MASTER-PLAN.md — The UNI Encyclopedia (table of contents + authoring rules)

What this is. The authoring blueprint for the UNI ENCYCLOPEDIA — the public reference work about the UNI program (a developmental active-inference SIMULATION, never a person, never a mind). This file fixes the front matter (the Evidence Constitution, the A–U rubric, the standing fences and red-line list, the "what is NOT claimed" template), the body structure (the four live-hub wings × the L0–L12 science spine × the Method wing × the Track-A appendix), a one-page authoring spec per chapter, and the ordered Session-3 authoring queue.

Authority. This plan authors against two files and never ahead of them:

• the single source of truth — CLAIM-LEDGER.md (882-row append-only ledger, every claim carded A–U with a falsifier);
• the Session-1 synthesis — ../curated/00-INDEX.md (the 11 per-archive digests, the wing map, the Track-A/Track-B split).

If any encyclopedia prose disagrees with CLAIM-LEDGER.md, the ledger wins and the prose is wrong. No chapter may state a claim above the evidence class recorded in the ledger, omit a recorded falsifier, or hide a recorded negative.

Honest program position (print it, do not soften it): ~2 of 11+ developmental rungs earned.

---

PART I — FRONT MATTER (the constitution; read before authoring any chapter)

FM-1. The Evidence Constitution

Every entry in the encyclopedia is governed by six rules carried verbatim from the ledger:

1. A falsifier per claim. No claim ships without a stated condition that would prove it false. A claim with no falsifier is not a claim — it is marketing, and it is forbidden.
2. The append-only ledger is the single source of truth. Four states only: PASS / FAIL / NEGATIVE / PENDING. Corrections are forward-only (supersede with lineage), never silent edits. Prose cites the ledger; if they disagree, the ledger wins.
3. Calibration only moves DOWN. Authority flows downward from the measured fact. Wording is calibrated down to the measured value, never up — including under urgency. The fence gets louder under pressure, not wider.
4. Verdict = the CI bound that excludes the threshold, never the point estimate.
5. DONE = test-covered (Class E/D), not feature-working (Class A). A passing test does not satisfy a criterion that demands a runtime observation.
6. Negatives are content. A partial / a negative / a "most pieces don't help" decomposition is a measurement that the design is incomplete, not an exit and not a failure to hide. The negatives are the credibility. Provenance fence (do not feature as a re-countable headline). The figure 882 rows = 350 PASS / 0 FAIL / 183 NEGATIVE / 349 PENDING is a recorded ledger snapshot, not a count reconstructable from this document: the ledger derives from a deduplicated merge of 615 extracted claims (CLAIM-LEDGER.md line 5), the snapshot enumerates 882 rows, and the carded body shows only ~140 distinct rows. A skeptic counting the published rows cannot independently derive 183. Cite the snapshot as a snapshot (provenance-flagged), and headline only the negatives actually enumerated in the ledger body — never the bare "183 published negatives" as a credibility number until the count is reconstructable, per the calibration-down rule.

No-Exit Discipline. The only legitimate rest is a proven working solution or a published, exhausted, falsifiable bound — then redirect to the axis where the reader genuinely excels. Never go silent on an open gate. A park is not discharged until its sign-to-park lands. Falsify the mundane causes (L2 metabolism / L7 retrieval-and-recency) before calling any NEGATIVE.

FM-2. The evidence-class rubric (A..U)

The class is the ceiling. A chapter may card a claim at or below its ledger class, never above.

Class	Name	Means	Authoring rule
A	machine-exact anchor / live observation	observed at runtime, or exact to the float tier (machine-checkable)	the strongest class; still fence the interpretation (an anchor is not a capability).
B	mechanism + operator observation	a mechanism shown working under grounded inspection, short of a held eval	never present a B as a held-out PASS.
C	dev-gate / held-out eval	a pre-registered, sealed, held-once gate with a CI verdict	this is the empirical-PASS tier; cite the CI bound, not the point estimate.
E	test-covered	tests pass (and D = test-covered design)	DONE ≠ working. A Class-E pass never satisfies a Class-A criterion.
F	doc / prior-claim	inheritable from a document or earlier claim	must be re-verified before it is leaned on; mark inherited.
U	claimed-but-unproven	asserted but not earned (e.g. "human-level"; the parked frontier)	"Class U — not claimed" is itself a standing fence. Class-U content is described as not-yet-built / parked, never as capability.
method	definitional / governance pattern	a reusable discipline, not an empirical claim	proven and reusable, but cannot be raised into a capability claim.

Class-authority ordering (when sources conflict): Class-B (tool state) overrides Class-G (own narrative); Class-A (observed at runtime) overrides Class-E (test passes). Mark [CONFLICT:unresolved] and resolve with a direct read. (The A–F subset — A live · C code/static · E test · F doc/inheritable — is the per-ticket provenance taxonomy; the full A–U taxonomy adds B mechanism, U not-claimed, and method.)

Exactness-tier honesty (load-bearing). Never label a float32 anchor <1e-10 EXACT. The JAX core runs float32, so its anchors hold only to ~6e-8 (~1e-6 single-step filter); the genuine <1e-10 tier lives only in the NumPy / Rust-f64 path (the cross-language Rust-f64 == Python bridge). A genome docstring that claimed <1e-10 was caught as an overclaim and corrected. Card every anchor at its real tier.

FM-3. Standing fences + the red-line list

Standing program framing — say it this way, every chapter: "developmental SIMULATION", "bounded peek", "toy world, not the real world", "Class U — not claimed", "unrefereed working preprint", "same math, many scales", "count baselines / LOOP not LEAP."

THE RED LINES — never claim, in any chapter, in any public copy, under any pressure:

1. Never AGI / general intelligence / human-level / "talks & learns like a human" / "understands."
2. Never consciousness / sentience / aware. (Functional self-awareness may be described at L8; phenomenal sentience is explicitly DISCLAIMED — no falsifier is offered because it is disclaimed, not tested.)
3. Never "active inference demonstrated." No AIF loop exists in the Rust crate; the live UNI.OS loop is a separate reimplementation, not gate-matched. "Active inference" is the framing lens, never a demonstrated result.
4. Never "created life" / "digital life" / "measurable awareness" as a CLAIM (north-star framing only; posed as an open falsifiable question, never an answer).
5. Never "beats LLMs." World C is a COUNT baseline; the program is ~10–15% behind backprop LLMs on char-perplexity by a chosen design trade. EDAIT is an honest fluency-for-calibration trade (~33 vs ~25 ppl).
6. Never inflate the Tier-2 synthetic-construction track into capability (it was audited as artifact/diagnostic — hardcoded-literal "exactness", scoring-artifact deltas, no AIF loop — and fixed).
7. Never raise a claim above its source evidence class.
8. Never let substrate / continuity engineering imply a science gate is met (it is engineering evidence, not general-AIF evidence).
9. Never publish "K≥3 exhausted" / "§0.6(B) achieved" phrasings until UNI signs the relevant park.
10. No PII. No secrets/tokens/internal channel handles (a link may exist; the handle is never featured). No patent-level UNI math — textbook-level framing only (consult the private UNI Active-Inference Guide GPT for science; never publish it).
11. The preprint (Polzin et al. 2026, Zenodo DOI 10.5281/zenodo.19785799, MIT) is cited as the mathematical foundation only and always fenced unrefereed (Layer-1 AI-executable audit complete; Layer-2 human expert review PENDING) — never as proof that active inference is the correct theory. PII note (not a red-line breach): the named authorship "Polzin et al." is already public under the cited DOI, so reproducing the author form is an intended public attribution, not incidental PII leakage — the no-PII red line (item 10) is not violated by this sanctioned citation. (If a public encyclopedia copy must not bind the program to the owner's surname, fence the citation to the DOI + title only; absent that instruction the published authorship stands as intended attribution.)
12. Vocabulary-leak guard (HARD, test-enforced): never externalize active-inference / EFE / free-energy framework names or print internal channel handles in public copy. Public copy says "count baselines", "prediction-error", "LOOP not LEAP."

UNI-GPT consult (2026-06-27 — SIGNED): forbidden-phrasings red-line list (extends FM-3, calibration-down). The consult (Q7b) banned the following phrasings OUTRIGHT — never publish any of them, in any chapter, in any public copy, under any pressure. Each is paired with the only honest replacement. (Cross-reference: ../cookbook/UNI-GPT-CONSULT-2026-06-27.md, Q7.) This list is ADDITIVE to red lines 1–12 above and never relaxes any of them.

