diff --git a/docs/assessment/50-execution-plan.md b/docs/assessment/50-execution-plan.md index 9cd7c2bd..6b4954b2 100644 --- a/docs/assessment/50-execution-plan.md +++ b/docs/assessment/50-execution-plan.md @@ -3,7 +3,7 @@ **Planner:** Opus 5 · 2026-07-27 · verified at `forgejo/main` @ `ed06dd64` **Governs:** everything in `30-gap-register.md` (G-1…G-21) and `31-hindrance-audit.md` (H-1…H-14), sequenced by `40-assessment.md` §6. **Method:** `docs/conceptualizing_engineering_mastery.md` — scrub → **delete** → simplify/enforce → accelerate → automate last. Nothing is automated that Waves 0–3 have not proven. -**Status (2026-07-28, at `d0bedfc1`):** **Wave 0 is closed.** Twelve rulings adopted, two stricken, the §5 NO-GO ratified — recorded in `50-rulings.md`. Every plan item that needed neither a ruling nor an ADR was already landed; the ruling-gated items are now *unblocked and owed*, in the order fixed in §2.1 below. +**Status (2026-07-28):** **Wave 0 closed, and the entire adopted docket is executed** — R-7, R-12, R-3+R-4, R-9+R-2, R-13, R-8, R-1/R-5/R-6/R-11. **Wave 0 is closed.** Twelve rulings adopted, two stricken, the §5 NO-GO ratified — recorded in `50-rulings.md`. Every plan item that needed neither a ruling nor an ADR was already landed; the ruling-gated items are now *unblocked and owed*, in the order fixed in §2.1 below. | Landed | Blocked, and on what | |---|---| @@ -11,7 +11,7 @@ | **Track A — ADR-0252 §5 run to NO-GO**, criterion pre-registered, artifact + digest committed, **verdict ratified 2026-07-28** | PR-8 — a small ADR (H-3/G-20) · PR-10 — a refactor ADR (H-4) | | PR-4 membership + reachability ratchets · PR-6 three doctrine pins · PR-7 M2 trust table · PR-9 count-the-swallow · **PR-5 the flag register (R-3+R-4)** | PR-12 — an ADR **amendment** (R-13 authorizes it; the amendment is still owed) | | PR-3 + PR-3b (**R-7**, Wave 1) · **PR-11 the soak to committed evidence (R-9+R-2)** · H-13 · H-8e · G-22 · G-23 | Track B — the fabrication ADR (standing instruction, unchanged) | -| **Closed:** G-5, G-6, G-7, G-8, G-9, G-10, G-15, G-19, G-22, H-1, H-6, H-7, H-8 (**all five**), H-11, H-13 · **Answered:** G-1 · **Discharged:** H-10 · **Pinned:** G-23 | Track C — invention, deliberately last | +| **Closed:** G-5…G-12, G-15, G-17, G-19, G-22, H-1, H-6, H-7, H-8 (**all five**), H-11, H-13 · **Answered:** G-1 · **Discharged:** H-10 · **Pinned:** G-23 | Track C — invention, deliberately last | **New this arc, not in the original assessment:** **N-8** (the §5 experiment had already returned GO twice, on unmerged branches, both unsound), **N-9** (the gate "drift" was a recorded decision — PR-4's pin 1 **withdrawn** rather than built, R-14 dissolved), **G-21** (the math reader decides **1.0%** of `holdout_dev/v1`), **G-22** (`main` was red before the arc, and `CLAIMS.md` published a superseded evidence digest for two days), **H-13**, **H-14**, **H-8e**. @@ -220,9 +220,11 @@ The packet's original order was by register number. That is the wrong axis: it i | 4 | ~~**R-9 + R-2**~~ | PR-11 — the soak re-run at 5000 beats under the **corrected** profile, artifact committed, digest pinned, cadence enforced by a source-hash pin, H1–H4 on the gate | **M** | **EXECUTED 2026-07-28.** Closed **G-5**. R-2 landed in the contract *before* the re-run, as designed. The staleness pin **would have caught PR-5** — verified by sabotage. | | 5 | ~~**R-13**~~ | PR-12 — distinct-evidence counting at the seal boundary, the honest re-count, demotions applied, the producer hardened against re-padding | **L** | **EXECUTED 2026-07-28. 25 licensed bands → 4.** Served capability shrank by 21 bands and `wrong` stayed 0: no answer changed, only the claim attached to it. | | 6 | ~~**R-8**~~ | PR-14 — the outcome-mix rule as two capabilities. **No ledger sealed: under the rule, zero bands license** | **M** | **EXECUTED 2026-07-28.** Overturns N-5 — the four bands cleared only on *pooled* evidence, by 0.000046. Closed **G-10**. | -| 7 | **R-1 · R-5 · R-6 · R-11** | Posture statements: efferent-action deferral with entry criterion (G-12) · the identity-enforcement discrimination bar (G-11) · non-text ingest deferral with the falsification bench as its standard (G-17) · **R-11 B — measure before gating** | **S** each | All four are prose with a criterion. None blocks anything; leaving them silent is the cost. They go last because they are the only items where lateness is cheap. | +| 7 | ~~**R-1 · R-5 · R-6 · R-11**~~ | `docs/specs/postures.md` — three posture statements, each with the criterion that would change it; **R-11's interim gate WITHDRAWN on measurement** | **S** each | **EXECUTED 2026-07-28.