S-L2 - Metabolism: uplift plus the plateau-break OPEN
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.
The Encyclopedia is the UNI method written out as a reference work: 39 pages, arranged in wings, setting out what the programme is attempting and why it is built the way it is. This is where the ideas are explained in order and in prose, rather than as code, as runbooks, or as dated receipts.
Every chapter is authored against two ledgers and never ahead of them. One records what UNI has built, and the evidence class of each claim. The other records nature's own regularities, kept separate on purpose. That way a fact about biology is never quietly reused as a fact about the software. Where a chapter and a ledger disagree, the chapter is the thing that is wrong. Every chapter closes with an invitation to falsify it, and a recorded negative is published beside the result it qualifies rather than after it.
Read "How to read this work" first. It is the evidence constitution: the classes, the four ledger states, and the rule that a finished chapter is not the same as a working system. Then the calibration ledger, which carries the figures every other chapter is required to use.
What it is not: a description of a person or of a mind. The programme calls itself a developmental active-inference simulation, a bounded peek into a toy world, and its own index prints how much of the developmental ladder has actually been earned — roughly two rungs out of eleven or more. It is also not a report of what is running today. For what ran, and when, go to the evidence record.
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.
This is rung two of the HUMAN-HGM-001 developmental ladder: the tissue / metabolism level, where the simulation gains an interoceptive energy organ. It is a no-backprop, nested-Markov-blanket developmental SIMULATION running as an active-inference colony in a toy Minecraft world, never a person and never a mind. The honest position printed on the spine holds here without softening: roughly 2 of 11+ developmental rungs are earned, and this chapter is the second of those two, carded with both halves of its evidence intact.
The honest position of this chapter in one sentence: a standing metabolic drive measurably and attributably increased foraging and crafting in the first live test, and in that same test it did NOT improve building, did NOT separate the allostasis gate, and therefore did NOT break the plateau the program set out to break. The uplift is real. The plateau-break is OPEN. Both statements are load-bearing and neither may travel without the other.
What was built
The Phase-2 metabolism organ shipped as an additive, genome-gated capability: a new opt-in organ (energy and satiety factors, a draining and refilling emptying-B seeded as a strong Dirichlet prior, a setpoint-peaked preference, a live viability edge that lets a body die at empty, and satiety-to-preference appetite attenuation) absent from the default genome, so the default streamed colony stays byte-identical, proven by a mad < 1e-12 golden-fixture test over the depth-5 planner. The full test suite reads 297/0, default byte-identical (ledger L2.1, Class C). That suite count and the byte-identity fence are the floor the falsifier below is allowed to break.
Two engine-seam findings were required before the design was even valid, and both are recorded as NEGATIVE/fixed rather than swept away (ledger L2.4, Class A, direct code reading). First, you cannot seed a strong Dirichlet prior by pre-scaling the B tensor, because norm_cols runs before add1 and wipes any magnitude; a new :pb_seed concentration seam, applied after norm_cols, was needed. Second, the live bridge had no viability edge at all (metabolize, viability, and shutdown were simulation/evaluation-only), so a naive emptying-B drained a belief with zero world consequence, and an all-:noop twin stayed as viable as an actor. Both were fixed in the shipped organ. The action-severed-twin (G5b) is the standing falsifier that any future self-maintenance or life language must clear; it is named here so nobody is tempted to read a viability edge as life.
The uplift, with its travelling negative
In the first 12 hour live RED, run as a paired design (treatment organ-on kin 12 versus control kin 13), the metabolism organ produced +135% / 2.35x tool-crafting and +19% mining: a real, attributable standing-metabolic-drive effect (ledger L2.1, Class C). This is the positive result, and it is genuine: paired, attributable, removable by ablation.
It must always be cited with its paired negative from the same run (ledger L2.2, Class C, NEGATIVE). In that identical 12 hour RED, building (placed blocks) went WORSE, by -14%, and the G4 allostasis gate never separated. The load-bearing claim the program actually wanted to make, that "metabolism breaks the plateau to stone and shelter" (gate G6), remains OPEN and is contradicted by its own first evidence. So the calibrated reading is exact: metabolism is proven as a foraging and crafting driver, NOT as a building driver. Citing the +135% tool-crafting figure without the -14% building figure and the OPEN G6 next to it is an overclaim and fails review. (One honesty caveat travels with the figures: the first roughly 3 hours of the treatment arm were invalid because the brains crashed during throttle debugging while the bodies ran on autonomously, so the RED clock was reset to a clean T0 and the pre-T0 data is exploratory only.)