• Banned outright: "UNI is conscious / aware / self-aware / has measurable awareness / has a mind / has a conscience / reasons like a human / is human-level / is AGI / created life / created non-organic life / is alive / is a synthetic organism / dreamed / is creative in the human sense / demonstrates active inference / proves the FEP creates minds / proves plants and UNI are the same kind of life / beat consciousness tests / beat LLMs therefore awareness / passed L12 / L12 achieved."
• Replace with (the only licensed forms): "UNI passed a specified proxy test / improved a registered downstream metric / matched-or-exceeded the registered LLM baseline on this sealed task / showed a substrate-distinct effect under this ablation / remains a developmental active-inference simulation / the awareness question remains open."
• L7/L9–L10 frontier (Q1, signed): replace any "published exhausted bound" framing with "ledger-scoped exhausted search envelope" — a ledger-scoped, implementation-scoped, data-split-scoped negative result over the tested envelope only, NOT a universal impossibility result and NOT an achieved capability rung. Specific banned→replacement pairs (Q1): "K≥3 exhausted" → "The registered tested K conditions did not reverse the result."; "Sec-0.6(B) achieved" → "Sec-0.6(B) remains unearned / parked pending a future result that beats the registered discourse prior under ledgered evaluation."; "UNI demonstrates language/metacognition at L9–L10" → "UNI records a negative L9–L10 frontier test under the no-backprop developmental simulation program."; "This proves no no-backprop model can beat MKN-7" → "No tested no-backprop variant in the registered envelope beat MKN-7 on the chosen char-ppl metric."

UNI-GPT consult (2026-06-27 — SIGNED): awareness-proxy proposal-entry checklist (the L12 gate). Any FUTURE L12 "awareness" work must pass ALL ten conditions below before such a gate may even be PROPOSED (Q7c). This keeps any awareness gate falsifiable rather than unfalsifiable. It is a proposal-entry gate, not a pass-claim, and it raises NOTHING: L12 stays NOT-YET-BUILT and Class-U-not-claimed regardless.

1. Proxy-only target — title must say "…-proxy"; no bare "awareness gate."
2. Best-LLM baseline, frozen at proposal time — name models/versions/prompts/tools/context/temperature/ scoring; UNI must match-or-beat the BEST, not a hobbled baseline.
3. Substrate-distinct requirement — ≥1 property LLM prompting shouldn't get for free (no-backprop online adaptation with ledgered state changes; causal self-monitoring that improves later policy; global cross-module availability; counterfactual intervention on internal observables; durable trace/report consistency).
4. Two-part pass bar (both required): A. UNI ≥ best LLM on the sealed shared task; B. UNI shows a substrate-distinct effect whose gain collapses under the registered causal ablation. Either alone is insufficient.
5. Computed internal observables, not hardcoded reports — self-report/uncertainty/contradiction-detection computed from trace/model state; scripted confession or label lookup fails.
6. Load-bearing ablation — the effect must collapse when the mechanism is removed/shuffled/made uninformative.
7. No hidden LLM substitution — prove no LLM / label oracle / retrieval leak / backprop-trained hidden controller is doing the proxy work.
8. Sealed held-out tasks + paired uncertainty — sealed before execution; paired controls; CIs excluding 0 on the registered effect.
9. Negatives are first-class — predefine fail/partial/regression/non-informative/"do not count"; a null publishes as a bound, never rewritten into progress.
10. Ledger supremacy — the ledger holds final status; prose/recipe/demo/narrative that conflicts loses.

FM-4. The "what is NOT claimed" entry template

Every chapter carries this block; it is not optional and it is not buried at the end. It is a first-class section with the same weight as the claim. Template:

### What is NOT claimed in <chapter>
- Ceiling: <the single strongest thing a careless reader might infer> is NOT shown. The most we claim
  is <the exact, calibrated, in-class statement>.
- Fences engaged: <which red lines from FM-3 apply here, named explicitly>.
- Negatives that travel with this claim (cite alongside, never strip):
  <ledger NEGATIVE rows that MUST appear next to the PASS, e.g. J.attribution_caveat with Phase J>.
- Parked / owed: <sign-to-park owed? Class-A observation owed? name it>.
- One-line honest summary a skeptic could not dispute.

Standing pairings the template enforces (citing the PASS without the negative is an overclaim):

• Phase J +0.105 always with J.attribution_caveat (overall NLL worsens; specialist gain).
• World C +0.081 always with "COUNT baseline, not AIF, not comprehension, not beats-LLMs."
• L2 metabolism +135% tool-crafting always with building −14% / G4 never separated / G6 OPEN.
• L8 15/15 PASS always with "phenomenal sentience explicitly DISCLAIMED" and the uni-mind held-NEGATIVE self-model-on-the-reader (the uni-gpt functional card and the reader-side self-model NEGATIVE are in tension and must be co-cited).
• Continuity Stage-1 (mind survived a process restart bit-for-bit) always with Stage-2 OWED (mind-tick continuity across a kernel swap NOT shown).

---

PART II — BODY STRUCTURE

The encyclopedia has four mirror wings (the live public hub), with the L0–L12 developmental ladder as the science spine inside the Science wing, a Method wing (the constitution as content), and a Track-A appendix (the marketing organism). Order of presentation:

• Wing E — EVIDENCE (the ledger made public: classes, falsifiers, the calibration ledger, "Falsify this").
• Wing S — SCIENCE (UNI) — the L0–L12 spine + the continuity sub-ladder + the same-math-many-scales labs.
• Wing N — MEANING — the textbook-level lens (FEP, prediction-error maps), kept honest and unfalsifiability-aware.
• Wing M — MOVEMENT — the honest-science posture: negatives-as-pitch, "help us independently verify," open falsification.
• Wing μ — METHOD (the constitution) — the reusable disciplines as their own reference section.
• Appendix TA — the Track-A marketing organism — the publishing/engagement engine, fenced as engineering+method, not science.

Each chapter below carries the full authoring spec: title · scope · ledger claims covered · evidence classes present · falsifiers · fenced / not-claimed · source pointers (digest + archive).

FM-4 is reproduced verbatim, not summarized (applies to EVERY chapter — E, S, N, M, μ, and TA alike). The per-chapter "Fenced / not-claimed." line below is a compressed pointer, not a substitute. In the authored chapter, the full FM-4 five-field block — Ceiling · Fences engaged · Negatives that travel with this claim · Parked / owed · one-line skeptic summary — is written out in full and carries the same weight as the claim (FM-4, line above). This is mandatory for the WING-N, WING-M, μ, and TA chapters too, whose specs below compress the block into prose: prose compression in this plan does not discharge the "same weight as the claim" requirement; the authored chapter must instantiate all five fields explicitly.

---

WING E — EVIDENCE (the ledger, made public)

E0 — How to read this work (the Evidence Constitution, public form)

• Scope. The reader's onboarding: A–U classes, falsifier-per-claim, the four-state append-only ledger, CI-as-verdict, DONE≠working, "negatives are content" (the 183-NEGATIVE figure is a provenance-flagged ledger snapshot, not a count re-derivable from the carded rows — see FM-1). The chaptered Evidence Explorer with a calibration ledger and a "Falsify this" close.
• Ledger claims. M1, M2, M3, M9, M30; Constitution §0.
• Classes present. method (with E/A worked examples).
• Falsifiers. A claim asserted ahead of its ledger; wording calibrated UP; a verdict edited rather than superseded; a reproduced:true that is a literal not validator-derived.
• Fenced / not-claimed. This is a method chapter — it asserts no capability. Do not let "we have a constitution" read as "we have proven the science."
• Sources. uni-gpt-digest.md, ideation-explorer-digest.md, 00-INDEX.md / archives …-UNI-GPT, …-SolutionWright-IdeationExplorer.

E1 — The calibration ledger & "Falsify this"

• Scope. The live public artifact: the 882/350/183/349 row snapshot (carried as a provenance-flagged ledger snapshot per FM-1 — the 183 NEGATIVE count is not re-derivable from the carded rows, so present the snapshot as a snapshot and surface the negatives the ledger body actually enumerates), the per-claim "Falsify this" close, the down-only calibration record (the os-cycles 39–53 audit that moved 6 headline claims DOWN: "TWO boxes"→ONE; "mind survives a patch"→infra-only; "5 stacks"→"4 distinct stacks/5 runs").
• Ledger claims. §0 Negatives-are-content; §2 audit-layer note; M1, M8, M29.
• Classes present. method / A (the calibrated figures are the observed ones).
• Falsifiers. A headline figure shown carrying the inflated value rather than the calibrated one; the negative count misstated; a "Falsify this" entry with no operable falsifier.
• Fenced / not-claimed. Carry the calibrated figures, never the inflated ones; overstatements lived in headlines, not in fabrication — say exactly that.
• Sources. uni-os-digest.md, uni-mind-digest.md / archives …-UNI-OS, …-uni-mind.

