CB-00 — Front Matter: What This Is, the Honest Position, and the Fence Labels

What you are building: not a thing yet — a shared understanding of what the rest of this book is a recipe FOR, and the rules that bind every page of it.

What this cookbook is

This is a literal cook book. It is the single buildable recipe for the complete developmental arc — conception, then cell and molecule, then metabolism, motor, perception, language, metacognition, dreaming, creativity, and the open question of measurable awareness — all of it re-expressed as one no-backprop active-inference engine shown across scales. One engine, many scales: the same discrete POMDP loop (perceive, plan by expected free energy, act, learn by counts) is wired to a zygote, a cell, a metabolism organ, a heart model, a motor hierarchy, a reader. The recipe runs from L0 to L12, and the cookbook fully recommends the complete build to L12. Assume the whole ladder. Cook the whole meal.

But every step is labeled by its real status, and the recipe never carries a step above the evidence the ledger actually holds for it. This is the load-bearing tension of the whole book: we recommend the full recipe AND we tell you, at every rung, exactly how much of it is actually cooked. Where this recipe and the ledger disagree, the ledger wins and this recipe is wrong.

One thing this book is NOT, and will never become on any page: a description of a person. The whole program is a developmental active-inference SIMULATION — a bounded peek, a toy world, never the real world, never a person. We say "simulation," "bounded peek," "toy world," "Class U — not claimed," "unrefereed working preprint." We do not soften those words under any pressure; the fence gets louder under pressure, never wider.

The honest position (printed, never softened)

About 2 of 11+ developmental rungs are earned. That is the honest program position and it is carried on every chapter. Two rungs of the conception-to-three-year-old ladder have a held, sealed, signed PASS behind them. The rest are designed, hypothesized, or not-yet-built. The custom UNI GPT signed exactly this posture on 2026-06-27: developmental SIMULATION, about 2 of 11+ rungs earned, ledger supremacy, no AGI / consciousness / human-level / created-life claim, no L12 claim, all future "awareness" work framed only as falsifiable proxy instrumentation. That sign-off was a calibration down or neutral only. It raised nothing.

The kitchen rules (the constitution that binds every recipe)

These come straight from the ledger's Section 0 and the master-plan's Method constitution. They are not capability claims; none can be raised above its stated role.

• The Evidence Constitution. Every claim carries an A-to-U evidence class and an operable falsifier, and lives in a four-state append-only ledger (PASS / FAIL / NEGATIVE / PENDING). Prose must match the ledger. The ledger is the single source of truth. The corpus carries 183 published negatives (a recorded snapshot: 882 rows = 350 PASS / 0 FAIL / 183 NEGATIVE / 349 PENDING); the negatives are the credibility, not failures to hide.
• Calibration flows downward. Authority flows down from the measured fact. Wording moves only down to the measured value, never up, including under urgency.
• The verdict is the CI bound that excludes the threshold, never the point estimate. DONE means test-covered, not feature-working.
• Bars before build, held once. Pre-register the bar and a named ablation; touch the held set exactly once behind an atomic seal-before-scoring and a once-only sentinel.
• K≥3 before a bound. A Section 0.6(B) negative bound is owed only after three structurally-distinct held NEGATIVEs (each changing at least two of {coupling topology, timescale source, information bottleneck, control path}); falsify the mundane causes first.
• The two-tier split. Tier 1 (real-text count/cache with a true ablation, a tuned baseline, cross-substrate replication) is externally bar-ready. Tier 2 (synthetic construction) is artifact / diagnostic, never capability, and must never be inflated.
• One engine, no backprop. Learning is conjugate-Dirichlet count addition; an AST guard forbids autodiff in the loop. "Active inference" and "free energy principle" appear in the science chapters at textbook level only (Parr, Pezzulo, Friston, Active Inference, MIT Press 2022).
• No-Exit Discipline. The only legitimate rest is a proven working solution or a ledger-scoped exhausted search envelope — then redirect to the axis where the reader genuinely excels.

