UNI Universal Natural Intelligence

Wiki · The Colony & the Method

Reproducibility Report

The Colony & the Method · docs/reports/reproducibility_report.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-four pages about the colony. Each agent is an Elixir process holding a generative model and doing inference, attached to a body that logs into a Minecraft world as an ordinary player. Around that sit the broadcast suite that films them and the runbooks that keep the whole thing running. There are typed specifications for each organ of the model, plus the world and genome specs. There are also the adversarial review personas used to attack a proposed change before it ships.

It is for the reader curious how a running system is put together and how it is held to account. The accountability half is the more distinctive. There is a lab protocol governing evidence and attribution, and a claim fence that restricts the vocabulary a claim is allowed to use. There is a public gate log. And there is a standing invitation to reproduce any verdict from the commit and the seed named in its receipt.

Start with the public read, then the lab protocol, then the falsification invitation. If you want the mathematics rather than the operations, go straight to the typed organ specs.

What it is not: a description of a mind, and not all one kind of document. A large part of this corpus is design and planning — specs marked as proposed rather than applied, organs designed but not built, plans that were later superseded — and each page states which it is. A specification is not a running system, and these pages are careful about the difference; the reader should be too. Eight documents were withheld from publication because they describe private infrastructure.

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.

Purpose

Verify Validation Invariant #13: reset with the same seed reproduces the same trace up to the declared tolerance; different seeds diverge.

Method

  • Pure SplitMix64 PRNG threaded explicitly (SP.Determinism); no :rand, no wall-clock, no external deps.
  • Compare full episode traces and world states for identical vs different seeds.
  • Pin a seeded episode in a golden artifact and re-derive it.

Artifacts used

  • SP.DeterminismTest, SP.World.DynamicsTest, SP.SimTest, SP.GoldenTest.
  • scripts/evidence.exs reproducibility section.
  • config/golden/reference_episode.json.

Result summary

From scripts/evidence.exs:

same-seed full-episode trace identical: true
same-seed world step identical:        true
different-seed traces differ:           true
  • SP.SimTest: same seed ⇒ identical Sim.points/1 and Sim.summary/1; different seed ⇒ differs.
  • SP.World.DynamicsTest: World.generate(77) |> step_n(120) identical across two independent runs.
  • SP.GoldenTest: the reference episode (seed 314, MorphologySeeking, 250 ticks) reproduces with structural metrics exact and float metrics within 1e-6. Because the golden artifact is generated by mix run and re-derived under the full mix test suite, this is a genuine cross-VM reproducibility check (it caught and now guards against atom-keyed map-iteration-order drift; see reproducibility.md).

Declared tolerance

  • Integer/structural metrics: exact.
  • Float metrics: within 1e-6 (absorbs serialization rounding only).

Pass/Fail

PASS.

Residual risks

  • BEAM does not guarantee float bit-identity across OTP versions in general; our arithmetic reproduces within tolerance across the CI matrix (OTP 27, Elixir 1.17/1.18). Integer/PRNG state is exact everywhere. See reproducibility.md.

sha256 b5d0fef1ba92c517 — 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)

This short report checks one thing: that starting again with the same seed reproduces the same run, and that different seeds diverge.

The method is stated in a line. All randomness comes from one generator threaded through the program by hand, with no system randomness, no clock, and nothing external. Whole episode traces and world states are then compared for identical and for different seeds, and one seeded episode is pinned in a stored artifact and re-derived.

The results are three plain true or false lines, all as hoped, plus the note that the pinned episode reproduces with whole-number metrics exact and decimal ones inside a stated tolerance.

The tolerance is declared rather than assumed, and the report ends by saying where exactness is not promised: decimal arithmetic is not bit-identical across all runtime versions, though whole numbers and generator state are.

Plain · written 2026-08-01 by claude-opus-5 · not yet checked by a person · about the document whose sha256 is b5d0fef1ba92c517

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)

This is one of a set of short validation reports, and it verifies a single invariant: resetting with the same seed reproduces the same trace up to a declared tolerance, and different seeds diverge.

The method rests on a design choice stated first. All randomness flows through one generator threaded through the program as an explicit value, with no system random source, no clock reading and no external dependency. On that basis, whole traces and world states are compared for identical and for different seeds, and one seeded episode is pinned in a stored artifact and re-derived.

The results are three plain statements: the same seed gives an identical trace, the same seed gives an identical world step, and different seeds differ. The pinned episode reproduces with structural metrics exact and decimal ones inside a small stated tolerance. Because the artifact is produced one way and re-derived another, the report calls this a genuine check across separate runs of the virtual machine, and notes that it caught a real ordering problem it now guards against.

The tolerance is declared rather than left implicit. The verdict is a pass, and the residual risk is that decimal arithmetic is not bit-identical across runtime versions in general, so the claim is limited to the tested matrix, while integer and generator state are exact everywhere.

Clear · written 2026-08-01 by claude-opus-5 · not yet checked by a person · about the document whose sha256 is b5d0fef1ba92c517