** Closed **G-11, G-12, G-17**. R-11's measurement dissolved its own premise — see below. | -**R-11's measurement is owed now, not later — and R-11 is the one adopted ruling that does not end in a decision.** Option B is *measure first, then re-ask*: an instrumentation PR (counts only, no behavior change) reporting how often the out-of-inventory reading path is taken on the practice corpora, after which **R-11 is put back to Shay with the number in hand.** So the docket owes two things here, not one: the measurement, and the second ruling. Until the number exists, R-11 has been ruled and nothing follows from it — which is a state worth naming, because a ruled-but-inert item is indistinguishable from a settled one at a glance. It lands with the R-1/R-5/R-6 batch at the latest and may land sooner, since the instrumentation is small and independent. +**R-11 — MEASURED 2026-07-28, and the result was not a rate.** Option B was *measure first, then re-ask*, and the instrumentation returned something more useful than a number: **the question could not be asked in that form.** Across **11,199** distinct serving-path inputs and **23,562** clauses, **zero** fell outside the verified inventory — because `atom_fact`'s template is `{p}`, a slot with no literal anchor that matches any string. *Nothing* is syntactically outside the 19. The proposed gate would have refused nothing while presenting as a safety mechanism, which is this repository's dominant defect class. **Second ruling (delegated): option A is WITHDRAWN as unimplementable as specified — not deferred** — and the reframing is the finding: G-2's fabrications are **mis-reads of admissible surfaces**, not admissions of out-of-inventory constructions, so no gate over the admissibility set can catch them. The fabrication ADR inherits the boundary question. Recorded as **P-4**. + +*(Superseded framing, kept for the record:)* **R-11's measurement is owed now, not later — and R-11 is the one adopted ruling that does not end in a decision.** Option B is *measure first, then re-ask*: an instrumentation PR (counts only, no behavior change) reporting how often the out-of-inventory reading path is taken on the practice corpora, after which **R-11 is put back to Shay with the number in hand.** So the docket owes two things here, not one: the measurement, and the second ruling. Until the number exists, R-11 has been ruled and nothing follows from it — which is a state worth naming, because a ruled-but-inert item is indistinguishable from a settled one at a glance. It lands with the R-1/R-5/R-6 batch at the latest and may land sooner, since the instrumentation is small and independent. ### Two items the docket added that were not rulings diff --git a/docs/assessment/50-rulings.md b/docs/assessment/50-rulings.md index b23c5bb8..b3a9da12 100644 --- a/docs/assessment/50-rulings.md +++ b/docs/assessment/50-rulings.md @@ -9,9 +9,9 @@ **Also ratified:** the ADR-0252 §5 **NO-GO** verdict. **Execution order (corrected):** R-12 → R-7 → R-3+R-4 → R-9+R-2 → R-13 → R-8 → R-1/R-5/R-6/R-11. -**Executed so far:** R-7 (PR-3, PR-3b) · **R-12** (both ADR amendments + the §5 verdict banner + the §6 retirement-condition amendment) · **R-3 + R-4** (PR-5 — the flag register, the profile mechanism, the daemon's fourth flag) · **R-9 + R-2** (PR-11 — the soak's committed artifact, pinned digest, change-triggered cadence) · **R-13** (PR-12 — the Wilson re-count; 21 of 25 licences revoked) · **R-8** (PR-14 — the outcome-mix rule; the four expected licences dissolve under it) — all landed 2026-07-28. +**Executed so far:** R-7 (PR-3, PR-3b) · **R-12** (both ADR amendments + the §5 verdict banner + the §6 retirement-condition amendment) · **R-3 + R-4** (PR-5 — the flag register, the profile mechanism, the daemon's fourth flag) · **R-9 + R-2** (PR-11 — the soak's committed artifact, pinned digest, change-triggered cadence) · **R-13** (PR-12 — the Wilson re-count; 21 of 25 licences revoked) · **R-8** (PR-14 — the outcome-mix rule; the four expected licences dissolve under it) · **R-1/R-5/R-6/R-11** (`docs/specs/postures.md` — three posture statements with criteria, and R-11's interim gate withdrawn on measurement) — all landed 2026-07-28. -**Remaining:** R-1/R-5/R-6/R-11 (the four posture statements) + R-11's measurement and its second ruling. +**Remaining: none.