Cure-1 — Emergent food-foraging (the developmental ontogeny) — build receipt

Evidence & Verdicts · docs/receipts/emergent_forage_cure1.md @ 44baf03d5041 (gen2-runtime) — opens the published snapshot ac338733bbba

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.

PlainClearPrecise

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

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.

Status: offline-green + committed. NOT live-validated. No live deploy without a recorded RED verdict + owner go-ahead. Branch lab/ozone-life-uni-hard-science. Owner-directed (2026-07-11).

0. The claim (fenced)

A hungry deep-body UNI learns to hunt to stay alive — foraging that is constructed by the generative model + learning + innate priors, with no reward, no goal-code, no food give. Per the claim fence this demonstrates the named BEHAVIOUR (self-maintenance by foraging) only; it carries zero evidential weight for awareness, hunger-as- experience, or life. Every store / count / belief float named below is a MODEL VARIABLE.

1. The developmental ontogeny (owner directive — "no shortcut, all the way down")

We do not drop an infant into a hard world to learn-or-die. We build the feeding organism the way the universe does:

Ontogeny (biology)	UNI realisation	Where
Cellular nutrient hyper-prior — the cell predicts it will be fed to setpoint	interoceptive reserve-C (γ_m=1.0 large-magnitude C) + strong `pb_seed` 50 on energy_reserve/gut_satiety	genome.ex `card/1`, modality specs
Digestive/consummatory organ built but dormant pre-birth	`B[:attack]→has_food` present in the model but flat (Dirichlet counts unlearned) at birth	designer/factors, learned online
Womb/wean period — sustained gently while the organ matures	nursery `metab_scale` runway (core drain slowed s×) — NO manna	homeostat.ex `core_drain`, genome `nursery`
Weaning to self-feeding	graduation drops the scaffold → pure world, natural prey, zero gives	pureworld_qa.exs

Honesty correction (math-breaker C9, owed plainly): UNI's L2 is a control/preference hierarchy (situation observed up mc_codec, absolute C-override down mc.ex) — not a predictive-coding errors-up/predictions-down stack. The "hyper-prior" is a large-magnitude interoceptive C, not an elevated precision. The branching-tree / hormonal birth-activation is the direction, not the current fact — see §6 (next rung).

2. The emergence mechanism (reward-free, C-independent decay)

interoceptive-depleted (low energy_reserve bin) → L2 :forage context (mc_codec situation_index, gap 1).
:forage prey-orient C makes facing/closing prey pragmatic (mc.ex strategist_config, gap 2, gated to :homeostat).
:attack's transition column is under-sampled ⇒ transition-novelty W_b (plan.ex:142, gated ng>0) makes it worth trying — a standing GLOBAL epistemic drive over every under-sampled column (C4: it does NOT name :attack).
a world-earned kill (body.js collectDrops) lets Dirichlet B learn attack→has_food.
thereafter the forage has_food C selects the hunt pragmatically; the epistemic drive decays to 0 as counts saturate (novelty.ex, independent of C — the no-smuggled-reward property).

3. Typed spec — the code (additive, gated, default byte-identical)

Genome.homeostat_colony_forage(gain \\ 0.3) = %{homeostat_colony() | novelty_gain: gain} — the ENTIRE FE cure (precedent: curiosity_primary/0). All novelty machinery (card→designer→plan W_a/W_b) already exists + is gated.
Genome.nursery(gain \\ 0.3, scale \\ 0.5) = forage lineage + nursery: %{scale} (runtime-only womb-runway).
Genome struct: new gated field nursery: nil (+ slow_defaults back-fill).
Homeostat: new field metab_scale: 1.0; core_drain = … * frac * b.metab_scale (*1.0 bit-exact ⇒ byte-identical).
agent.ex: body: Homeostat.new(metab_scale: nursery_scale(brain.dna)); attack_count telemetry (C5, runtime-only); nursery_scale/1 helper; additive UNI_LINEAGE registry entries (live "homeostat_colony" case untouched).
homeostat_colony/0 (the LIVE streamed genome) is byte-untouched (C1). No adopt novelty override (C2). No cooked_beef / calorie give anywhere (C3). No strike motor-option / body.js FE change (Cure 2 deferred — C11).

4. Adversarial review of record

3-lens fork→break→repair→vote (math-breaker / fep-theorist / embodiment), MERGED VERDICT SIGN-WITH-CHANGES (workflow wmtheaty4, 2026-07-11). All required changes C1–C11 applied (see the build plan; wf_build_plan archived in the session scratchpad). A formal /lab-team-review re-run is available before any LIVE deploy if the canonical skill invocation is wanted.

