UNI Universal Natural Intelligence

Wiki · Evidence & Verdicts

RED pre-registration — glands-phase5

Evidence & Verdicts · docs/receipts/red_preregistration_glands_phase5.md @ 44baf03d5041 (gen2-runtime) — opens the published snapshot ac338733bbba

How to read this page

Three ways to read this page. Precise is the document itself, exactly as it is written in the repository. Plain and Clear were written for this website to help you meet that document — they are about it. They are not it, and they are not evidence.

Eighty-seven dated pages: receipts, pre-registrations, handoffs, validation records and review verdicts. A receipt is written at the moment a piece of work was checked. It names what was claimed, the commit and the seed, what was actually run, and the outcome in one of a small set of controlled words. Then it names what the work did not achieve. That last part is what makes it a receipt rather than an announcement. A pre-registration is the same discipline run in advance: the conditions that would count as a pass and the conditions that would falsify the claim are written down before the run, so neither can be adjusted once the numbers arrive.

That is why so many small dated stubs are an audit trail rather than noise. No one of them is meant to be a good read. The value is in the sequence and in the dates, because you can watch a prediction be registered, then the run happen, then the verdict land — sometimes against the prediction. Pages here record a falsified result, a rejected fix, a retracted overclaim, and a green receipt that turned out not to be reproducible from the commit that carried it. A record that carried only successes would be worth a good deal less than this one.

A gentle way in is to read a pre-registration first, so the shape becomes familiar, then a result page, then one of the corrections. This section sits off the main navigation on purpose: it is the record you check the rest of the site against, not the place to begin.

What it is not: documentation, and not a summary. Nothing here has been tidied in hindsight. Every entry reads as of its date, a later entry may overturn an earlier one, and the presence of a page is not a claim that its result stood.

Your browser cannot switch reading levels, so the document itself is shown.

Precise — the source document

This is the document. Rendered from the repository at the commit above, with nothing rewritten for the web. A gate re-renders it on every deploy and fails the build if a single byte differs.


verdict: WITHHELD evidence_class: pending

RED pre-registration — glands-phase5

  • Gate name: glands-phase5
  • Phase: Phase 5 (endocrine / persistent glands)
  • Pre-registered: 2026-07-13
  • Runner: runs/glands_red.exs
  • Related: docs/UNI_MISSION_DEEPENING.md, lib/sp/brain/hormones.ex

Motivation

Persistent-gland-state organ biases policy through satiety-attenuated positive C on a whitelist. The risk: this is the classic place reward can be smuggled. The RED asserts NO scalar-per-action leakage AND byte-identity preserved.

PASS condition

  • Gland lineage engages: satiety trajectory shows attenuation of positive C on the whitelist over the pre-registered window.
  • No-smuggled-reward: the action_clone_invariance_test.exs PASSES with gland organ present.
  • Byte-identity of default_genome PASSES with gland organ absent.

FALSIFIES condition

  • Any evidence of scalar-per-action leakage into policy logits (clone/inject-cost/perturb-one guards fail); OR
  • default_genome byte-identity breaks; OR
  • The gland attenuation signature is not observable (organ inert).

Protocol

  1. glands_lineage/0 genome with the endocrine organ; coupling 0.0 default.
  2. Full invariant suite: decider_byte_identity, action_clone_invariance, novelty.
  3. Live diagnostic: satiety trajectory + policy logits under matched-vs-satiated conditions.
  4. Verdict:
    • PASS: all invariants pass AND attenuation signature observable.
    • FAIL: any invariant fails OR attenuation not observable.

Ship-gate

This is a Phase 5 organ. Absolutely no FE code changes to hormones.ex / precision.ex before the RED PASS + MERGED VERDICT.

sha256 81346099c958dc5a — of the original file, so what was ingested stays checkable.

Plain — written for this website, not the source document

Written for this website — not the document. This is a plain-language retelling, written to help you meet the document. It is not the source, and it is not evidence. It has not yet been checked by a person. (or choose Precise in the reading-level control above)

A promise made before an experiment, and not a report of one. It sets out a test of a proposed part of a system that would nudge behaviour according to an internal state, something like being fed or being hungry. The authors name the danger themselves: this is exactly the place where a hidden reward signal could be smuggled in, and they do not want one. So the test is written to catch that. The page carries no verdict, because the work had not been done.

Plain · written 2026-08-01 by claude-opus-5 · not yet checked by a person · about the document whose sha256 is 81346099c958dc5a

Clear — written for this website, not the source document

Written for this website — not the document. This is a clearer retelling, written to help you meet the document. It is not the source, and it is not evidence. It has not yet been checked by a person. (or choose Precise in the reading-level control above)

Written as a pre-registration, meaning the conditions are set down first, so nothing on it has been run. It describes an organ that had not yet been built, and the checks that would have to pass before it could be. The verdict is withheld.

The header names the gate, the phase, the date it was registered, the script that would run it, and the related notes. The short motivation paragraph is the honest heart of the page. The proposed organ biases choice according to an internal state. The authors say straight out that this is the classic place a reward signal can be smuggled into a system that is not supposed to have one.

So the pass condition has three parts. The organ must do something observable, a stated signature over a stated window. An invariance test must still pass with the organ present, which is the part that would catch a hidden per-action reward. And a default genome must stay byte-identical with the organ absent. Any one of those failing is a failure. A closing rule forbids touching the related code at all until this test passes and a review agrees.

Clear · written 2026-08-01 by claude-opus-5 · not yet checked by a person · about the document whose sha256 is 81346099c958dc5a