** All twelve adopted rulings are executed. --- @@ -38,7 +38,7 @@ ## R-1 · CR-3 efferent action — deferred, or out of telos? -**Status:** **RULED A** — 2026-07-28, Shay (explicit adoption) · *deferred with the entry criterion named* · **Register:** G-12 · **Blocks:** nothing; leaving it silent is the cost. +**Status:** **RULED A — EXECUTED 2026-07-28** · *deferred; criterion recorded as **P-1** in `docs/specs/postures.md` — no efferent surface until the chooser exists AND an efferent falsification bench exists* · **Register:** G-12 · **Blocks:** nothing; leaving it silent is the cost. **The question.** CORE's telos ends at articulate/learn/replay. AGI-grade generality ordinarily implies acting on the world. No system-level statement exists either way. @@ -130,7 +130,7 @@ Two flags' documentation describes a production profile the production profile d ## R-5 · The identity-enforcement authorization bar -**Status:** **RULED A** — 2026-07-28, Shay (explicit adoption) · *discrimination bar, corpus and floor named in advance* · **Register:** G-11 · **Blocks:** nothing; it makes an honest posture *stay* honest. +**Status:** **RULED A — EXECUTED 2026-07-28** · *bar recorded as **P-2** in `docs/specs/postures.md` — held-out benign/adversarial corpus, separation on the certified metric, floor named BEFORE the run* · **Register:** G-11 · **Blocks:** nothing; it makes an honest posture *stay* honest. **The question.** `identity_wave_gate` is off and `identity_action_surface` is off and documented as **"NOT authorized for live activation."** That is a deliberate, well-reasoned posture. What is missing is the *criterion*: no document states what evidence would authorize live refusal. @@ -149,7 +149,7 @@ Two flags' documentation describes a production profile the production profile d ## R-6 · Non-text ingest — entry criterion, or explicit deferral? -**Status:** **RULED A** — 2026-07-28, Shay (explicit adoption) · *deferral with the falsification bench as the standard* · **Register:** G-17 · **Blocks:** nothing. +**Status:** **RULED A — EXECUTED 2026-07-28** · *deferral recorded as **P-3** in `docs/specs/postures.md` — a modality enters serving on the SAME falsification standard as text* · **Register:** G-17 · **Blocks:** nothing. **The question.** 59 sensorium modules exist and reach no serving path. Projection heads do not exist. The position paper is honest about this. There is no entry criterion and no deferral ruling — so it reads as neither built nor deliberately postponed. @@ -263,7 +263,7 @@ Two flags' documentation describes a production profile the production profile d ## R-11 · An interim defensive gate for the fabrications? -**Status:** **RULED B** — 2026-07-28, Shay (explicit adoption) · *measure the out-of-inventory rate first, then re-ask* · **Register:** G-2 · **Blocks:** nothing; it is a posture choice while the ADR is pending. +**Status:** **RULED B — MEASURED, AND OPTION A WITHDRAWN 2026-07-28 (DELEGATED second ruling)** · *the measurement dissolved the premise: `atom_fact`'s template is `{p}`, a universal match, so "outside the verified inventory" is not a well-defined property and the proposed gate would refuse nothing while looking like a safety mechanism. **C** is operative; the fabrication ADR inherits the boundary question. See **P-4** in `docs/specs/postures.md`.* · **Register:** G-2 · **Blocks:** nothing; it is a posture choice while the ADR is pending. **The question.