L2 — Tissue / metabolism (interoception and energy)

The Cookbook · cookbook/recipes/L2-metabolism.md @ 575fc93d9d31 (main) — opens the published snapshot e850f872196d

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.

The Cookbook is the method carried out step by step: 34 pages of recipes for building a developmental active-inference SIMULATION — a bounded peek at a toy world, never a person. The front matter says that word is never softened under any pressure, so it is not softened here. The recipes run from the molecular and cellular rungs up through metabolism, motor control, perception, language and metacognition, and on to rungs that are still open questions. Around them sit a set of kitchen rules, a shared pantry of engines and primitives, and a second family of recipes about nature itself — rocks, water, air, stars, DNA, ants, whales, bats, humans.

It is for the reader asking what building this would actually take. Each recipe names its ingredients, the order of work, the tests to run at that stage, and the point at which a step stops being something already carried out and becomes something proposed.

Begin with the front matter and then the kitchen rules. Those two pages fix the honest position and the fence labels that every later recipe leans on, and without them the status markers on a recipe are easy to skim past. After that the recipes can be read in any order.

The nature recipes sit slightly apart and should be read that way. They cite outside science — geology, chemistry, biology, astrophysics — and a nature citation is never a UNI gate: those chapters contain zero UNI claims and raise no rung.

What it is not: a claim that the whole ladder has been cooked. The book recommends the complete recipe and, on the same page, labels every rung by its real state — that tension is deliberate and is the thing the book is built around. Where a recipe and the claim ledger disagree, the ledger wins.

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.

Recipe CB-L2. A chapter of the literal UNI Cookbook. The Cookbook fully recommends the complete build to L12, but every step is labeled by its REAL status drawn verbatim from ../../encyclopedia/CLAIM-LEDGER.md. Where this recipe and the ledger disagree, the ledger wins and this recipe is wrong.

Honest program position (printed, never softened): ~2 of 11+ developmental rungs earned. This whole program is a developmental active-inference SIMULATION — a bounded peek, a toy world, never a person.

What you are building

A standing metabolic drive for the embodied colony: an interoceptive organ that gives a body a real energy/satiety budget and a viability edge (it can die), so that maintaining itself becomes metabolically necessary. The pantry framing is WORLD ⊥ BODY ⊥ MIND (M12): interoception is a body signal, never a felt state. Affect is modeled, never felt.

You are building this as a proven foraging/crafting driver and, in the same breath, recording its first-class NEGATIVE as a building driver. The headline temptation — "metabolism breaks the plateau to stone and shelter" — is the gate G6, and G6 is OPEN. Do not let the +135 percent uplift spin into a plateau-break.

Ingredients

Drawn from the shared pantry by name:

The metabolism organ (strings): standing-metabolic-drive interoception organ; energy/satiety factors; a draining/refilling emptying-B; setpoint-peaked preferences C; the :pb_seed strong-Dirichlet seam; a live viability edge in the bridge. Additive + genome-gated (default byte-identical).
The Z affect modulator (uni-gpt / uni-mind): the global [energy, arousal, valence, fatigue, pain, threat, safety, inflammation] vector that carries the interoceptive signals into precision / preferences / habits / learning-rate / horizon.
The Heart Lab (uni-precision / worldmodels) as the physiological mirror of the same one engine (the reduced Karaaslan cardio-renal homeostasis loop) — "same math, many scales," used here only as the cross-scale teaching companion, not as a metabolism gate.
One-cure-at-a-time paired RED discipline (M21): paired kin-N treatment vs kin-N+1 control; never stack changes so the winning outcome is unattributable; an offline RED pre-check before any live burn.
The JAX POMDP + EFE + Dirichlet engine (core.py) for the no-backprop loop; the read-only counterfactual-EFE (shadow-EFE) audit path for diagnosing the plateau on the real brains.