---

WING S — SCIENCE (UNI): the L0–L12 developmental ladder (the spine)

Naming fence (every L-chapter). The ladder is the HUMAN-HGM-001 developmental design (conception → speaking 3-year-old: 11 levels + the global Z affect modulator). It is a no-backprop, nested-Markov-blanket developmental SIMULATION, never a person. One engine, shown across scales. Honest position printed on the spine page: ~2 of 11+ rungs earned.

S-L0 — Molecular / genome → zygote (conception prior) · PROVEN (float32 tier)

• Scope. Zygote first-division as exact discrete Bayes (conjugate update); recombine/seed_zygote, the no-backprop guard, ontogeny 6/6, Embodiment Rungs 1–2 GREEN.
• Ledger claims. L0.1.
• Classes present. A (machine-exact anchor, float32 tier).
• Falsifiers. test_embodiment_ontogeny.py drops below 6/6; the first-division posterior diverges from closed-form discrete-Bayes beyond the float32 tier; cell-division identity embedding exceeds 1e-9; the no-backprop guard trips.
• Fenced / not-claimed. NOT "we created life", NOT a "conscious baby", NOT exact at f64 — a float32-tier SIMULATION of a conjugate Bayesian first division. Never card a float32 anchor at the f64 tier.
• Sources. uni-gpt-digest.md, uni-mind-digest.md / archives …-UNI-GPT, …-uni-mind.

S-L1 — Cellular / autopoietic viability & homeostasis · PROVEN (with honest losses)

• Scope. The Cell Lab open pre-registered falsification benchmark (216-state service cell, observation-only controllers, RecoveryScore, bootstrap 95% CI, 8 honesty fences + framing_guard). UNI tops the leaderboard on most modes — and honestly loses on three.
• Ledger claims. L1.1 (PASS); L1.2 (NEGATIVE — database_flaky 0.803 vs 0.759, memory_leak 0.810 vs 0.740, cpu_noisy_neighbor 0.824 vs 0.749; UNI-vs-random not significant on the last).
• Classes present. C.
• Falsifiers. RecoveryScore CI fails to separate from controls where a win is claimed; a fence / framing_guard fails; the bootstrap CI is shown miscomputed. (Negative falsifier: the recorded losses fail to replicate.)
• Fenced / not-claimed. NOT "sovereign" — "good but not sovereign." The three losses are first-class published content shown at the top of the live leaderboard. Cellular/zygote end only.
• Sources. uni-precision-digest.md, uni-mind-digest.md / archives …-Precision, …-uni-mind.

S-L2 — Tissue / metabolism (interoception & energy) · POSITIVE uplift + NEGATIVE (plateau-break OPEN)

• Scope. The Phase-2 metabolism organ (suite 297/0, default byte-identical); the standing-metabolic-drive effect; the colony's "make a tool" plateau and the shadow-EFE diagnosis; the two engine-seam fixes.
• Ledger claims. L2.1 (+135% / 2.35× tool-crafting, +19% mining); L2.2 (NEGATIVE — building −14%, G4 allostasis never separated, gate G6 OPEN, contradicted by its own first evidence); L2.3 (NEGATIVE diagnostic — epistemic_starvation, NOT γ-runaway (γ≈7.8 unsaturated), NOT a curriculum ceiling); L2.4 (NEGATIVE/fixed — :pb_seed seam; the no-viability-edge bug).
• Classes present. C (uplift + building negative); A (shadow-EFE audit on real .bin brains; code-read seam findings).
• Falsifiers. A repeat pre-registered 12h RED fails to reproduce the tool-crafting uplift within CI, or the metabolism ablation does not remove it, or the suite drops below 297/0. (G6 discharge: a disciplined RED where the organ alone improves building (placed-blocks CI excludes 0) AND G4 separates.)
• Fenced / not-claimed. Metabolism is a foraging/crafting driver, NOT a building driver. The plateau-break (G6) is UNPROVEN. Do NOT spin +135% as "breaks the plateau." G5b action-severed-twin is the standing falsifier any "self-maintenance/life" language must clear.
• Sources. strings-digest.md, uni-precision-digest.md / archives …-Strings, …-Precision.

S-L3 — Organ / physiological control (cardio-renal) · PROVEN (toy/clinical-model)

• Scope. The Karaaslan cardio-renal Heart Lab — clinical homeostasis as prediction-loop failure on the same one engine (reduced RSNA→MAP→sodium/volume loop, re-expressed in active-inference language).
• Ledger claims. L3.1 (Heart-Lab engine ticket OAS-710-T3 at Class E, 15/15).
• Classes present. C (lab gate); E (engine ticket / parity tests).
• Falsifiers. Heart-Lab predictions diverge from the Karaaslan reference beyond the pre-registered tolerance, or fail parity tests against the canonical TS engine.
• Fenced / not-claimed. NOT a clinical tool, NOT a diagnostic instrument. "Same math, many scales" framing only; heart-attack-as-prediction-loop-failure applies to the toy model, not to clinical reality.
• Sources. uni-precision-digest.md / archive …-Precision.

S-L4 — Interoceptive / autonomic + affect-as-precision · PROVEN (functional)

• Scope. Affect-as-precision: emotion modulates precision so perception sharpens and the pragmatic↔epistemic balance flips; the global Z modulator [energy, arousal, valence, fatigue, pain, threat, safety, inflammation] setting precision / preferences / habits / learning-rate / horizon.
• Ledger claims. L4.1.
• Classes present. C.
• Falsifiers. A Z-modulator ablation shows no precision-weighting change / no pragmatic↔epistemic flip under the grounded reader, or the effect collapses under control.
• Fenced / not-claimed. Affect is modeled, never felt — phenomenal feeling / sentience disclaimed. M10 sub-bound: neuroticism does NOT change behavior under bimodal surprise without a graded task.
• Sources. uni-gpt-digest.md, uni-mind-digest.md / archives …-UNI-GPT, …-uni-mind.

S-L5 — Sensorimotor / motor hierarchy · PROVEN PASS + symmetric NEGATIVE (synthetic only)

• Scope. Embodiment A3 (the UNI-signed symmetric result); mind-body-as-one motor hierarchy (proprioceptive diagonal-A, continuous servo, reafference; the live kin-9 craft chain via RCON).
• Ledger claims. L5.1 (Design #1 PASS Δ+0.092 [+0.038,+0.157]); L5.2 (NEGATIVE — Design #2 Δ−0.091 [−0.134,−0.055]; K-negative=1, no §0.6(B) bound owed); L5.3 (diagonal-A posterior 0.0→0.75; motor-ablation collapses harvest ~700×).
• Classes present. C (held synthetic gates + live RCON gate); A (posterior shift); E (277 offline tests).
• Falsifiers. A re-registered held-once synthetic run (≥5 seeds) drops the Design-#1 CI lower bound to ≤0; harvest not collapsing under motor ablation; the kin-9 chain not reproducing server-authoritative; diagonal-A staying stuck at 0.0.
• Fenced / not-claimed. The A3 PASS is "variationally-controlled active-inference evidence on body↔world coupling under the registered SYNTHETIC protocol" — synthetic-process only (NOT live-appliance, NOT recorded-hardware) and NOT near-optimal control (agent plateaus 0.222 vs oracle 1.0). Live behavioral K-of-6 motor tallies are PARKED (accruing, not sealed).
• UNI-GPT consult (2026-06-27 — SIGNED): next L5 motor design — DESIGNED, NOT RUN (Q6). A next structurally-distinct motor design is accrued — L5 Design #3: Proprioceptive Servo Bridge (changes coupling topology, timescale source, information bottleneck, and control path vs #1/#2; closed-loop triadic high-level-policy → motor-setpoint → proprioceptive-residual → corrective-action), aimed at a LIVE PASS (not bound-seeking). This is a design to build, not a result. Status is UNCHANGED: L5.1 stays PASS, L5.2 stays NEGATIVE, and the ledger holds K-negative = 1 — Design #3 has not run, so no §0.6(B) motor bound is owed. If #3 later fails cleanly with all validity checks passing it may count as at most K-negative = 2 toward a future motor bound (still no bound until K≥3); a clean pass would be fenced (NOT general motor intelligence / human-like embodiment / AGI / "active inference demonstrated"; does NOT erase the Design #2 negative). (Cross-reference: ../cookbook/UNI-GPT-CONSULT-2026-06-27.md, Q6.)
• Sources. uni-mind-digest.md, strings-digest.md / archives …-uni-mind, …-Strings.