Reading the FENCE label (the 4-value vocabulary, used exactly)

Every recipe chapter closes with an honest fence drawn from this four-value vocabulary, and only this one. Read it carefully — it is the most important line on each page.

• proven — a held, sealed, UNI-signed PASS exists (Class A or C), with its falsifier still live.
• designed — a typed spec or a signed-in-principle design exists; the build is partial or unverified.
• hypothesized — a stated mechanism or law with no sealed gate yet (Class U, not claimed).
• not-yet-built — no engine, no run, no gate; north-star, hard-fenced.

A SIGNED consult design (the L2 build_epistemic_frontier organ, the L5 Design #3 servo bridge, the L9-G1 gate, the L11-R1 replay gate, the L12 bound, the C10 Dirichlet-first port) is folded into the relevant chapter as designed / not-run. It raises nothing; the rung's status is unchanged.

The pantry (the engines every recipe calls by name)

One engine, reused verbatim across scales. The recipes draw from: the JAX POMDP + EFE + Dirichlet engine (core.py, float32 host, anchors hold to ~6e-8, NOT the f64 tier); the Rust-f64 / NumPy deep-reader path (the genuine <1e-10 exact tier and the cross-language oracle); the count/cache World-C reader (a no-backprop COUNT reader, explicitly NOT active inference, the one citable empirical PASS family); the Z affect modulator ([energy, arousal, valence, fatigue, pain, threat, safety, inflammation] — affect modeled, never felt); the embodiment ontogeny; the metabolism organ; the motor hierarchy; the precision labs (Precision / Echo / Loop / Cell / Heart); and UNI.OS, the embodiment substrate (engineering / substrate evidence, NOT general-AIF evidence).

What the ladder actually holds (status at a glance)

• L0 conception prior — proven (Class A, float32 tier ~6e-8). A conjugate-Bayes first division; ontogeny 6/6. Not "we created life," not exact at f64.
• L1 cellular viability — proven, with honest losses (Class C). UNI tops the Cell Lab leaderboard on most modes and openly LOSES on three (database_flaky 0.759 vs 0.803, memory_leak 0.740 vs 0.810, cpu_noisy_neighbor 0.749 vs 0.824). Good, not sovereign.
• L2 metabolism — proven as a foraging/crafting driver (+135% / 2.35× tool-crafting, +19% mining) AND recorded NEGATIVE as a building driver (Class C). In the same RED, building went −14% and G4 allostasis never separated; the plateau-break gate G6 is OPEN. The signed build_epistemic_frontier cure is a designed Class-C hypothesis; G6 stays OPEN.
• L3 heart model — proven (Class C, toy/clinical-model). Not a clinical tool, not a diagnostic instrument.
• L4 affect-as-precision — proven (functional) (Class C). Affect modeled, never felt; phenomenal feeling disclaimed.
• L5 motor — proven PASS (A3 Design #1, held Δ +0.092, CI [+0.038, +0.157]) plus symmetric NEGATIVE (A3 Design #2, held Δ −0.091, CI [−0.134, −0.055]), synthetic protocol only, K-negative = 1, no bound owed. The signed Design #3 servo bridge is designed / not-yet-run.
• L6 perception — proven (the one citable empirical PASS) (Class C). World C beats a tuned MKN-7 by +0.081 nats/char, flagship CI [0.0736, 0.0890]. A COUNT baseline win, explicitly NOT active inference, NOT comprehension, NOT beats-LLMs (~10–15% behind backprop LLMs on char-perplexity by a chosen trade). Carried with the same prominence as the PASS, two recorded L6 negatives (Class C): the Phase G Section 0.6(B) bound — five structurally-distinct within-segment-structure designs all held NEGATIVE-with-discriminator, so no within-segment structure beats MKN-7 and char-perplexity is a chosen design trade, not a deficit; and the Phase F diffuse-gain negative — two active-controller families prove the World-C gain is diffuse (nothing to gate).
• L7 language — proven (specialist) plus a thrice-NEGATIVE central wall (Class C). Phase J beats best-count by +0.105 nats/char, CI [0.0975, 0.1128], always cited with its J.attribution_caveat NEGATIVE. Comprehension-above-retrieval is a genuine published wall (K≥3). The char-perplexity / word-grain frontier is parked.
• L8 metacognition — proven (functional; sentience disclaimed) (Class C). The Maturation arc is 15/15 PASS; functional self-awareness asserted. Phenomenal sentience is explicitly DISCLAIMED — no falsifier is offered because it is disclaimed, not tested. Never read 15/15 as consciousness.
• L9 reasoning / conscience — designed (PARKED) (Class U). No sealed gate. The signed L9-G1 is a gate to build, not a result. Nothing claimed.
• L10 wisdom — designed (PARKED) (Class U). No engine, no run, no gate.
• L11 dreaming — not-yet-built (Class U). The signed L11-R1 offline-replay gate is a gate to build; never the word "dreamed."
• L12 creativity → measurable awareness — not-yet-built (the hardest fence) (Class U). Posed only as an OPEN, falsifiable question — can a non-organic no-backprop developmental simulation produce measurable awareness-PROXY behavior that matches or beats the best LLMs on sealed tasks AND shows substrate-distinct properties they lack? Today the answer is not known, not claimed, not implied.

