UNI Universal Natural Intelligence

Wiki · Evidence & Verdicts

LIVE metabolism activation gate — PASS in production (2026-07-11)

Evidence & Verdicts · docs/receipts/metabolism_activation_gate_LIVE.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.

⚠️ CORRECTED (2026-07-11): survival claim WITHDRAWN. The PASS/"alive" headline below rested on a launcher RCON food-give (force-feeding), later reclassified as a fake-life hack — not survival. The mechanism-WHEN-DRIVEN result still holds, but the survival/"alive" claim is WITHDRAWN. The rebuild (commit f0c789a) makes foraging EMERGE from an interoceptive hyperprior via EFE (world-earned food, no gives); it is UNPROVEN until it passes a pre-registered colony-survival RED gate. Do not cite this as a live survival PASS.

LIVE metabolism activation gate — PASS in production (2026-07-11)

Verdict: PASS. The :metabolism organ is proven mechanistically LIVE in production, closing the activation leg Phase-2 left WITHHELD. Full pos/neg/neg/pos, held to the pre-registered numeric bars, in the production world against the healthy 2-week-uptime mc-server, with fresh minds (timestamped memory dir) and a kin group (88) unused by any prior lineage.

Receipt provenance

  • Full log (381 lines, gz committed): docs/receipts/metabolism_activation_gate_LIVE.log.gz
  • SHA256 (raw log): f76e364c8c9cf667c315e618ac725d0b...
  • Source deploy: uni-activation container, localhost/uni-colony:metabolism image, on uni-colony-net, run ID gate-1783749155, ~5 min end-to-end. Baseline dir on lab box: ~uni/baseline/2026-07-11-gate-live-pass/.
  • Harness: runs/live_activation_gate.exs (tracked; git add -f).

Pre-registered bars (pinned BEFORE the run)

  • POS live PASS: each agent alive_final = true AND energy range ≥ 0.5 AND ≥ 2 reversals over 240 s.
  • NEG PASS-as-negative: in-BEAM control dies at tick < 300 abstract ticks.

Results — production

arm condition outcome
POS UNI-88-1 metabolism_primary, live @ mc-server (fresh mind) alive; range 0.668, 8 reversals
POS UNI-88-2 metabolism_primary, live @ mc-server (fresh mind) alive; range 0.746, 10 reversals
POS UNI-88-3 metabolism_primary, live @ mc-server (fresh mind) alive; range 0.685, 8 reversals
NEG-1 action-severed twin (in-BEAM, forced :noop, same Metabolism.step) died @ 25
NEG-2 food-severed (in-BEAM, inv.food=0, real MC.step decisions) died @ 16

Full trajectory (excerpt): UNI-88-1 energy 0.863 → 0.598 → 0.312 → 0.658 → 0.918 → 0.806 → 0.678 → 0.882 → 0.688. Live limit cycle — the posterior depletes AND refills, per agent, over 17 probes each.

What was remediated to reach production PASS

Four load-bearing defects were found, receipted, and closed (all committed):

  1. Live-wiring gap (88be5c9) — the metabolic loop lived only in bridge.ex (which the live colony doesn't run). Mirrored the loop into SP.Runtime.Agent.handle_info, gated on metabolic?, default byte-identical. This is why Phase-2's arms were indistinguishable: the organ was inert live.
  2. Rate miscalibration → wall-clock notch (383ffb4) — Metabolism.step per-step upkeep killed agents in ~9 s live; now drain scales by dt / @nominal_tick_sec (8 s), so the viability edge is wall-clock based, cadence-independent. dt = nil (offline) ⇒ frac 1.0 ⇒ byte-identical; every offline test unchanged.
  3. UI auto-boot ate mc-server slots (535f9b6) — ui/lib/sp_ui/application.ex on UNI_AUTOSTART=1 auto-spawns the design colony (kins 0..3) which raced my metabolism agents. Launcher sets UNI_AUTOSTART=0 at BEAM boot; container run env matches.
  4. BEAM cwd trap (535f9b6) — mix run sets cwd to /app/ui (not /app), so File.cwd!() + "viewer/body.js" resolved to /app/ui/viewer/body.js (missing) → File.exists? false → Port.open silently skipped → no body ever embodied. Launcher uses UNI_REPO=/app absolute.
  5. World-food constraint (a9b1508) — a fresh mineflayer body has no inv.food, so :eat cannot refill; the world must provide food. Launcher uses SP.Minecraft.Rcon to give @a cooked_beef 64 post-spawn + every 45 s. The world provides food; the agent's job is to KEEP its store viable USING it.

