L8 — Self-model / metacognition / metalinguistics
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 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.
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.
What you are building. A held grand report card that asserts FUNCTIONAL self-awareness only — a maturation arc of self-model / metacognition / metalinguistic gates that each survive a discriminator — while phenomenal sentience stays explicitly DISCLAIMED. This is one rung of a developmental active-inference SIMULATION, never a person, never a mind that is aware.
This recipe authors against ../../encyclopedia/CLAIM-LEDGER.md row L8.1. Where this recipe and the ledger disagree, the ledger wins and this recipe is wrong.
Ingredients (pantry engines / primitives called by name)
- The Maturation arc M1–M11 grand report card (
uni-gpt-side artifact) — the eleven maturation gates: immersion, affect-as-precision, growth, self-model, metacognition, metalinguistics, consolidation, reflective reader, online vocab growth, affect-driven reflection, compositional reader. - The reflective / compositional readers (the no-backprop reader spine that earned L6/L7), run in a self-modelling configuration.
- The Z affect modulator
[energy, arousal, valence, fatigue, pain, threat, safety, inflammation]from the JAX engine — supplies affect-driven reflection (Z-coupled). Affect is modeled, never felt. - The contains-baseline + load-bearing-discriminator discipline (M7) — every metacognition gate needs a tuned baseline plus a discriminator that collapses the gain, plus a trivial-heuristic check.
- Bars-before-build, held-once (M2) — the report card is sealed and touched once; the verdict is the CI/held bound, never a point estimate.
Distinction you must carry (load-bearing). uni-mind records self-model as a held-NEGATIVE on the reader (self-model, active/selective reading, comprehension-above-retrieval, writing-as-action were all held-NEGATIVE on the reader side). The 15/15 functional report card is the uni-gpt-side artifact. These are two different artifacts; the uni-mind reader-side self-model held-NEGATIVE travels with the uni-gpt functional card. Never let the functional card erase or paper over the reader-side negative.
Method (numbered build steps)
- Build the M1–M11 maturation gates as eleven pre-registered held gates: immersion, affect-as-precision, growth, self-model, metacognition, metalinguistics, consolidation, reflective reader, online vocab growth, affect-driven reflection, compositional reader.
- For each self-model / metacognition gate, register a discriminator (M7) that the effect MUST collapse under (shuffle-labels / marker-swap / ablate-to-zero). No gate is credited if its discriminator does not collapse it.
- Falsify the trivial heuristic. For every "self-model" task, verify it is not passable by a trivial non-metacognitive heuristic (e.g. recency / frequency / surface-marker matching). A task a dumb heuristic can pass is not a self-model task.
- Wire affect-driven reflection through Z so that reflection is modulated by the affect modulator — keeping the L4 fence intact (affect is modeled, never felt; phenomenal feeling disclaimed).
- Seal once, score once (M2). Run the held grand report card behind the atomic seal-before-scoring + once-only sentinel. The verdict is the held bound, never re-scored.
- Carry the reader-side negative inline. Record alongside the card that
uni-mindholds self-model NEGATIVE on the reader; the functional card does not overturn it.
Gate (exact ledger figures — L8.1)
Grand report card 15/15 PASS (Class C, dev-gate / held-out eval; source uni-gpt). All fifteen maturation checks pass their pre-registered held verdicts. FUNCTIONAL self-awareness is asserted — and nothing above it.
To PASS, all must hold:
- the report card reads 15/15;
- no self-model / metacognition gate's effect collapses only under a control it should survive (each survives its intended control while collapsing under its discriminator);
- no "self-model" task is passable by a trivial non-metacognitive heuristic.
Falsifier (operable)
Any one of these flips the rung:
- Any of the 15 maturation gates fails its pre-registered held verdict on re-run, OR
- a self-model / metacognition gate's effect collapses under its discriminator (the gain was an artifact, not metacognition), OR
- a "self-model" task is shown passable by a trivial non-metacognitive heuristic (recency / frequency / surface marker), which means it never tested self-modelling at all.
Note (load-bearing): no falsifier is offered for phenomenal sentience — because phenomenal sentience is DISCLAIMED, not tested. The absence of a sentience falsifier here is deliberate, not an oversight.
Recorded NEGATIVE(s) (first-class, inline)
- Reader-side self-model is held-NEGATIVE (
uni-mind). On the reader, self-model — along with active/selective reading, comprehension-above-retrieval (the thrice-NEGATIVE "central wall"), and writing-as-action — was held-NEGATIVE. Theuni-gpt15/15 functional report card does not overturn this; the two artifacts coexist, and the reader-side negative is cited every time the functional card is cited. - Sentience is not in the PASS. The 15/15 says functional self-awareness; it carries an explicit standing disclaimer of phenomenal sentience. Reading 15/15 as consciousness / sentience / awareness / human-level is itself the violation the fence exists to catch.