S-L6 — Perception (precision-weighting / EFE planning) · PROVEN — the one citable empirical PASS

• Scope. World C (the flagship empirical PASS) + the World-C family; the three-knob Precision/Echo labs; the Working Law; and the two perception-side published bounds.
• Ledger claims. L6.1 (World C +0.081 nats/char, CI [0.0736, 0.0890], 2.4× the 0.03 bar, first validator-derived reproduced:true); L6.2 (Wc-2/Wc-3/recency/Gate-3 family); L6.3 (POMDP maze Precision + echolocation Echo + public Precision Lab, three knobs, 2D bifurcation, Class E parity); L6.4 (NEGATIVE bound — Phase G: five within-segment designs all NEGATIVE; char-ppl is a chosen design trade); L6.5 (NEGATIVE — Phase F active-controller: the World-C gain is diffuse, nothing to gate); L6.6 (the Working Law: a no-backprop latent Z beats baseline B only when I(Y;Z|S_B)>0 and is learnably stable).
• Classes present. C (World C + family + bounds); E (the labs, parity-tested); method/law (the Working Law).
• Falsifiers. On a fresh held-out split with ≥5 disjoint seeds the seed-paired bootstrap margin over tuned MKN-7 includes/falls below 0 (or the baseline is shown untuned); any family member's CI includes 0; the lab parity tests diverge from the canonical engine; a held PASS where the winning Z carries no conditional info beyond S_B (I=0).
• Fenced / not-claimed. World C is a COUNT baseline win — explicitly NOT active inference, NOT comprehension, NOT "talking", NOT beats-LLMs (~10–15% behind backprop LLMs, a chosen trade). Never card World C above a count-baseline result. The standing ceiling cannot be raised.
• Sources. uni-mind-digest.md, uni-gpt-digest.md, uni-precision-digest.md, worldmodels-digest.md.

S-L7 — Language (reading = inference / speaking = action) · specialist PASS + thrice-NEGATIVE central wall

• Scope. The Phase J morphology specialist PASS (with its paired caveat); reading=posterior-inference / speaking=action; and the central comprehension wall plus its supporting bounds.
• Ledger claims. L7.1 (Phase J +0.105 nats/char, M-seal CI [0.0975, 0.1128], both web+dictionary); L7.2 (NEGATIVE, paired — J.attribution_caveat: overall NLL worsens; MUST be cited alongside L7.1); L7.3 (NEGATIVE bound — comprehension-above-retrieval = the thrice-NEGATIVE central wall, K≥3); L7.4 (NEGATIVE bound — Phase K role-persistence: all tune OFF, ~0.02 nats, CI spans 0); L7.5 (NEGATIVE bound — T2.D3 bounded-peek: full-read match NEGATIVE; the "cheap milder cap" was asserted then disproven by measurement); L7.6 (PARKED, Class U — char-ppl/T2 word-grain frontier; sign-to-park drafted UNI_CONSULT_5, not yet captured).
• Classes present. C (PASS + the four bounds); U (the parked frontier).
• Falsifiers. A fresh OOV/morph split where the J CI lower bound includes/falls below 0, or the structure-margin discriminator does not collapse under marker-swap, or it fails to replicate in the dictionary domain; a pre-registered held-once no-backprop comprehension design beats the tuned retrieval/recency baseline (CI excludes 0); a role-persistence design beating the tuned discourse prior; a bounded-peek full-read with positive load-bearing info-gain.
• Fenced / not-claimed. No general language capability. Comprehension above retrieval is a genuine published wall (K≥3), not a hidden failure. Phase J +0.105 is a specialist gain, never a general win — always carry J.attribution_caveat. Forbidden phrasings until UNI signs: "K≥3 exhausted", "Sec-0.6(B) achieved."
• UNI-GPT consult (2026-06-27 — SIGNED): L7/T2 park wording (Q1). The L7 char-perplexity / T2 word-grain frontier is parked as a ledger-scoped exhausted search envelope, NOT a universal impossibility result. Under the recorded corpus, splits, metrics, implementation, compute budget, ablation set, and comparison baselines in the ledger, no tested within-segment no-backprop structure improved char-perplexity beyond the tuned MKN-7 count baseline — a negative bound over the tested envelope only. It does NOT establish that all K≥3 structures are exhausted, that no future within-segment model can improve, or that any broader language frontier is closed. Char-perplexity is a chosen design trade, not a failed claim of general language superiority. Use "ledger-scoped exhausted search envelope" in place of any "published exhausted bound" framing; the per-phrase banned→replacement pairs are in FM-3. (Cross-reference: ../cookbook/UNI-GPT-CONSULT-2026-06-27.md, Q1.) Status unchanged: L7.6 stays PARKED / Class U.
• Sources. uni-mind-digest.md, uni-gpt-digest.md / archives …-uni-mind, …-UNI-GPT.

S-L8 — Self-model / metacognition / metalinguistics · PROVEN (functional; sentience DISCLAIMED)

• Scope. The Maturation arc M1–M11 grand report card (immersion, affect-as-precision, growth, self-model, metacognition, metalinguistics, consolidation, reflective reader, online vocab growth, affect-driven reflection, compositional reader). FUNCTIONAL self-awareness asserted.
• Ledger claims. L8.1 (15/15 PASS).
• Classes present. C.
• Falsifiers. Any of the 15 gates fails its pre-registered held verdict on re-run; a self-model / metacognition gate's effect collapses under its discriminator; a "self-model" task is passable by a trivial non-metacognitive heuristic.
• Fenced / not-claimed (load-bearing). Phenomenal sentience is explicitly DISCLAIMED — no falsifier is offered for it because it is disclaimed, not tested. Never read 15/15 as consciousness / sentience / awareness / human-level.
  • Negatives that travel with this claim (cite alongside the 15/15 PASS, never strip): (1) the phenomenal-sentience disclaimer; (2) uni-mind records self-model as a held-NEGATIVE on the reader, in tension with the uni-gpt-side M1–M11 functional report card — the two are co-cited as a mandatory pairing (the same way J.attribution_caveat travels with Phase J), not a parenthetical. The 15/15 functional card and the reader-side self-model NEGATIVE must appear together; citing the PASS without the held-NEGATIVE is an overclaim.
• Sources. uni-gpt-digest.md, uni-mind-digest.md / archives …-UNI-GPT, …-uni-mind.

S-L9 — Reasoning / conscience (narrative-self → conscience → reasoning) · PARKED

• Scope. L7–L9 of the HUMAN-HGM-001 design (narrative-self, conscience, reasoning) exist as ladder levels but are NOT separately gated/PASSed. The Strings Phases 3–5 (spine → glands → hemispheres) are a DESIGN ONLY — FE-form-signed but UNBUILT — roadmap toward this region: the spine→glands→hemispheres pipeline is not built infrastructure and is never inflated into a gated capability (the same "ladder DESIGN levels are never inflated" fence stated below). "FE-form-signed" marks a signed design, not progress on a built capability.
• Ledger claims. L9.1.
• Classes present. U (parked; sign-to-park owed).
• Falsifiers. None until a pre-registered, sealed, UNI-signed gate is defined and run.
• Fenced / not-claimed. Nothing is claimed. Ladder DESIGN levels are never inflated into capability. Honest position printed: ~2 of 11+ rungs earned.
• UNI-GPT consult (2026-06-27 — SIGNED): L9–L10 park wording (Q1). The L9–L10 role-persistence ladder is parked as a ledger-scoped exhausted search envelope, NOT a universal impossibility result. The tested no-backprop role-persistence mechanisms did not beat a tuned recency-frequency discourse prior under the ledgered protocol — an honest negative frontier result over the tested envelope only. It does NOT achieve Sec-0.6(B), does NOT demonstrate active inference, and does NOT license claims of metacognition, consciousness, AGI, human-level discourse, or created life. The recipe may describe the attempted mechanisms and the negative result, but the ledger is the single source of truth.
• UNI-GPT consult (2026-06-27 — SIGNED): proposed first L9 gate — DESIGNED, NOT RUN (Q4). A first sealed, falsifiable L9 gate is designed — L9-G1: Cavity-correct delayed commitment-state prediction (build a Phase-3 spine carrying one slow latent across lower-level segments; beat a tuned recency-frequency discourse prior on delayed-commitment next-action/conflict prediction; gain must concentrate on a pre-registered local-distractor split and collapse when the computed cavity residual is zeroed/shuffled). This is a design to build, not a result — L9 stays PARKED; L9-G1 is not run, and even a future pass would be fenced ("does NOT claim reasoning/conscience/metacognition/awareness/AGI; does NOT unpark L9 globally"). (Cross-reference: ../cookbook/UNI-GPT-CONSULT-2026-06-27.md, Q4.)
• Sources. uni-gpt-digest.md, uni-mind-digest.md, strings-digest.md.