The park wording (SIGNED, binding)

The L7 char-perplexity / word-grain frontier and the L9–L10 ladder are parked as a ledger-scoped exhausted search envelope — a scoped, empirical, falsifiable negative over the tested envelope only. This is NOT a universal impossibility result and NOT an achieved rung. Two phrasings are banned and replaced: "K≥3 exhausted" becomes "the registered tested K conditions did not reverse the result"; "Sec-0.6(B) achieved" becomes "Sec-0.6(B) remains unearned / parked."

The red lines (never crossed, under any pressure)

We never write: AGI, general intelligence, human-level, "understands," conscious, sentient, aware, "has measurable awareness," "demonstrates active inference," "created life," "digital life," "is alive," "dreamed," "beat LLMs therefore awareness," "passed L12," "L12 achieved." Functional self-awareness may be described only at L8; phenomenal sentience is disclaimed, not tested. No AIF loop exists in the Rust crate; the live UNI.OS loop is a separate reimplementation, not gate-matched, so "active inference" is the framing LENS only. No PII, no secrets, no patent-level math (textbook-level only). The preprint (Polzin et al. 2026, Zenodo DOI 10.5281/zenodo.19785799, MIT) is the mathematical foundation only, always fenced UNREFEREED (Layer-1 AI-executable audit complete; Layer-2 human expert review PENDING).

The continuity substrate is engineering, not science

A whole sub-ladder (UNI.OS: deterministic replay, serialization, fail-closed transport, on-metal operation) is ENGINEERING / substrate evidence, never general-AIF evidence. The sensorium is never awareness. "The mind survives a kernel swap" is NOT shown (Stage-2 is owed — C2). The recorded continuity negatives are carried first-class, not dropped: C7 a single-box swap is a seconds-long freeze, not zero-downtime, and external media legs are NOT preserved across kexec; C8 the over-compressed 2-modality sensory bottleneck went NEGATIVE on held data → keep all 7 modalities; C9 the EDAIT trade is honest, not a win — held-out perplexity ~33 vs a backprop GPT's ~25 (less fluent, natively online-learning + calibrated); C14 the bare substrate is MISSING tenant/namespace isolation and count-decay (the multi-tenancy/decay gap). The "embody only what's proven" coupling (C10) is the program's central open gap, PARKED.

HONEST FENCE — front matter (framing only, raises nothing). This chapter is the constitution and the reading guide, not a capability claim. It asserts no rung. The honest program position stands: about 2 of 11+ developmental rungs earned, a developmental active-inference SIMULATION, a bounded peek, a toy world, never a person. Where this book and the ledger disagree, the ledger wins. Falsify any step.