** The fixes are held for their ADR. In the meantime CORE *serves* readings it fabricated — `member(every_dog, mammal)` from `every dog is a mammal`, and `asserted(furthermore)` recited back as a premise. Is there an interim posture that is neither the held fix nor the status quo? diff --git a/docs/assessment/README.md b/docs/assessment/README.md index a280a18f..cf80026a 100644 --- a/docs/assessment/README.md +++ b/docs/assessment/README.md @@ -19,12 +19,17 @@ A read-only, evidence-bearing assessment of CORE's cognitive-cycle design versus | [`40-assessment.md`](40-assessment.md) | 5 | The synthesis | | [`50-execution-plan.md`](50-execution-plan.md) | 6 | **The execution plan** — five waves + five frontier tracks over every G/H entry, with the dependency gates and the risks. §0 carries **nine** corrections to the assessment found while sizing and executing it; **§2.1 carries the adopted ruling docket and its execution order** | | [`50-rulings.md`](50-rulings.md) | 6 | **The ruling packet** — R-1…R-14, each with evidence, options, a recommendation, and the exact diff that follows from each choice. **RULED 2026-07-28: twelve adopted, R-10 and R-14 stricken, §5 NO-GO ratified.** Carries the **standing delegation** under which residual sub-questions are decided, and marks each ruling `EXECUTED` as it lands | +| [`../specs/postures.md`](../specs/postures.md) | — | **The posture statements** (R-1/R-5/R-6/R-11) — efferent-action deferral, the identity-enforcement discrimination bar, non-text-ingest deferral, each with the criterion that would change it; plus **P-4**, the interim fabrication gate withdrawn because measurement dissolved its premise | | [`../specs/flag_register.md`](../specs/flag_register.md) | — | **The flag register** (PR-5, R-3 + R-4) — all 32 `RuntimeConfig` booleans by class, governing ADR, and *what evidence flips it*; profiles as the unit of decision; §5 indexes every declared table in the repository and the pin that makes each true | **Maintenance contract** (from §8 of the synthesis): a card whose `verified_at` falls behind a load-bearing arc is testimony, not evidence. Update cards when their subsystems move, or this directory becomes the next dead instrument it was built to replace. -**Execution status (2026-07-28).** Phases 0–6 produced the evidence; the arc that followed executed everything in it that needed no ruling, **Wave 0 closed** (twelve rulings adopted, two stricken, the §5 **NO-GO ratified**), and the docket then began executing in its corrected order. Landed: Track A's §5 verdict (pre-registered), PR-4 (+ pin 3), PR-6, PR-7, PR-9, PR-1, PR-3 + PR-3b (**R-7**), **R-12** (two ADR record amendments + the §5 verdict banner + the §6 retirement-condition amendment), **PR-5** (**R-3 + R-4** — the flag register, the profile mechanism, and the daemon's missing fourth flag), and four defect fixes (H-13, H-8e, G-22, G-23). +**Execution status (2026-07-28).** Phases 0–6 produced the evidence; the arc that followed executed everything needing no ruling, **Wave 0 closed** (twelve rulings adopted, two stricken, the §5 **NO-GO ratified**), and **the entire adopted docket is now executed** in its corrected order — R-7 → R-12 → R-3+R-4 → R-9+R-2 → R-13 → R-8 → R-1/R-5/R-6/R-11. Landed: Track A's §5 verdict (pre-registered), PR-1, PR-3+PR-3b, PR-4 (+pin 3), **PR-4b**, **PR-5**, PR-6, PR-7, PR-9, **PR-11**, **PR-12**, **PR-14**, **R-12**'s ADR amendments, the **posture statements**, and five defect fixes (H-13, H-8e, G-22, G-23, plus pin 3's own union blind spot). -Closed: **G-6, G-7, G-8, G-9, G-15, G-22, H-6, H-7, H-8 (all five instances), H-11, H-13**. Added: **N-8, N-9, G-21, G-22, G-23, H-13, H-14**. **Wave 4's F-6 half-gate is lifted.** What remains is **owed work, not open questions** — see [`50-execution-plan.md`](50-execution-plan.md) §Status for the board and **§2.1 for the adopted docket and its execution order**. +Closed: **G-5…G-12, G-15, G-17, G-19, G-22, H-1, H-6, H-7, H-8 (all five instances), H-11, H-13**. Added: **N-8, N-9, G-21, G-22, G-23, H-13, H-14**. **Wave 4's F-6 half-gate is lifted.** The gate is now four steps (smoke + warmed_session + deductive + teaching). + +**Three results worth reading before the registers.