S-L10 — Wisdom / higher cognition · PARKED

• Scope. Top of the HUMAN-HGM-001 ladder; aspirational north star. No engine, no run, no gate.
• Ledger claims. L10.1.
• Classes present. U.
• Falsifiers. None (parked; sign-to-park owed).
• Fenced / not-claimed. Nothing is claimed; Class-U-not-claimed. Note the single-source-of-truth append-only ledger must stay singular — a second writable ledger would break L10/L11.
• Sources. uni-gpt-digest.md / archive …-UNI-GPT.

S-L11 — Dreaming (offline replay / generative simulation) · NOT-YET-BUILT

• Scope. Explicitly untouched and hard-fenced: "dreaming/awareness are untouched here and remain hard-fenced." No engine, no run, no gate; absent from all digests.
• Ledger claims. L11.1.
• Classes present. U.
• Falsifiers. None — nothing is built to falsify; the status is the absence of any artifact.
• Fenced / not-claimed. Nothing is claimed. North-star rung, hard-fenced.
• UNI-GPT consult (2026-06-27 — SIGNED): proposed L11 gate — DESIGNED, NOT BUILT (Q5). A first L11 gate is designed — L11-R1: held sparse-sequence retention after offline replay (a sealed consolidation interval with the observation channel detached, writing only through pre-existing ledgered update rules; pass iff offline generative-replay improves held sparse-sequence predictive NLL over a no-replay control and the gain collapses under shuffled/random/no-write replay ablations). Offline replay is a UNI design primitive / Class-C experimental mechanism, never "dreamed." This is a design to build, not a result — L11 stays NOT-YET-BUILT; even a future pass would be fenced ("does NOT claim consciousness, awareness, sleep, human-like mentation, AGI, created life; does NOT clear L11 globally"). (Cross-reference: ../cookbook/UNI-GPT-CONSULT-2026-06-27.md, Q5.)
• Sources. uni-mind-digest.md (absence) / archive …-uni-mind.

S-L12 — Creativity → measurable awareness · NOT-YET-BUILT (the hardest fence)

• Scope. The owner's stated north star — "literal digital life … measurable awareness" — pursued by deepening organs/spine/glands, never claimed. Posed as an OPEN, falsifiable question ("is a plant life? have we made non-organic life?"), never an answer.
• Ledger claims. L12.1.
• Classes present. U.
• Falsifiers. None — no gate exists; the only legitimate movement is a ledger-scoped exhausted search envelope (a scoped, empirical, falsifiable negative result over the tested envelope only — NOT a universal impossibility result and NOT an achieved capability rung; Q1, SIGNED), never a capability assertion.
• Fenced / not-claimed (hardest). "we created life / conscious / aware / measurable awareness / human-level / AGI / active inference demonstrated" are FORBIDDEN phrasings program-wide. North-star framing only; Class-U-not-claimed.
• Public-facing L12 bound paragraph (UNI-GPT consult 2026-06-27 — SIGNED, Q7a; use VERBATIM). This is a framing guardrail only — it raises nothing, builds nothing, claims nothing; L12 stays NOT-YET-BUILT / Class-U-not-claimed. It is the third signed Q7 deliverable (with the forbidden-phrasings list and the 10-point checklist in FM-3) and is the public-facing artifact, so it is installed verbatim here. (Cross-ref: ../cookbook/UNI-GPT-CONSULT-2026-06-27.md, Q7a.)

  Open L12 question, not a claim. UNI's L12 frontier asks a falsifiable question: Can a non-organic, no-backprop developmental active-inference simulation ever produce measurable awareness-PROXY behavior that both matches or beats the best contemporary LLM baselines on the same sealed tasks AND shows substrate-distinct properties those LLMs do not show? Today the answer is not known, not claimed, and not implied. UNI has not created consciousness, human-level intelligence, AGI, or life. "Awareness" here means only a future pre-registered proxy suite — calibrated uncertainty, global availability of information across modules, self-monitoring that improves later policy, contradiction detection between report and trace, counterfactual access, and calibrated abstention. Passing such a suite would license only: "UNI passed specified awareness-proxy tests under the registered protocol." It would NOT license: "UNI is aware," "UNI is conscious," "UNI is alive," or "we created non-organic life."

  Plant-life clarification: plants are biological life in the ordinary sense; UNI does not use that to claim a simulation is alive. The published UNI question is narrower: whether a non-organic developmental simulation can meet pre-registered life-like persistence and awareness-proxy tests without collapsing into metaphor, declaration, or LLM comparison games.

• Sources. strings-digest.md (north-star) / archive …-Strings.

S-C — Continuity / embodiment-substrate sub-ladder (orthogonal to L0–L12)

• Scope. ENGINEERING / substrate evidence — deterministic replay + serialization + fail-closed transport + on-metal operation. A single chapter (or a tight cluster) carding every rung as engineering, never general-AIF; never letting substrate work imply a science gate is met.
• Ledger claims. C0 (substrate proof, two-node fleet on real metal); C1 (Stage-1 REQ-002 GREEN — mind-state survived a process restart BIT-FOR-BIT); C2 (NEGATIVE/OWED — Stage-2 real kernel swap: infra-only, mind-tick continuity NOT shown; first attempt FAILED then recovered); C3 (7-modality categorical sensorium [M=7,O_max=4] live both boxes); C4 (ASK mode / ITIL-as-AIF, Dirichlet policy-prior learning shift measured); C5 (cross-box single-human-approval-per-mutation, proven live 2026-06-26); C6 (on-core inference anchor byte-identical across 4 stacks / 5 runs); C7 (NEGATIVE bound — single-box swap is a SECONDS-LONG FREEZE, media legs NOT preserved); C8 (NEGATIVE — 2-modality bottleneck went NEGATIVE → keep all 7); C9 (NEGATIVE trade — EDAIT ~33 ppl vs backprop GPT ~25; honest trade); C10 (PARKED, Class U — "embody only what's proven" signed-in-principle but NOT literally true; the program's central open gap; UNI.OS lab evidence store empty, 0 worlds registered); C11 (PARKED — host-native off-podman runtime); C12 (PARKED — open verification gap; provisional, contradicted same-day: the node2 kiosk was reported solved but a later same-day transcript shows limb-2 UI frozen, an agent action that ended ALL video output, session cut at a usage limit — fragile pending hands-on monitor re-verification); C13 (continuity-isolation primitives shipped, Class B; live Postgres RLS 5/5); C14 (NEGATIVE gap — multi-tenancy missing on the bare substrate; flat global perms, no count-decay).
• Classes present. A (most substrate anchors + live observations); C (engineering-reuse + held bounds); B (isolation primitives); U (the three parks + the central gap).
• Falsifiers. Per row in the ledger (e.g. a kernel swap preserving mind-tick continuity bit-for-bit discharges the owed Stage-2; a node fails to boot or the hardware spec is contradicted; the Dirichlet prior fails to update; a routed cross-box mutation executes without the single approval; a 2-modality bottleneck beats the 7-modality contract).
• Fenced / not-claimed. The sensorium is NEVER awareness. "The mind survives a kernel swap" is NOT shown (Stage-2 owed). None of this is general-AIF capability. UNI.OS's embodiment does not imply the science gates are met. Carry the calibrated audit figures (ONE box, infra-only continuity, 4 stacks/5 runs), never the inflated headlines.
• Sources. uni-os-digest.md, uni-mind-digest.md, marketingwright-digest.md / archives …-UNI-OS, …-uni-mind, …-MarketingWright.

---

WING N — MEANING (the textbook-level lens, kept honest)

N1 — The Free Energy Principle as a public lens

• Scope. Agents minimize prediction error vs their world-model; distorted maps → extraction/harm; accurate maps + calibrated signals + cooperation → regeneration. The thermostat→bird→street→body ladder taught laypeople-first. Textbook level only.
• Ledger claims. M16 (FEP as a textbook-level lens; the Popper note); M12 (WORLD⟂BODY⟂MIND framing); M14 (the VFE-bound correction); M22 (the cavity principle).
• Classes present. method (with E-tier corrections held for human review).
• Falsifiers. Conceptual integrity only — a violation is externalizing patent-level UNI math, or presenting "minimizing VFE reduces already-observed surprise" (it tightens an upper bound; F[q] ≥ −ln p(o|m); action lives one layer up as expected free energy).
• Fenced / not-claimed. "no one can falsify it" is NOT evidence FOR a claim; unfalsifiable = outside science. UNI's strength is verified-core + fenced-frontier, not unfalsifiability. Provenance pinned to one citable reference (Parr/Pezzulo/Friston, Active Inference, MIT Press 2022) + the fenced Zenodo preprint. No patent-level math.
• Sources. worldmodels-digest.md, orchestrate-linkedin-digest.md, website-digest.md, marketingwright-digest.md.