Method

Ship the organ additive + genome-gated. Install the metabolism organ as a new opt-in genome organ absent from default/0, so the default streamed colony stays byte-identical — verify with the mad < 1e-12 golden-fixture tests over the depth-5 planner. The full suite must read 297/0 before you proceed (L2.1).
Seed the strong Dirichlet metabolic prior only via the :pb_seed seam. You CANNOT seed a strong prior by pre-scaling B: norm_cols runs before add1, which wipes any magnitude you tried to bake in (L2.4). The :pb_seed concentration field is applied after norm_cols. This seam is the only place a "10-100x lifetime strong prior" can be expressed.
Wire a real viability edge into the live bridge. The naive design is a trap: in the live bridge, metabolize / Viability / shutdown were Sim/Eval-only, so a naive emptying-B drained a belief with zero world consequence — an all-:noop twin stays exactly as "viable" as an actor (L2.4). Wire drain-per-tick, refill-only-when-the-body-has-food, and die-at-empty into bridge.ex so foraging becomes metabolically necessary and a body can actually die. Add satiety→C appetite attenuation so a sated body's preferences relax.
Run an offline RED pre-check before any live burn (M21). Run the real engine in a synthetic world first. This is load-bearing: the offline pre-check previously caught an "eat-every-tick attractor" (the seeded drain model was ~6x too pessimistic vs the real store) before a live RED was wasted. Fix the generative model (gentle drain; "eating is need-driven, not a habit," mirroring the :noop-is-non-habitual rule) until offline G0/G1/G2/G3/G5b are green.
Run the pre-registered 12 h live paired RED, one cure at a time. Unit = a matched kin pair: treatment = organ-on (kin-N), control = kin-N+1 with the organ off, identical code / world / body, only the gated metabolic coupling differing. Register up front: the G4 allostasis separation gate, the G6 plateau-break gate, and the G5b action-severed-twin falsifier. Collect server-authoritative behavior via RCON (<=10-min cadence) plus brain probes at start/mid/end so the record survives context compaction. Reset the RED clock to a clean T0 if the treatment arm's brains crash during setup (pre-T0 data is exploratory only).

Gate

Two distinct gates live at L2. Carry both at their exact ledger figures and never collapse one into the other.

The proven uplift gate (L2.1, Class C). The first 12 h live RED is a PASS for metabolism as a foraging/crafting driver: +135% / 2.35x tool-crafting and +19% mining — a real, attributable standing-metabolic-drive effect, with the default colony byte-identical and the suite at 297/0.

G6 — the load-bearing plateau-break gate (OPEN). A disciplined RED in which the metabolism organ alone improves building, with the placed-blocks CI excluding 0, AND the G4 allostasis index separates in the intended direction. Until both hold, G6 is OPEN and no plateau-break may be claimed. Per the ledger, G6 is presently contradicted by its own first evidence (see the recorded NEGATIVE).

Falsifier (operable)

The proven uplift is falsified if any of:

a repeat pre-registered 12 h live RED fails to reproduce the tool-crafting uplift within CI; OR
the metabolism-organ ablation does not remove the uplift (the gain was not attributable to the organ); OR
the suite drops below 297/0 (byte-identity / regression broken).

G6 is discharged (not falsified — earned) only by a subsequent disciplined RED where the organ alone improves building (placed-blocks CI excludes 0) and G4 allostasis separates.

The standing G5b action-severed-twin is the falsifier that any "self-maintenance / life" language must clear: an :noop-twin with action severed must NOT stay as viable as an actor. Any such language that has not cleared G5b is an overclaim.

Recorded NEGATIVE(s) (first-class, inline)

These are content, not failures to hide — they are the credibility of the rung.

L2.2 (NEGATIVE, Class C) — metabolism is NOT yet a building driver. In the same 12 h RED that produced the +135% crafting uplift, building (placed blocks) went WORSE: −14%, and G4 allostasis never separated. The load-bearing claim "metabolism breaks the plateau to stone/shelter" (gate G6) is therefore OPEN and contradicted by its own first evidence. Do NOT spin the +135% as a plateau-break. Falsifier-to-discharge: a subsequent disciplined RED shows placed-blocks CI excludes 0 AND G4 separates.
L2.3 (NEGATIVE, diagnostic; Class A shadow-EFE audit on real brains) — the plateau is epistemic starvation, not gamma-runaway. The colony plateaus at "make a tool" (one UNI hoarded 32 pickaxes, never built). A read-only counterfactual-EFE (shadow-EFE) audit on the real hoarder .bin brains diagnosed epistemic_starvation — NOT gamma-runaway (gamma ≈ 7.8, unsaturated) and NOT a curriculum ceiling; the EFE landscape is pragmatic-saturated and flat, and the information drive is ~100x too weak. Falsifier: a gamma-saturation finding, a curriculum-ceiling flip, or an info-drive scaling that breaks the plateau without organs would overturn the diagnosis.
L2.4 (NEGATIVE, fixed; Class A direct code reading) — two engine seams that invalidated the naive design. (1) You cannot seed a strong Dirichlet prior by pre-scaling B (norm_cols runs before add1), so the new :pb_seed seam is required. (2) The live bridge had no viability edge — metabolize / shutdown were Sim/Eval-only, so a naive emptying-B drained a belief with zero world consequence. Both fixed in the shipped organ. Falsifier: a seam allowing strong-Dirichlet seeding without :pb_seed, or evidence the live bridge already had a viability consequence.