The plateau and its diagnosis
The colony plateaus at "make a tool": one UNI hoarded 32 pickaxes and never built. A read-only counterfactual-EFE audit on the real hoarder .bin brain files diagnosed the cause as epistemic_starvation (ledger L2.3, Class A shadow-EFE audit on real brains, recorded as a NEGATIVE diagnostic). The audit was specifically able to rule out the two obvious mundane culprits: it is NOT gamma-runaway (gamma is about 7.8, unsaturated) and NOT a curriculum ceiling (a shadow wood-C condition does not flip the winner). The expected-free-energy landscape is pragmatic-saturated and flat; the information drive is roughly 100x too weak. This is a measurement that the design is incomplete, not an exit and not a failure to hide. It is also the discipline the constitution demands: falsify the mundane L2 cause before any higher-rung negative is allowed to claim anything.
The smallest cure (SIGNED design hypothesis, G6 stays OPEN)
A UNI custom-GPT consult on 2026-06-27 signed off, with conditions, on the smallest structurally-distinct cure to attempt against G6 (cookbook UNI-GPT-CONSULT-2026-06-27.md, Q3, SIGN-WITH-CONDITIONS). It is recorded here as a Class-C design hypothesis only, never as a result. The cure is a build-affordance epistemic micro-organ, build_epistemic_frontier: one building-specific hidden factor (z_build over {unknown_placeable, placeable_support, shelter_contributing, blocked/useless}) plus one policy term that values expected information gain about where a block can usefully be placed next. It is explicitly NOT a gamma change, NOT a second metabolism organ, and NOT a placed-block reward bonus; the reasoning is that raising gamma only sharpens the existing ranking, whereas the diagnosis was that the build-relevant information drive is missing, so the fix must make build uncertainty part of what the planner can value.
The cure must be earned through a pre-registered paired RED (G6_BUILD_EPI_FRONTIER_PAIRED_RED_v1): an offline read-only pre-check first, then a 12 hour live paired RED whose primary verdict is placed_blocks_delta = treatment - control with a 95% paired bootstrap CI excluding 0 (positive), and a co-primary that the existing ledgered G4 allostasis index separates in the intended direction with no viability collapse. The gain must collapse under the required ablations (information-gain-zeroed, shuffled-affordance, and gamma-only), or it does not count. Until that RED clears both bars and the ablations collapse the gain, this is a Class-C design hypothesis supported by a read-only counterfactual diagnosis, NOT an achieved plateau-break. G6 stays OPEN.
What is NOT claimed in S-L2 - Metabolism
- Ceiling: that metabolism "broke the plateau" or drove the colony to stone and shelter is NOT shown. A careless reader might infer that a +135% / 2.35x tool-crafting uplift means the building plateau was solved; it was not. The most we claim is that, in one pre-registered 12 hour live RED, a standing metabolic drive produced an attributable +135% / 2.35x increase in tool-crafting and +19% in mining (Class C), while in the same run building went -14% and G4 allostasis never separated, so gate G6 (plateau-break to stone/shelter) remains OPEN.
- Fences engaged: never "created life" / "digital life" / "measurable awareness" as a claim (the viability edge is a model variable, never a felt state); never raise a claim above its source evidence class (the uplift is Class C, the diagnosis Class A, the G6 cure a Class-C design hypothesis); never let substrate or engineering imply a science gate is met; the
G5baction-severed-twin is the standing falsifier any self-maintenance or life language must first clear. - Negatives that travel with this claim (cite alongside, never strip): L2.2 (building -14%, G4 never separated, G6 OPEN, contradicted by its own first evidence) must appear next to L2.1's +135%; L2.3 (the plateau diagnosis
epistemic_starvation, NOT gamma-runaway, NOT a curriculum ceiling); L2.4 (the two fixed engine-seam negatives: no strong-Dirichlet seam without:pb_seed, and no live viability edge in the naive design). - Parked / owed: a disciplined paired RED on
build_epistemic_frontieris OWED to discharge or refute G6; the G4 allostasis gate is owed a clean separation using the existing ledgered metric; until then G6 is OPEN, not parked, and certainly not passed. - One-line honest summary a skeptic could not dispute: a standing metabolic drive measurably increased foraging and crafting in one paired live test and in that same test failed to improve building or separate allostasis, so the plateau-break it was built to achieve is still OPEN.