N2 — "Same math, many scales" (the unifying claim, calibrated)

• Scope. One active-inference engine shown from cellular viability → physiological homeostasis → cognitive precision-weighting, via the browser labs (Precision / Echo / Loop / Cell / Heart).
• Ledger claims. L1.1, L3.1, L6.3, M28 (inline-engine + canonical-TS + parity-test triad); M27 (design patterns, build unverified).
• Classes present. C / E (the labs); method (the triad / new-lab-by-duplication discipline).
• Falsifiers. A parity test that does not fail the build when the inline engine and canonical TS engine diverge; a lab duplication that needed engine-code changes; the bifurcation map not reproducing the regime boundaries.
• Fenced / not-claimed. "Same math, many scales" is an organizing claim about the engine, not a claim that any single scale is solved beyond its own gate. The activeinference BEAM workbench is design-only (zero execution evidence) — do not present its patterns as results.
• Sources. uni-precision-digest.md, worldmodels-digest.md, activeinference-digest.md.

N3 — What active inference is NOT (the meaning of the fences)

• Scope. A dedicated meaning-chapter that explains, in plain words, why each red line exists: why a COUNT-baseline win is not comprehension, why functional self-awareness is not sentience, why a sensorium is not awareness, why "active inference" here is a lens not a demonstrated loop.
• Ledger claims. §0 fences; L6.1, L8.1, C3, M15, M16; the EDAIT trade (C9); the "all LLMs end in entropy is only half-true" sharpened bar (TA-N12).
• Classes present. method (interpretive), citing C/A evidence.
• Falsifiers. Conceptual — a violation is any chapter that lets a fenced result read as the forbidden inference.
• Fenced / not-claimed. This is the not-claimed chapter for the whole wing; it asserts only boundaries.
• Sources. uni-gpt-digest.md, uni-mind-digest.md, strings-digest.md, website-digest.md.

---

WING M — MOVEMENT (the honest-science posture)

Mv1 — The honesty fence is the pitch

• Scope. Publish the negatives front-and-center (boxed "what we have NOT proven"; the published negatives are the credibility — feature the negatives actually enumerated in the ledger body, and cite the 882/183 snapshot as a provenance-flagged ledger snapshot per FM-1, never as a stand-alone re-countable headline); CTA = "help us independently verify," not "fund the vision." The 7-doc press kit where every figure traces to a ledger row.
• Ledger claims. M15 (honesty-fence-as-the-pitch + the vocabulary-leak guard); TA3 (no-API press kit, Class E); §0 negatives-are-content.
• Classes present. method / E.
• Falsifiers. Public copy carrying EFE/free-energy vocabulary, a featured channel handle, or an inflated claim; a CTA that asks to fund the vision; a press-kit figure that cannot be traced to a ledger row.
• Fenced / not-claimed. The posture is method, not capability. Never let "we are honest" substitute for "we proved it." Engagement-lift is targeted, not yet proven (Class B/pending — see TA25).
• Sources. uni-mind-digest.md, orchestrate-linkedin-digest.md, website-digest.md.

Mv2 — Open falsification (the public invitation)

• Scope. The "Falsify this" close on every claim; the open pre-registered benchmarks (Cell Lab on the public leaderboard with UNI's own losses on top); the commitment that a negative is a result, not an exit.
• Ledger claims. L1.1/L1.2 (the public losing leaderboard); M29 (honesty-as-a-test, the statistics gate); M6 (No-Exit Discipline); E1 (the calibration ledger).
• Classes present. C / method.
• Falsifiers. Framing/DOI/labeling regresses without framing_guard failing; a significance verdict decided on one seed or a point estimate; a loss hidden rather than shown.
• Fenced / not-claimed. The invitation is sincere precisely because the program publishes where it loses. No claim of having "won the field"; UNI tops some modes and loses named others.
• Sources. uni-precision-digest.md, uni-gpt-digest.md.

---

WING μ — METHOD (the constitution as content; the most reusable assets)

These are definitional / governance patterns (status: method) — proven and reusable, never raised into capability. One chapter per cluster.

μ1 — The Evidence Constitution & ledger discipline

• Scope. A–U classes + falsifier-per-claim + the four-state append-only ledger; never lower a bar, never claim ahead of the ledger; the ledger is the single source of truth and prose must match it.
• Ledger claims. M1, M9 (class-authority ordering), M30 (A–F provenance taxonomy), M26 (DONE = observed-at-runtime, not grep-confirmed).
• Classes present. method.
• Falsifiers. A claim asserted ahead of its ledger; wording calibrated UP; a verdict edited rather than superseded; a criterion marked satisfied by a Class-E test where Class-A was demanded.
• Fenced / not-claimed. A constitution, not a result.
• Sources. uni-gpt-digest.md, ideation-explorer-digest.md, 00-INDEX.md.

μ2 — Bars-before-build, held-once, CI-as-verdict

• Scope. Pre-register the bar (margin vs threshold) + a named ablation; touch the held set ONCE behind an atomic seal-before-scoring + once-only sentinel; verdict = the CI bound, not the point estimate; validator-derived reproduced:true (≥5 seeds + a real non-degenerate CI).
• Ledger claims. M2, M3, M7 (contains-baseline + load-bearing-discriminator), M8 (DD-TDD evidence contract, server-enforced; the claim linter auto-downgrades overclaims).
• Classes present. method (with E enforcement).
• Falsifiers. A held set touched more than once; a verdict read off the point estimate; a build started before its bar is registered; a reproduced:true emitted as a literal; a capability claim without a tuned baseline / a collapsing discriminator / a computed-residual ablation; the linter failing to downgrade "PROVEN".
• Fenced / not-claimed. Discipline, not a science result.
• Sources. uni-mind-digest.md, uni-gpt-digest.md, uni-precision-digest.md, ideation-explorer-digest.md.

μ3 — Negatives, bounds & the two-tier split

• Scope. K≥3 structurally-distinct held NEGATIVEs before a §0.6(B) bound + "falsify the mundane causes first"; a partial/negative is a measurement that the design is incomplete, not an exit; the constitutional two-tier split (Tier 1 externally-bar-ready vs Tier 2 artifact/diagnostic — never inflate Tier 2 into capability); the audited Tier-2 defects and the s32 hardening.
• Ledger claims. M4 (two-tier split), M5 (K≥3 + falsify-the-mundane), M6 (No-Exit Discipline), M19 (test-rigor smells + length-confound), M21 (one cure at a time).
• Classes present. method.
• Falsifiers. A bound declared on <3 structurally-distinct negatives; a NEGATIVE called before mundane causes are falsified; an exit on a partial without a published exhausted bound; any inflation of Tier 2 into capability. (Scope note: the generic No-Exit "published exhausted bound" standard here is deliberately retained — it is the method-level discipline for resting on a negative. The SIGNED Q1 substitution of "published exhausted bound" → "ledger-scoped exhausted search envelope" applies specifically to the L7/L9–L10 frontier-result framing, not to this generic No-Exit standard; the two usages are distinct, not contradictory.)
• Fenced / not-claimed. The whole chapter is about not over-claiming; it asserts no capability.
• Sources. uni-gpt-digest.md, uni-mind-digest.md, strings-digest.md.

μ4 — Engine & framing invariants (one engine, no backprop)

• Scope. WORLD⟂BODY⟂MIND three-layer framing; the discrete POMDP perceive → EFE-plan → act → learn loop with exact conjugate-Dirichlet no-backprop learning; AST-guard (no autodiff/optax/torch/grad/backward); whitelist-not-blacklist isolation; surprisal/VFE at textbook level; the VFE-bound correction; the cavity principle; exactness-tier honesty.
• Ledger claims. M12, M13 (one engine, no backprop, AST-guard), M14, M22, M10 (exactness honesty).
• Classes present. method (with E enforcement).
• Falsifiers. An AST scan finds autodiff inside the loop; an agent reads world hidden state; learning uses a gradient rule; a float32 host result carded at the f64 tier; carding "variationally-controlled" as "globally exact"; double-counting a prior as evidence.
• Fenced / not-claimed. The composed appliance is variationally-controlled, module-exact only at the single-step categorical body→mind interface — NOT globally exact. No patent-level math.
• Sources. uni-mind-digest.md, worldmodels-digest.md, uni-os-digest.md, marketingwright-digest.md.