UNI-GPT consult (2026-06-27 — SIGNED): the smallest G6 cure is a Class-C DESIGN HYPOTHESIS; G6 stays OPEN

The diagnosed cure for the plateau is a structurally distinct organ — NOT a gamma change, NOT a second metabolism organ, NOT a placed-block reward bonus. Cross-ref ../UNI-GPT-CONSULT-2026-06-27.md Q3 (SIGN-WITH-CONDITIONS). This is DESIGNED / not-run. It RAISES NOTHING. The rung's status is UNCHANGED and G6 stays OPEN.

The organ — build_epistemic_frontier. Add one building-specific hidden factor + one policy term valuing info gain about where a block can usefully be placed next:

hidden factor: z_build in {unknown_placeable, placeable_support, shelter_contributing, blocked/useless}
drive: maximize expected info gain over z_build for candidate inspect/move/place policies
scope: active only when shelter/stone plateau preconditions are near but not achieved
epistemic term per policy: G_new = G_old - beta_build_IG * E_Q[ D_KL( Q(z_build|o,pi) || Q(z_build|pi) ) ]
scale beta by matching, not hand-waving: beta_build_IG = median(|delta pragmatic G for food/tool|) / median(|delta build IG term|); clamp initial to {25x, 100x}; 100x only if the offline RED shows 25x underpowered.

Why epistemic, not gamma. Gamma (≈7.8, unsaturated) is a precision over G — raising it only sharpens the existing ranking. If build-relevant info gain is missing or ~100x too small, gamma sharpens the wrong ranking. The cure makes build-relevant uncertainty part of what the planner can value. (The organ metaphor stays Class-C design language; the standard active-inference part is only the EFE decomposition + policy selection, at textbook level.)

The paired RED — G6_BUILD_EPI_FRONTIER_PAIRED_RED_v1 (designed). Matched kin-pair seeds; control = kin-N+1 (current organ/gamma/build machinery, no frontier term); treatment = kin-N (identical + build_epistemic_frontier, beta fixed from the offline RED, gamma unchanged). One cure at a time (M21).

Offline RED pre-check (replay prior traces, read-only) — pass only if all hold: a build-policy rank shift appears (treatment > control); gamma non-diagnostic (unchanged, unsaturated); the info term is causally responsible (delta-G_build_IG explains the shift); foraging/crafting not cannibalized. Suggested bar: >=25% relative increase in build-relevant policies entering top-3, food/tool rank within +/-5%, gamma unsaturated.
Live paired RED: unit = matched kin pair; same 12 h / world dist / paired seeds; no peeking-based tuning. Primary: placed_blocks_delta = T - C, 95% paired bootstrap CI excludes 0 (positive). Co-primary (G4): the existing ledgered allostasis_index delta CI excludes 0 in the intended direction, no viability collapse (do NOT invent a metric after the run). Secondary: stone/shelter progress > 0, crafting/mining non-inferior, distance-to-shelter improves.
Required ablations (gain MUST collapse): IG-zero (beta_build_IG = 0), shuffled-affordance (z_build permuted across sites), gamma-only (no organ, gamma matched). Credited only if the gain appears with build IG present and collapses when the IG channel is zeroed/scrambled, while gamma-only fails to reproduce it.

Falsifiers (Q3). Placed-blocks CI includes 0 / is negative; OR blocks improve but G4 never separates; OR the ablation fails to collapse the gain; OR gamma-only reproduces it (the "not gamma" diagnosis was wrong); OR blocks bought by damaging survival/tooling beyond non-inferiority. If the offline RED predicts no rank shift but live improves, log "behavioral improvement observed; mechanism not proven."

Status. Until the RED clears both bars and the ablations collapse the gain, build_epistemic_frontier is a Class-C design hypothesis supported by a read-only counterfactual diagnosis, NOT an achieved plateau-break. G6 stays OPEN.

HONEST FENCE

proven (as a foraging/crafting driver) + NEGATIVE (as a building driver). Class C.

proven: a held, signed result exists (Class C, L2.1) — the first 12 h live RED is a real, attributable standing-metabolic-drive effect: +135% / 2.35x tool-crafting and +19% mining, default colony byte-identical, suite 297/0. Its falsifier is still live (a repeat RED must reproduce within CI; the organ ablation must remove the uplift).
NEGATIVE: in the same RED, building went WORSE (−14%) and G4 allostasis never separated (L2.2). The plateau diagnosis is epistemic_starvation, not gamma-runaway (gamma ≈ 7.8 unsaturated, info-drive ~100x too weak; L2.3).
designed (RAISES NOTHING): the SIGNED build_epistemic_frontier organ and its paired RED are a Class-C design hypothesis, folded in as DESIGNED / not-run. The rung's status is UNCHANGED. G6 stays OPEN.