** (1) **ADR-0252 §5 returned NO-GO** against a pre-registered criterion — the geometric structure-mapper is refuted for its embedding class, and §6's organ-retirement condition was amended because a NO-GO made it unsatisfiable as written. (2) **Served capability shrank on purpose**: the Wilson re-count took deduction from **25 licensed shape-bands to 4**, and the curriculum outcome-mix rule dissolved the four licences N-5 predicted — in both cases `wrong` stayed 0, so no answer changed, only the claim attached to it. (3) **R-11's instrumentation dissolved its own question**: nothing is syntactically outside the verified inventory, because one of the 19 constructions matches any string — so the fabrications are mis-reads of admissible surfaces, and the fabrication ADR inherits the boundary question. + +What remains is **the next arc, not this one** — see [`50-execution-plan.md`](50-execution-plan.md) §Status and §2.1. **Standing note:** the PR #138 fabrication findings appear throughout as *measured & pinned, fix held for ADR + ratification* — recorded, never re-discovered, never fixed here, per explicit instruction. diff --git a/docs/specs/postures.md b/docs/specs/postures.md new file mode 100644 index 00000000..f074c590 --- /dev/null +++ b/docs/specs/postures.md @@ -0,0 +1,76 @@ +# Posture Statements + +**Authority:** R-1, R-5, R-6, R-11 (`docs/assessment/50-rulings.md`), ruled 2026-07-28. +**What this file is for:** four decisions that cost nothing to make and something real to leave silent. Each names a **criterion** — the evidence that would change the posture — because an unstated boundary reads as an unexamined one, and a future reader cannot tell a deliberate limit from an accident. + +None of these flips a flag. Three record where CORE deliberately stops; the fourth records a proposal **withdrawn because measurement dissolved its premise**. + +--- + +## P-1 · Efferent action is deferred, with an entry criterion (R-1, ruled A · G-12) + +CORE's telos ends at *articulate / learn / replay*. AGI-grade generality ordinarily implies acting on the world, and no system-level statement existed either way. + +**Posture: deferred, in scope eventually, and nothing moves until the criterion is met.** + +> **Entry criterion.** No efferent surface beyond typed, trace-folded tool operators (ADR-0018) until **(i)** the chooser (G-4) exists and is governed, and **(ii)** an efferent falsification bench exists at the standard `docs/specs/` sets for M2. + +Both halves are load-bearing. Without (i) CORE would act with no principled account of *what it should do next* — the Candidate Register's open seat. Without (ii) it would act with no way to be shown wrong, which is the one thing this architecture never permits of a serving capability. + +*Why deferral rather than exclusion:* declaring efferent action out of telos is coherent and cheap, but it forecloses a scope decision on a system whose chooser has not been designed. Deferral costs nothing exclusion saves and preserves the option. **ADR-0211's prohibition on motor/efferent units in the falsification bench v1 is narrower than this and remains in force** — it is a bench-level rule, not the system-level statement recorded here. + +--- + +## P-2 · Identity enforcement stays scoring-only until a discrimination bar is cleared (R-5, ruled A · G-11) + +`identity_wave_gate` is off; `identity_action_surface` is off and documented **"NOT authorized for live activation."** That is a deliberate, well-reasoned posture. What was missing was the criterion, and a posture with no criterion decays into an unexamined default. + +**Posture: scoring without blocking, until live refusal is authorized by measured discrimination.** + +> **Authorization bar.** Live identity *refusal* requires a discrimination report on a **named, held-out** corpus of benign and adversarial traffic, showing a separation between the two on the certified metric, at a **floor named before the run**. γ_id is the only certified metric today; `identity_action_surface`'s thresholds are uncertified placeholders and ADR-0246 §6.3 shows it refuses benign and adversarial traffic alike on the declared placeholder frame. + +*Why the bar is stated in advance:* the pre-registration discipline that made ADR-0252 §5's NO-GO full credit applies here for the same reason. A refusal gate authorized on a floor chosen after seeing the numbers is not evidence, and identity refusal is a surface where being wrong is expensive in both directions — a false refusal is a denial of service to a legitimate user, a false accept is the attack succeeding. + +**Scoring-without-blocking stays honest only while the path to blocking exists.