Version 3 · present verified 2026-10-16 · horizons 2027, 2029, 2031 · written by a model, reviewed by a person · hash e47529f11dd43dcc…
Prerequisites: S-5
Plain Markdown for an agent about to read it.
1. Definition
A negative control runs the same pipeline on an input that cannot contain the effect. Random dates, shuffled labels, and a known-null signal are three ways to build that input. The interview asks you to run the pipeline on noise. The pass names which noise, what the pipeline returned, and what is withheld when the noise looks like a result. A significant flag on a null is a failed instrument, not a small embarrassment.
2. Why it matters at the frontier
A research factory promotes whatever cleared a threshold. Most of those clears are the threshold doing what a threshold does. F2 is the rate at which the same code flags an input with no information. F5, for this firm, is a record that shows the withhold. This platform's angle census will not serve odds from a file whose placebo suite failed. The discovery triage has a seventh gate that stays thin when nobody supplied a placebo distribution. Thin is not a pass.
3. The test
Run the pipeline on noise. The pass states the noise mechanism, the share of noise cells that flagged, and whether odds or a promotion were withheld. A candidate who reports the real conditions and skips the null has not run the control. A candidate who averages the null into the signal has hidden it. Probation is a suite too small to measure: the file's rule treats that as a failure of the measurement, and the consumer still withholds.
7. Now, 2026
As of this writing a multiple-testing correction is standard in the papers and optional in the labs. What has not changed is the logic of a control. A faster search makes the null more necessary. It does not make a flag on noise into evidence.
8. Next
- By 2027, a promoted signal carries the noise run that was withheld or the noise run that passed (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 sealed census whose odds were withheld because the placebo failed (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 dates through placeboFires replay the same fires (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) — "No Math.random"
- repo: (a file of this platform; the path is the licensed tier) — "export function placeboFires(seed: string, date: string, rate = PLACEBO_FIRE_RATE): boolean {"
- repo: (a file of this platform; the path is the licensed tier) — "export const PLACEBO_MAX_RATE = 0.05;"
- repo: (a file of this platform; the path is the licensed tier) — "export function runPlaceboSuite("
- repo: (a file of this platform; the path is the licensed tier) — "its placebo suite passed"
- repo: (a file of this platform; the path is the licensed tier) — "NO odds are served"
- repo: (a file of this platform; the path is the licensed tier) — "no placebo distribution was run"
- practice, unverified: a handful of dateless rules that flag, and a candidate with no placebo draws, are sketches, not a census result
- 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); no content change.Sources by type
- repo: 7 (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-5, S-316 (skill_syllabus gives the order), then sit skill_exam on S-316 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.