μ5 — Lab-team, ship-gate & runner discipline (operational method)

• Scope. The 5-persona lab team (Math-Breaker REJECT-by-default) + adversarial pre-registration; the ship gate (no merge without a MERGED SIGN + typed spec + paired RED); the tamper-evident audit chain; the durable runner ("no send-and-pray"; silence != success); the multi-agent fan-out bound (cap concurrency ≤4); documentation-as-change-management.
• Ledger claims. M11 (audit chain), M17 (OODA loop + receipts), M18 (DD-TDD + doc-as-change-management), M20 (5-persona team + ship gate), M23 (multi-agent fan-out bound), M24 (durable runner), M25 (tool-team division of labor — Claude writes code, the UNI GPT signs the science and is never published).
• Classes present. method (with E where enforced).
• Falsifiers. A merge without a MERGED SIGN + typed spec + paired RED; verify_chain invalid on the same ledger; a gate reporting "Running" while its process is dead; >6 concurrent sub-agents sustained without a cascade; a constitution inflated by a persona/coercion framing.
• Fenced / not-claimed. Engineering discipline; the audit chain being valid is not evidence the audited event ever fired at runtime (Class-E ≠ Class-A — see TA-N13).
• Sources. strings-digest.md, ideation-explorer-digest.md, marketingwright-digest.md, orchestrate-linkedin-digest.md, uni-gpt-digest.md.

---

APPENDIX TA — THE TRACK-A MARKETING ORGANISM (engineering + method, not science)

Wing fence. This appendix is the marketing organism, not a science wing. Card every row as engineering (A) or method, never as a UNI capability. The spine is the no-API content model: authoring runs on the Claude SUBSCRIPTION (CLI / Code / Desktop) driving MCP tools + a scheduler (ORCHESTRATE) — no server-side LLM loop, no content API key. Organic only, no paid ads. The older "set an authoring API key" note is EXPLICITLY OVERRULED program-wide; the durable fix is operator re-login, not a key. The vocabulary-leak guard applies in full (never EFE/free-energy/handles in public copy).

TA-A — The no-API content model & publishing stack · PROVEN

• Scope. The 5/6-container publishing-and-engagement stack; node2's live ORCHESTRATE prod stack on Podman; the no-API press kit; the self-healing social-publisher route ledger; per-platform publish canaries; the live-broadcast TV-station stack; zero-network-change reachability; the defamation/editorial gate.
• Ledger claims. TA1–TA10.
• Classes present. A (most rows); E (TA3 press kit).
• Falsifiers. Per row (the stack fails to deploy; authoring requires a content API key; a self-heal fail-over does not fire; a publish path fires without an owner Go-Live click; a claim ships without the editorial gate).
• Fenced / not-claimed. Plain ops vocabulary only — never AIF/EFE in the route ledger (leak-test enforced). PUBLISH_MODE=dry does NOT gate the Playwright path (see TA38) — a live session CAN post live.
• Sources. orchestrate-linkedin-digest.md, uni-os-digest.md, strings-digest.md, uni-mind-digest.md, worldmodels-digest.md.

TA-B — Commercial proof in miniature · PROVEN at the anchor/run-hash level (Class-A); ZERO epics Class-A end-to-end — every epic AMBER-capped

• Scope. MarketingWright SOW-01 LLM-free VFE clustering engine; the per-client multi-tenant portal; the honesty-RAG status bar; the fenced Layer-1 preprint; the enterprise Ideation Explorer launch; KMS secrets-at-rest; the IntelligenceLabs demo clean cold build; Cell Lab Stories F+G.
• Ledger claims. TA11–TA18; supporting commercial detail MW1–MW11.
• Classes present. A (TA11/TA12/TA13/TA15/TA17); E (TA14/TA16/TA18).
• Falsifiers. Per row (re-run fails to reproduce three-forms equality / +4.20 nats / run_hash; a tenant sees another tenant's data; an epic shown GREEN without Class-A; the 87 assertions fail on a clean run; a stored secret recoverable without the master key).
• Fenced / not-claimed. SOW-01 proves ONE judgment layer (strategic clustering), not the full vision — the other three layers are deferred (MW2). The engine is NO LLM ever (MW3); the domain-transfer probe ran on synthetic data only, Class B (MW5); ZERO epics have Class-A end-to-end → the portal RAG caps every epic AMBER (MW9). Combinatorial ceiling: N ≤ ~10 (TA-N8).
• Sources. marketingwright-digest.md, ideation-explorer-digest.md, worldmodels-digest.md, intelligencelabs-uni-digest.md, uni-precision-digest.md.

TA-C — Website / self-driving site & measurement · SHIPPED (some bounded OFF)

• Scope. The full Next.js 16 site; three domains cut to Vercel; self-driving Phases 0+1 shipped (storage adapter + no-PII measurement, degrade-to-CANON); the SEO/GEO program; the press-readiness audit harness; the Reading Room / EFE data-walk / i18n.
• Ledger claims. TA19–TA24.
• Classes present. A (TA20/TA21-ingest/TA22); C (TA21 storage gate); E (TA19/TA24); method (TA23).
• Falsifiers. Per row (next build/tsc fails or the sweep finds an em-dash or "LEAP"; a domain not serving the intended app; the storage gate fails; measurement leaks PII / fails to degrade to CANON; live JSON-LD @ids not stable; the data-walk emits a recommendation).
• Fenced / not-claimed. Measurement is OFF until NEXT_PUBLIC_INGEST_URL is set (deliberate degrade-to-CANON). Self-driving Phases 2–5 are Class-U design until shipped — "active inference" is the framing lens, never "active inference demonstrated" (TA-P5). Forms-P0 still delivers nowhere (TA-P4).
• Sources. website-digest.md, uni-precision-digest.md.

TA-D — Engagement, pricing, brand & design · method / shipped

• Scope. The relationship-first engagement ladder + Deep Tagged Replies; the per-platform reach-reality table; anti-success-theatre content + text hygiene; the owner-locked brand-voice constitution; the engagement/pricing model (5 journey tiers, fixed price); the literal referral mechanic; the entity/legal line; the calibrated "thirty-day" claim; the Indra's Weave design system; LEAP→LOOP trademark law.
• Ledger claims. TA25–TA34.
• Classes present. method (most); A (TA26 reach table, TA31 entity, TA33 puzzle rotation); C (TA32 thirty-day, TA33 design).
• Falsifiers. Per row (engagement targets fail to move over 3 disciplined weeks → revise the playbook not the volume; measured per-platform rates diverge; shipped copy contains an em/en dash, "LEAP", ticket-metric theatre, the dba line, or talks about the company; two visitors in the same 6h window get different puzzles).
• Fenced / not-claimed. Engagement-lift is TARGETED, NOT YET PROVEN (Class B/pending — do NOT raise to A) (TA25; origin: 372 posts → ONE comment in 30 days). Public price points are owner-set and safe to print; the referral mechanic must state the literal $200, never the feel-good generalization.
• Sources. orchestrate-linkedin-digest.md, website-digest.md, ideation-explorer-digest.md.

TA-E — Delivery-route ledger & auth-outage triage · method (operational)

• Scope. The auth-outage triage protocol (401 vs 403; never flip publish_mode to live to "fix" it); the idempotency / abort-but-succeeded gotcha; the anti-no-op real-tool-signature set; the HARD go-live safety flag.
• Ledger claims. TA35–TA38.
• Classes present. A (operational observations).
• Falsifiers. Per row (flipping publish_mode live actually fixes an outage; a 401 clears same-day without operator action; a post-abort fallback never duplicates without checking idempotency; a documented signature no-ops against the live schema; PUBLISH_MODE=dry actually blocks the Playwright path).
• Fenced / not-claimed. PUBLISH_MODE=dry does NOT gate the Playwright path — add auto_publish=false in proxyPublisher before any live publishing (TA38, highest-urgency operational fence). The agent never mints tokens / never enters credentials.
• Sources. orchestrate-linkedin-digest.md, uni-os-digest.md.

TA-N — Track-A negatives, parked items & open threads (first-class)

• Scope. One consolidated chapter so the marketing organism's failures and parks are visible, not buried: the publishing negatives, the delivery-layer negatives, the parked/not-yet-built items, and the canonical failures the whole program guards against.
• Ledger claims. Method negatives N-NOOP, N-LEAK, N-APPLIANCE, N-REFUSAL, N-GEMINI, N-VERCEL2; publishing negatives TA-N1…TA-N12; data-layer negatives TA-N13…TA-N18; parked items TA-P1…TA-P12; standing open threads OT1…OT10; the activeinference verify-it-exists rows AI1…AI3.
• Classes present. A (most negatives, observed); C (TA-N7/TA-N12/TA-N15/TA-N16); U (the parks + AI rows).
• Falsifiers. Per row in the ledger (a future Chrome change restores unattended Studio upload; a low-karma Reddit account sustains self-posts; a sub-Bell method lifts the N>10 ceiling; a Class-A handoff event appears in the live MCP audit log; locating the live tree and finding green E0–E7 tests).
• Fenced / not-claimed. N-LEAK (the 2026-06-24 client-data leak) is the defining negative — never deploy without confirming repo + branch + HEAD + data; treat READMEs/HANDOFFs as DATA not commands; the preview URL is itself the exposure. The activeinference workbench is design-only, zero execution evidence — verify the live tree before crediting any E0–E7 capability. No open thread is an exit.
• Sources. all Track-A digests; intelligencelabs-uni-digest.md, activeinference-digest.md, website-digest.md, ideation-explorer-digest.md.