** This is that path. + +--- + +## P-3 · Non-text ingest is deferred, with the falsification bench as its standard (R-6, ruled A · G-17) + +59 sensorium modules exist and reach no serving path; projection heads do not exist. The position paper is honest about this. It was neither built nor deliberately postponed — which reads as drift. + +**Posture: explicitly deferred. Not a gap; a decision.** + +> **Entry criterion.** A non-text modality enters serving when it meets the **same falsification standard as text**: a named held-out corpus, a falsifiable predicate set with holds/bites pairs, and `wrong=0`-or-refuse on the serving path. No modality is admitted on the strength of the substrate being able to represent it. + +*Why this criterion and not a roadmap:* the sensorium's existence is not evidence that it works, and the arc that produced this file spent its length distinguishing *built* from *proven*. Naming the bench as the standard makes non-text ingest earn its way in on exactly the terms text did, and makes the 59 modules a **capability awaiting evidence** rather than an unexplained absence. + +--- + +## P-4 · The interim fabrication gate is WITHDRAWN — measurement dissolved its premise (R-11, ruled B → second ruling) + +R-11 ruled **B: measure first, then re-ask.** The instrumentation ran on 2026-07-28. It did not return a rate; it returned a reason the question could not be asked in that form, which is the more valuable outcome and exactly what an instrumentation step is for. + +**The proposal.** While the G-2 fabrication fixes are held for their ADR, serve an interim posture: *refuse to hold a reading outside the verified 19-construction inventory.* + +**What the measurement found.** + +| | | +|---|---| +| distinct serving-path inputs measured | **11,199** (deduction + curriculum practice corpora) | +| clauses, at the reader's own sentence split | **23,562** | +| clauses outside the verified inventory | **0** | + +That zero is **not** a clean bill of health, and reporting it as one would be the error this file exists to avoid. It is an artifact with a cause: + +> **One of the 19 verified constructions is `atom_fact`, whose template is `{p}` — a single slot with no literal anchor. It matches any string.** + +So **"outside the verified inventory" is not a well-defined syntactic property.** Nothing is outside it. The proposed gate would refuse nothing while appearing to be a safety mechanism — a mechanism whose failure state is indistinguishable from its success state, which is this repository's dominant defect class. + +**This reframes G-2, and the reframing is the finding.** `every dog is a mammal` → `member(every_dog, mammal)` is **not** the reader admitting an out-of-inventory construction. The surface is admissible; the reader assigns it the **wrong relation**. Likewise `asserted(furthermore)`. The defect is in the **mapping**, not in the admissibility set — and no gate over the admissibility set can catch it. + +**Second ruling (DELEGATED, 2026-07-28): option A is WITHDRAWN as unimplementable as specified — not deferred.** A deferred option is one that could be built later; this one cannot be built at all against the inventory as it stands, and recording it as "deferred" would leave a future reader to rediscover why. Option **C** (no interim gate; the fixes stay held for the fabrication ADR) is the operative posture. + +**What the fabrication ADR inherits, stated so it is not lost.** It must define the admissibility boundary itself, because the verified inventory does not supply one. Specifically: either `atom_fact` is narrowed so that admissibility is decidable, or the guarantee is relocated from *admissibility* to *mapping correctness* — a check that the relation assigned to an admissible surface is the one that surface licenses. The second is the shape the evidence points at. + +*Method note, recorded because the number nearly shipped wrong.* The first two passes of this measurement were both invalid — one matched templates against whole multi-sentence inputs (reporting 100% out-of-inventory, when every clause was in-inventory), the other misread the sentence-splitter's tuple and measured `"."` 23,562 times. Both produced confident, plausible, wrong numbers. The catch-all was found only by a **non-vacuity check** — asserting the matcher could still say *no* to `"most birds can fly"` — and it could not. A measurement that cannot fail is not a measurement.