Falsify this
Lead falsifier (the uplift): a repeat pre-registered 12 hour live paired RED fails to reproduce the +135% / 2.35x tool-crafting uplift within its CI, OR ablating the metabolism organ does not remove the uplift, OR the test suite drops below 297/0. Any one of these would overturn L2.1. The plateau-break is discharged only the other way: a disciplined RED in which the metabolism organ (or the build_epistemic_frontier cure) alone improves building with a placed-blocks CI excluding 0 AND G4 allostasis separates, with the gain collapsing under the registered ablations. Until that happens, G6 is recorded OPEN.
Sources
- Narrative grounding (PII-redacted):
curated/strings-digest.md(Phase-2 metabolism arc, the first-12h mixed live RED, the plateau diagnosis, the two engine seams, theG5bfalsifier);curated/uni-precision-digest.md(the Heart Lab physiological mirror and the paired-bootstrap-CI evidence discipline). - Ledger rows (single source of truth):
encyclopedia/CLAIM-LEDGER.md, section L2 - rows L2.1 (uplift, Class C), L2.2 (building/G4/G6 NEGATIVE, Class C), L2.3 (epistemic_starvationdiagnostic, Class A), L2.4 (engine-seam negatives, Class A). - SIGNED consult insert:
cookbook/UNI-GPT-CONSULT-2026-06-27.md, Q3 (SIGN-WITH-CONDITIONS) - thebuild_epistemic_frontiermicro-organ and theG6_BUILD_EPI_FRONTIER_PAIRED_RED_v1paired RED; Class-C design hypothesis only, G6 stays OPEN. - Source archives (local-only, PII, not read here):
...-Strings(TMDLRG/UNI.MineCraft, "THE STRATIFIED PALIMPSEST");...-Precision.
sha256 c929117335d6dacc — of the original file, so what was ingested stays checkable.
Plain — written for this website, not the source document
The second rung of the developmental ladder hands the simulated colony an energy organ it can feel from the inside, and the chapter reports what that bought and what it did not. In the first paired live run the organ raised foraging and tool-crafting by an attributable margin. In the very same run it left building slightly worse, the allostasis gate never came apart, and the plateau the organ was built to break stayed exactly where it was. Both halves are load-bearing and neither may be quoted without the other. A read-only audit of the real colony brain files then went looking for the cause. It ruled out the two obvious mundane explanations and found instead that the drive toward information is orders of magnitude too weak for building to be worth planning for. A smallest cure has been designed and signed at the design level only. It has not been run.
Plain · written 2026-08-01 by claude-opus-5 · not yet checked by a person · about the document whose sha256 is c929117335d6dacc
Clear — written for this website, not the source document
The colony here is a simulation, a toy world and not a person. The organ shipped into it as an additive, opt-in capability absent from the default genome, so the default colony stays byte-identical, checked by a golden-fixture test over the planner. Two engine-level findings were required before the design was even valid, and both are recorded as negatives that were fixed rather than swept away. A strong prior could not be seeded the obvious way, because of the order two operations run in. And the live bridge had no viability consequence at all, so a naive design drained a belief with no effect on the world.
The positive result came from a paired live run, treatment with the organ on against a control without it. Tool-crafting and mining both rose, and the effect is attributable and removable by ablation.
It must always be cited with the negative from the same run. Building went backwards, and the allostasis gate never separated, so the load-bearing claim the program actually wanted to make is contradicted by its own first evidence and remains open. The calibrated reading is exact: metabolism is shown as a foraging and crafting driver, not as a building driver. One honesty caveat travels with the figures, since the earliest hours of the treatment arm were invalid after a crash and the clock was reset to a clean start.
The diagnosis section is the most interesting part. The colony plateaus at making a tool, and one agent hoarded pickaxes and never built. A read-only counterfactual audit on the real brain files diagnosed the cause as a starvation of information value, and was specifically able to rule out the two mundane culprits, a runaway precision term and a curriculum ceiling. The planning landscape is saturated and flat, and the information drive is orders of magnitude too weak.
The proposed cure is recorded as a design hypothesis and nothing more: a small building-specific organ carrying one hidden factor and one policy term that values expected information gain about where a block can usefully be placed. It is explicitly not a precision change and not a reward bonus. It must be earned through a paired run whose bar is registered before it starts, with its own primary verdict and required ablations, and until that run clears both bars the gate stays open.
Clear · written 2026-08-01 by claude-opus-5 · not yet checked by a person · about the document whose sha256 is c929117335d6dacc