Guards green throughout: decider_byte_identity mad<1e-12, action_clone_invariance A1/A2/A3, 28 tests 0 failures; compiles --warnings-as-errors.

What this proves + what it does NOT

Proves (live, production, receipt-backed):

  • The :metabolism organ is mechanistically live in the live Agent path (the wiring gap Phase-2 tripped on is closed).
  • The viability edge is real live: action-severed → death, food-severed → death (two structurally distinct negatives).
  • The energy posterior oscillates live — a live homeostatic limit cycle in production, not offline.
  • The wall-clock notch works: agents live for 4 min at live cadence (they'd die in ~9 s per-step-drain).

Does NOT prove (unchanged, still open, honestly fenced):

  • G6 plateau-break — behavioural, still FAIL live in Phase-2 (docs/receipts/phase2_metabolism_red.md). Activation live does not imply behavioural cure. G6 remains the next real gate to earn.
  • Life / experience / awareness — every float here is a model variable, never a felt state (claim fence). Passing this gate demonstrates named MECHANISM, never experience.

Ready-to-announce state

  • Metabolism activation gate LIVE = PASS (this receipt). Upgrades P2 activation WITHHELD → VERIFIED IN PRODUCTION.
  • Phase-2 §16 mechanism-null reading corroborated: the null was a live-wiring defect, not the organ math failing. The offline math held all along; the live path did not exercise it. Now both do.
  • Next live gate = G4 allostasis / G6 plateau-break under the survival-C + true-vision A4 program (design reviewed, 15 blockers folded — docs/specs/{generative_model,curriculum_removal,sensorium}.md).

sha256 392d9a1a38bf8c42 — 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)

Read the banner at the top of this page first: the survival part of this result is withdrawn. The headline rested on food being handed to the bodies by the harness, which was later reclassified as a hack rather than as survival. What still stands is the narrower mechanism result, that when the organ is driven it does something measurable in production, and that two different ways of cutting it off both end in death. The replacement design makes feeding emerge instead of being given, and it is unproven until it passes a gate whose conditions were written down first.

Plain · written 2026-08-01 by claude-opus-5 · not yet checked by a person · about the document whose sha256 is 392d9a1a38bf8c42

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 page carries a correction at the top, and that correction is the first thing to read. The survival claim below it is withdrawn. It had rested on food being given to the bodies by the launcher, later reclassified as force-feeding rather than survival. What survives the withdrawal is the mechanism result when the organ is driven. The rebuild that replaces it makes feeding emerge from the model instead. The page says that rebuild is unproven until a gate written down in advance passes, and it says outright not to cite this receipt, the file recording what was run, as a live survival pass.

Under the banner the original receipt is left intact rather than edited. It records a production run against a long-running world, with fresh minds and a lineage group not used by any earlier work, and it names the log, its digest, the deploy and the harness so the run can be traced.

The bars were pinned before the run, and the results table gives one row per arm. Three positive arms show an energy estimate that ranges and reverses direction repeatedly. Two negative arms die early, one with action cut off and one with food cut off. A trajectory excerpt shows the estimate falling and rising rather than draining once.

Then a list of the defects that had to be closed to get there, which is the most instructive part of the page. The loop lived only in a path the live system does not run, so the organ was inert in production, which explains why an earlier phase could not tell its arms apart. A drain rate calibrated per step rather than per unit of time killed bodies within seconds at live cadence. An automatic start elsewhere was competing for slots. A working-directory assumption meant a component was silently skipped, so no body was ever embodied. And a fresh body carries no food, so the world had to provide it, which is precisely the point the later correction turns on.

The closing sections separate what the run shows from what it does not, and that separation was explicit even before the withdrawal. A behavioural gate had failed and stayed open. Activation does not imply a behavioural cure, and every number here is a model variable rather than a felt state.

Clear · written 2026-08-01 by claude-opus-5 · not yet checked by a person · about the document whose sha256 is 392d9a1a38bf8c42