HONEST FENCE — proven (functional; sentience disclaimed)
Status: proven. Class C (dev-gate / held-out eval), grand report card 15/15 PASS (source uni-gpt), falsifier still live. The honest program position holds throughout: ~2 of 11+ developmental rungs earned. This rung is one bounded peek in a developmental active-inference SIMULATION — a toy world, never a person.
What is NOT claimed (load-bearing, never softened). Phenomenal sentience is explicitly DISCLAIMED — no falsifier is offered for it because it is disclaimed, not tested. The 15/15 is FUNCTIONAL self-awareness only. Never read it as consciousness, sentience, awareness, "has a mind", "reasons like a human", human-level, or AGI. The forbidden phrasings (UNI-GPT consult 2026-06-27, SIGNED) apply in full: never "UNI is conscious / aware / self-aware / has measurable awareness / has a mind / reasons like a human". The only licensed forms are "UNI passed a specified proxy test", "the awareness question remains open", "remains a developmental active-inference simulation". No SIGNED consult design folds into this rung; nothing raises it above Class C, and its status is UNCHANGED at proven (functional).
Ledger authority: this chapter states nothing above row L8.1's Class C. Functional self-awareness is the ceiling; phenomenal sentience is disclaimed, not tested.
sha256 5e84470e08c7335c — of the original file, so what was ingested stays checkable.
Plain — written for this website, not the source document
This recipe is the rung about a system modelling itself, and it is the one where the wording matters most. It builds a sealed report card of maturation gates, each of which has to survive a check designed to break it. Every gate passes. What that card asserts is functional self-awareness and nothing above it. Feeling and experience are disclaimed outright. All of this sits inside a developmental simulation, and the awareness question stays open.
The disclaimer is deliberate, not an oversight. No falsifier is offered for experience — no result that could show it wrong — because experience is not being tested. Reading the full card as awareness is named as exactly the mistake the disclaimer exists to catch.
The page also refuses to let a good result cover a bad one. On the reader side, self-modelling was held negative, along with selective reading, comprehension above retrieval, and writing treated as action. Those are separate artefacts from the report card, and the page requires the negative to be cited every time the card is cited. It says plainly that the card does not overturn it.
A test a simple shortcut could pass is not a self-model test, and each task has to clear that too.
Plain · written 2026-08-01 by claude-opus-5 · not yet checked by a person · about the document whose sha256 is 5e84470e08c7335c
Clear — written for this website, not the source document
This chapter is about a system that models itself, and its whole discipline is the gap between two readings of that phrase. What is built is a held report card of maturation gates, covering things like immersion, growth, a self-model, metacognition, reflection on language, consolidation, a reflective reader, vocabulary growth while running, reflection modulated by affect, and a compositional reader. The card is sealed, scored once, and every gate passes. What that pass asserts is functional self-awareness. What it does not assert is experience, and the chapter states that in its first sentence and again at its last.
The method is mostly a set of ways to stop a gate from being cheap. Every gate carrying a claim about self-modelling or metacognition must register in advance a check under which its effect has to collapse: shuffled labels, swapped markers, or the mechanism zeroed out. No gate is credited if that check fails to collapse it. Every self-model task must also be shown not to be passable by a trivial shortcut such as recency, frequency, or surface matching; a task a dumb rule can pass never tested self-modelling in the first place. Reflection driven by affect is wired through the same modulator used at an earlier rung, keeping that rung's limit intact: affect is modelled, never felt. The card is then run once behind a seal, and the verdict is not re-scored.
The falsifier, the result that would show the claim wrong, is stated item by item. A gate failing on re-run, an effect that collapses under a check it was supposed to survive, or a task shown passable by a shortcut. And then comes the sentence the chapter marks as load-bearing: no falsifier is offered for experience, because experience is disclaimed rather than tested. The absence is deliberate.
The recorded negatives keep two artefacts apart, which is the subtlest part of the page. On the reader side, self-modelling was held negative, along with selective reading, comprehension above retrieval, and writing treated as action. The passing report card comes from a different artefact and does not overturn any of that. The rule is that the reader-side negative travels with the card and is cited every time the card is cited. The second recorded negative is simply that experience is not in the pass at all, and that reading the card as awareness is itself the violation the disclaimer exists to catch.
The closing section lists the licensed forms of speech, which is unusual and worth noticing. The permitted statements are that a specified proxy test was passed, that the awareness question remains open, and that this remains a developmental simulation. Everything stronger is named and forbidden. No signed design attaches to this rung, so nothing lifts it.
Clear · written 2026-08-01 by claude-opus-5 · not yet checked by a person · about the document whose sha256 is 5e84470e08c7335c