The not-claimed line. Metabolism is proven as a foraging/crafting driver, NOT a building driver. The plateau-break (G6) is UNPROVEN and is contradicted by its own first evidence. Do NOT spin the +135% uplift as "breaks the plateau." This is not life, not a synthetic organism, not self-aware, not consciousness, not "active inference demonstrated"; energy and satiety are modeled, never felt. The G5b action-severed-twin is the standing falsifier any "self-maintenance / life" language must clear. This rung remains a developmental active-inference simulation — a bounded peek, a toy world, never a person. ~2 of 11+ rungs earned.

sha256 21346fa3663ec72b — 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)

This recipe gives a simulated colony a metabolism: an organ that hands each body a real energy budget and an edge past which it dies, so that keeping itself going becomes necessary rather than optional. Energy and satiety are modelled here, never felt.

The one thing this page says is that a single run produced a win and a loss, and both are printed in the same breath. Metabolism worked as a driver of foraging and crafting. In that very same run, building got worse and a second measure never separated. So the headline everyone reaches for — that metabolism breaks the plateau through to stone and shelter — is a gate that stays open, and the page says outright that it is contradicted by its own first evidence.

There is a diagnosis and a proposed cure. A read-only audit of real saved brains suggests the colony is starved of the sort of curiosity that would make building worth exploring, rather than being over-sharpened or capped by its curriculum. A signed design proposes one extra organ to test exactly that. It has not been run, it raises nothing, and the gate stays open.

Plain · written 2026-08-01 by claude-opus-5 · not yet checked by a person · about the document whose sha256 is 21346fa3663ec72b

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)

This chapter builds a standing metabolic drive for an embodied simulated colony: an interoceptive organ giving a body an energy and satiety budget and a real viability edge, so that self-maintenance becomes metabolically necessary. The framing is stated up front — signals from the body are body signals, never felt states — and the honest programme position is printed in the opening block rather than the closing one.

The build steps are unusually candid about traps. The organ ships as an opt-in addition so that the default colony stays byte-identical, checked by golden-fixture tests and a full suite that must be green before anything else proceeds. A strong prior cannot be seeded the obvious way, because a normalisation step runs first and wipes any magnitude baked in beforehand, so a dedicated seam exists and is the only place that prior can be expressed. A viability edge must be wired into the live bridge. In the original design the relevant routines existed only in simulation and evaluation, so a draining budget drained a belief with no consequence in the world at all. A twin that did nothing stayed exactly as viable as one that acted. An offline pre-check is run before any live burn, and the page records that this pre-check already caught a bad attractor before a live run was wasted. Only then does the live paired experiment run, one cure at a time, with a matched pair of kin as the unit and only the metabolic coupling differing between them.

Two distinct gates live here and the chapter insists they never be collapsed into one. The first is the uplift gate, which passed: a real, attributable effect on tool-crafting and mining with the default colony unchanged. The second is the plateau-break gate, which requires that the organ alone improve building with its interval excluding no-difference, and that a separate measure of regulation separate in the intended direction. Until both hold, that gate stays open.

The recorded negatives are the substance of the page. In the same run that produced the crafting uplift, building went worse and the regulation measure never separated, so the plateau-break claim is open and contradicted by its own first evidence. A read-only counterfactual audit on real saved brains diagnosed the plateau as a starvation of information-seeking rather than as an over-sharpened ranking or a ceiling in the curriculum, with the planner's landscape flat and the information drive far too weak. And two engine seams that invalidated the naive design are recorded as negatives too, both since fixed.

The last section is a signed consultation design for the smallest cure. It adds one hidden factor about where a block could usefully go next, plus one policy term valuing information gain about it, with its weight set by matching rather than by hand-waving. A paired experiment is designed around it with required ablations that must collapse the gain, and a list of results that would sink it, including the possibility that the sharpening explanation was right all along. The status is stated plainly: designed and not run, raising nothing, with the gate still open.

The closing verdict carries both halves together: proven as a foraging and crafting driver, negative as a building driver, with the cure designed only.

Clear · written 2026-08-01 by claude-opus-5 · not yet checked by a person · about the document whose sha256 is 21346fa3663ec72b

L2 — Tissue / metabolism (interoception and energy)

How to read this page

Precise — the source document