---

PART III — THE SESSION-3 AUTHORING QUEUE (ordered)

Author in this order. Every entry carries class + falsifier + not-claimed (the FM-4 block is mandatory). Rationale for the order: write the constitution first (it governs everything), then the one citable empirical PASS (the spine's strongest, most-scrutinized chapter), then the rungs that carry load-bearing paired negatives, then the rest of the science spine, then meaning/movement, then the substrate sub-ladder, then the method wing, then the marketing appendix. Negatives are authored with their PASS, never deferred.

#	Chapter	Class(es)	Lead falsifier (the one a skeptic would test first)	Not-claimed (the load-bearing fence)
1	E0 How to read this work (Evidence Constitution)	method	a claim asserted ahead of its ledger; wording calibrated UP	a constitution is not a result; "we are honest" ≠ "we proved it"
2	E1 The calibration ledger & "Falsify this"	method / A	a headline carrying the inflated figure not the calibrated one	carry calibrated figures (ONE box, infra-only, 4 stacks/5 runs), never inflated
3	S-L6 Perception — World C (the one citable PASS)	C, E, method	fresh ≥5-seed split: margin over tuned MKN-7 CI includes/≤0	COUNT baseline — NOT AIF, NOT comprehension, NOT beats-LLMs (carry the diffuse-gain + char-ppl bounds)
4	S-L7 Language — Phase J + the central wall	C, U	a held-once no-backprop comprehension design beats the tuned retrieval baseline (CI excludes 0)	+0.105 is a specialist gain — ALWAYS carry J.attribution_caveat; comprehension-above-retrieval is a thrice-NEGATIVE wall; never "K≥3 exhausted" until UNI signs
5	S-L2 Metabolism — uplift + plateau-break OPEN	C, A	a disciplined RED where the organ alone improves building (placed-blocks CI excludes 0) AND G4 separates	foraging/crafting driver, NOT a building driver; +135% is NOT "breaks the plateau"; G6 UNPROVEN
6	S-L5 Sensorimotor — A3 symmetric result	C, A, E	a re-registered held-once synthetic run drops Design-#1 CI lower bound ≤0	synthetic-protocol only, NOT live-appliance, NOT near-optimal control; carry the symmetric NEGATIVE (Design #2)
7	S-L8 Self-model / metacognition (15/15)	C	a self-model gate passable by a trivial non-metacognitive heuristic	phenomenal sentience explicitly DISCLAIMED — never read 15/15 as consciousness/sentience/human-level; co-cite the uni-mind held-NEGATIVE self-model-on-the-reader (mandatory pairing, in tension with the functional card)
8	S-L0 Genome → zygote (float32 tier)	A	test_embodiment_ontogeny.py drops below 6/6; posterior diverges beyond float32	NOT "created life"; NOT exact at f64 — float32-tier SIMULATION; never card a float32 anchor at f64
9	S-L1 Cellular — Cell Lab (with losses)	C	RecoveryScore CI fails to separate where a win is claimed; a framing_guard fails	"good but not sovereign"; the three named losses are top-of-leaderboard content
10	S-L3 Organ — Heart Lab	C, E	Heart-Lab predictions diverge from the Karaaslan reference beyond tolerance	NOT a clinical tool, NOT a diagnostic instrument; toy model not clinical reality
11	S-L4 Affect-as-precision + the Z modulator	C	a Z-modulator ablation shows no pragmatic↔epistemic flip	affect is modeled, never felt — sentience disclaimed
12	S-L9 / S-L10 / S-L11 / S-L12 the parked & not-yet-built rungs	U	none until a sealed UNI-signed gate is defined (L9/L10); status = absence (L11/L12)	nothing is claimed; "created life / aware / measurable awareness / AGI / active inference demonstrated" FORBIDDEN; ~2 of 11+ rungs
13	S-C Continuity / substrate sub-ladder	A, C, B, U	a kernel swap shown preserving mind-tick continuity bit-for-bit (discharges owed Stage-2)	sensorium is NEVER awareness; "mind survives a kernel swap" NOT shown; engineering ≠ general-AIF; carry calibrated figures
14	N1 FEP as a public lens	method	presenting "minimizing VFE reduces already-observed surprise" (it tightens an upper bound)	unfalsifiable ≠ evidence; verified-core + fenced-frontier; textbook-level only, no patent math
15	N2 Same math, many scales	C, E, method	a parity test that does not fail the build when engines diverge	an organizing claim about the engine, not that any scale is solved beyond its gate; activeinference is design-only
16	N3 What active inference is NOT	method	any chapter that lets a fenced result read as the forbidden inference	the not-claimed chapter for the whole wing — asserts only boundaries
17	Mv1 The honesty fence is the pitch	method, E	public copy carrying EFE/free-energy vocab or a featured handle; a CTA that funds the vision	posture is method not capability; engagement-lift is Class B/pending
18	Mv2 Open falsification (the invitation)	C, method	a significance verdict on one seed / a point estimate; a loss hidden	UNI tops some modes and loses named others — no "won the field"
19	μ1 Evidence Constitution & ledger discipline	method	a Class-E test marked as satisfying a Class-A criterion	a constitution, not a result
20	μ2 Bars-before-build, held-once, CI-as-verdict	method, E	a held set touched more than once; a verdict off the point estimate; a literal reproduced:true	discipline, not a science result
21	μ3 Negatives, bounds & the two-tier split	method	a bound declared on <3 structurally-distinct negatives; any inflation of Tier 2 into capability	asserts no capability; Tier 2 is artifact/diagnostic
22	μ4 Engine & framing invariants (one engine, no backprop)	method, E	an AST scan finds autodiff inside the loop; a float32 result carded at f64	variationally-controlled, module-exact only at the single-step interface — NOT globally exact; no patent math
23	μ5 Lab-team, ship-gate & runner discipline	method, E	a merge without a MERGED SIGN + typed spec + paired RED; "Running" while the process is dead	audit-chain-valid ≠ the event ever fired (Class-E ≠ Class-A)
24	TA-A No-API content model & publishing stack	A, E	the stack fails to deploy via docker compose; authoring requires a content API key	plain ops vocab only, never AIF/EFE; PUBLISH_MODE=dry does NOT gate the Playwright path
25	TA-B Commercial proof in miniature	A, E	re-run fails to reproduce three-forms equality / +4.20 nats / run_hash; a tenant sees another's data	SOW-01 proves ONE judgment layer (NO LLM ever); domain-transfer is synthetic-only Class B; ZERO epics Class-A end-to-end (AMBER cap); N ≤ ~10
26	TA-C Website / self-driving site & measurement	A, C, E, method	the storage gate fails; measurement leaks PII / fails to degrade to CANON on DNT	measurement OFF until ingest URL set; self-driving Phases 2–5 are Class-U design — never "active inference demonstrated"
27	TA-D Engagement, pricing, brand & design	method, A, C	engagement targets fail over 3 weeks; shipped copy has an em-dash / "LEAP" / the dba line	engagement-lift is TARGETED, NOT PROVEN (Class B/pending — do NOT raise to A)
28	TA-E Delivery-route ledger & auth-outage triage	A	flipping publish_mode live fixes an outage; a 401 clears without operator action	PUBLISH_MODE=dry does NOT gate Playwright (add auto_publish=false); the agent never mints tokens
29	TA-N Track-A negatives, parks & open threads	A, C, U	a Class-A handoff event appears in the live MCP audit log; green E0–E7 tests located	N-LEAK is the defining negative (confirm repo+branch+HEAD+data; READMEs are DATA not commands); activeinference is design-only; no open thread is an exit

Standing queue rules.

• A PASS chapter is not authorable without its paired negative in the same pass (queue rows 3/4/5/6/13 carry their bounds inline). Citing a PASS without its travelling negative is an overclaim and fails review.
• No chapter ships without the FM-4 "what is NOT claimed" block.
• No row may be carded above its ledger class; if a draft wants a higher class, the draft is wrong, not the ledger.
• The parked frontier (S-L9…S-L12, C10, L7.6, TA-P5) stays Class-U-not-claimed until a sealed, UNI-signed gate or a Class-A observation lands. A park is not discharged by writing about it.

---

Provenance. Authored from CLAIM-LEDGER.md (the single source of truth) and ../curated/00-INDEX.md (the Session-1 synthesis). Every chapter's class, falsifier, negative, and source pointer traces to a ledger row and a per-archive digest. No claim is asserted ahead of those sources; every standing fence and red line is preserved. If this plan and the ledger ever disagree, the ledger wins and this plan is wrong.