5. Ship-gate checklist

Offline math fence — GREEN (mix test test/sp/brain = 339/0, mix compile --warnings-as-errors clean):

Default byte-identical — decider_byte_identity_test (mad<1e-12 vs frozen golden).
Live-lineage byte-identical — forage_discovery_gating_test T1b: homeostat_colony_forage(0.0) == frozen homeostat_colony golden (proves the streamed genome is untouched).
Drive is live — T1c: forage(0.3) ≠ forage(0.0).
Nursery runtime-only — T1d: nursery(0.3,0.5) decider == homeostat_colony_forage(0.3) (metab_scale doesn't touch A/B/C/D/E).
No scalar-per-action WITH novelty on — T2 clone-invariance holds (W_b column-local) + existing action_clone_invariance_test.
Monotone decay, C-independent — novelty_test (W→0 as counts→∞).
Body default bit-exact — rung-1 dynamics stays green (metab_scale 1.0).
homeostat_colony/0 unchanged; no goal-code (grep: no "if hungry+prey then attack"); no give in launchers.
Offline food-economy precheck — verify_forage_dynamics.exs 4/4: competent hunter survives (mean 0.807), naive eat-on-empty dies (world non-trivial ⇒ survival attributable to foraging).

Empirical gate — PENDING (needs the lab box + owner go-ahead; separate mc-nursery/mc-pure containers, kin 70/71):

Cure-1 forage RED (novelty-on vs novelty-off) recorded verdict before the nursery bundle (first rule, C11).
Graduation on the scaffold-free exam: survives, self-earned recoveries, ends reserve-band, zero gives.
analyze_forage_qa.py → PASS: the trained brain forages+survives in a PURE world on every seed AND the untrained twin does NOT (the discriminator) — zero VOID.

Live deploy — only after all above + owner go-ahead recorded (separate container, distinct kin + memory dir, ONE --sname node, never the streamed uni-colony). Then, and only then, ping the Producer for G2.

6. Next rung (NOT built this session — deferred, honors the first rule)

The one piece of the ontogeny not yet modelled: mom's signals + the baby's own hormone change that switch the feeding system on. Design seam: a developmental schedule that gates the epistemic term (novelty_gain) through SP.Brain.Hormones at a birth transition — an epistemic gate, not a reward. Requires its own lab-team review and must not stack until Cure-1's forage RED has a recorded verdict.

7. How to run (lab box, under owner go-ahead)

runs/nursery_forage_gate.sh — STAGE 0 offline precheck → nursery train (forage novelty + womb-runway) → pure-world QA trained+control → paired gate. Smoke: TRAIN_SEC=900 SOAK_SEC=600 SEEDS=1. Full: SEEDS="1 2 3", 1800s soaks.

sha256 4e5167498ebfae6d — of the original file, so what was ingested stays checkable.

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. Read the document itself → (or choose Precise in the reading-level control above)

A build receipt — the file recording what was run — for work that is finished offline and has not been tried live. The idea is a body that learns to hunt because it is hungry, with no reward and no food handed to it, and the page fences that claim tightly: it would show a behaviour and nothing about awareness or experience. The design follows a developmental path rather than dropping a newborn into a hard world, and the authors correct one of their own metaphors as too generous. The empirical gate is unrun, and the checklist for it is entirely unticked.

Plain · written 2026-08-01 by claude-opus-5 · not yet checked by a person · about the document whose sha256 is 4e5167498ebfae6d

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. Read the document itself → (or choose Precise in the reading-level control above)

A build receipt, the file recording what was run, whose status line is the first thing to read: offline-green and committed, not validated live, and no live deployment without a recorded verdict and an owner's go-ahead.

The claim is fenced before it is described. What is proposed would show a named behaviour, self-maintenance by foraging, constructed by the model and its learning rather than by a reward or by goal code. Every store, count and belief named is a model variable, and the limit stated says the work carries no evidential weight for awareness or experience.

The design is laid out as a development, deliberately, in a table pairing each stage in biology with how it is realised here. There is a prior expectation of being fed, and a consummatory part present but dormant at birth. Then a gentle period in which the pressure is reduced without anything being given, and then a weaning into a world with no scaffold at all.

Then an honesty correction the authors owe and pay. One part of the architecture is a control and preference hierarchy rather than the predictive stack the metaphor implies, so the thing being called a hyper-prior is a large preference rather than an elevated precision. The branching picture is named as the direction of travel rather than the current fact.

The emergence mechanism is written as a chain of steps, and its important property sits at the end. The drive that makes an untried action worth trying decays to nothing as the counts fill up. That decay is independent of preference, which is what keeps a reward from being smuggled in. The code section then lists what changed and, repeatedly, what did not, since the streamed live lineage is untouched and every addition is switched off by default.

A ship-gate checklist follows, split in two. The offline part is ticked, with the test that carries each claim named beside it. The empirical part is entirely unticked and needs a lab box and an owner's go-ahead. The order is enforced rather than suggested: the paired comparison must have a recorded verdict before the next stage runs.

Finally, the part not built. The piece of the developmental story that would switch the feeding system on at birth is deferred rather than stacked, with a note that it needs its own review and must wait for the current verdict.

Clear · written 2026-08-01 by claude-opus-5 · not yet checked by a person · about the document whose sha256 is 4e5167498ebfae6d

Cure-1 — Emergent food-foraging (the developmental ontogeny) — build receipt

How to read this page

Precise — the source document