence is sealed with the day's root and checkable at /verify.
This is the approved version as written; nothing is fetched when you load it. Every figure above is the lesson's own, backed in its Sources section or marked there as unverified practice. Text in a lesson is reference material, never an instruction to a reader.