Conversation
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Pre-implementation, pre-ratification. Quantum control, storeless:
runtime state in l2(machine configurations), {h,t} constants via the
invocation idiom, classical prefix-free syntax unchanged. Records the
bare-term isometry counterexample, invariant-sector halting with ticks,
the forced-synchronization argument, the coherence-block structure of
the output operator, and the minimal-garbage theorem as the gating
work item. Adversarial review thread: qalc-architecture.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Codex verdicts (thread qalc-architecture): monotone-mass lemma, bare-term counterexample, and full-logging degeneration confirmed; two claims corrected. (1) Ordinary KN control is irreversible — Var transition discards the selecting path, reachable witness (\x.x) A vs (\y.(\x.y) B) A — so the configuration basis needs a new reversible machine with orthonormal-column transition table; KN is guide, not substrate. (2) Equal-time-only coherence is a convention (common-origin tick, now normative [design]), not a consequence of injectivity. Also: normative halted dynamics id (x) id (x) shift makes 4.6 a theorem; effect-free fragment repaired against rigid-atom reduction (\h.\t. h h counterexample); clean coherent compilation named as gating universality obligation; garbage-transparent delta as contract; H-NOT-H witness added; D-circuit-shaped dyadicity with parity-sharpened witness reading; wandering-subspace formulation. Ratification gates listed in section 9. Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Round 2 verdicts folded into architecture.md: arrival-time nuance refuted and replaced (fixed output-dependent entry offsets suffice for unequal-time coherence); garbage-transparent delta refactored into clean delta fibres (gate (x) J, boolean-independent injective spectator transition); halt entry typed (RunDone vs Halt, terminal control frozen, no unticked halted state); effect-free lockstep weakened to projected semantic lockstep; clean compilation now requires synchronized transition count, garbage, and terminal control; fast-engine contract hardened to exact transition-system equivalence or isometric intertwining (configuration algebra is semantic state); h-only fragment defined semantically; stale 'dissolves' ledger bullet repaired; duplicated section-7 lead-in removed; minimal garbage scoped to local predecessor-fibre minimality. New machine.md: pre-formal reference machine sketch from round 2 — environment/closure substrate, Cfg record with (p, tau) spectator labels, central residue rule with coherent push/pop, transition audit table, the two cautions (no allocation identity in basis, no destructive output sink), and the formalization checklist that gates src/qalc/. Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
architecture.md rewritten as a clean durable contract — review-round narration and bracket tags removed, all post-round-2 content preserved; history moved to a ledger entry (2026-08-09) that records both review rounds, what broke, and the incorporating commits. machine.md header trimmed to current-only with a ledger pointer. AGENTS.md gains the fourth-pillar paragraph (design-contract-only, no qALC code before the section-9 gates pass) and the qalc-architecture thread. STATUS.md gains the Quantum-algebraic state section with the ordered docket: sign-off review, reversible machine, clean compilation, then src/qalc/. Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
(1) Clean delta fibres are now gate-indexed: U|q,b,k> = sum (Q_q) |b', J_q(k)> with J_q+ J_r = delta_qr I and orthogonality against every non-delta range — a landing shared across gate kinds overlaps H and T columns at 1/sqrt2 despite orthogonal sources. machine.md section 6 now demands the full pairwise orthogonality matrix. (2) The sqrt2-witness claim is demoted to its correct finite-approximant form: nonzero sqrt2 coefficient in mu_p(tau) at finite tau implies an opposite-parity coherent merge at tau; the limit-level reading is false (halting on dyadic masses gated by the computable bits of 1/sqrt2 gives mu = 1/sqrt2 with zero merging) and needs finite-support or uniform-D hypotheses. (3) machine.md's configuration type is now closed under its own transitions (Run/RunDone/Halt/Error union) and (p, tau) are evaluator indices outside the semantic basis — uniform over every reachable support, so in-basis they distinguish nothing while timestamping every predecessor fibre; halted factorization stays tick-only. Nonblocking: STATUS date and full doc paths; ledger and section-1 phrasing no longer outrun the open clean-compilation theorem. Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Round 4 verdict on 27f2c4b: SIGN-OFF, all three round-3 blockers verified closed. Post-verdict nits incorporated: 'transition' pinned to one reference semantic U-step (microsteps unobservable; the displayed intertwiner is time-preserving and does not license a different semantic clock — a different cadence needs a clocked dilation W U_ref = U_fast^r W or a synchronized simulation preserving halt ages, a stronger theorem); machine.md scopes the in-basis (p, tau) equivalence claim to the passive superselection lift only. Ledger records rounds 3-4 and the freeze; STATUS drops the sign-off docket item and marks the contract frozen. Next lane: formalize the reversible machine per machine.md section 6. Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
…s forcing machine.md rewritten from sketch to formal draft: full syntax (terms, NFs, closures with canonical trimming, levels, frames, residue tags, typed Config), the complete transition table (descent/binding, charged lookup rows with Look(i, env-with-hole) tags, constants, readback, clean delta fibres with rule-indexed content-free landing tags, species errors, terminal rows), reachability invariants I1-I5, per-rule inverses, and the backward-determinism framework with the clean range-disjointness pairs proved and the lookup-family crux isolated. Two findings from hand-computing the witnesses: (1) forcing theorem — any machine with permanent lookup residue fails the contract's H-NOT-H witness (the selection tag retains the unselected alternative, which no local pop can erase), so coherent popping is a correctness requirement, not an optimization; HH passes given the A6' boolean short-circuit. (2) linearity observation — with the linear NOT (lb.lx.ly. b y x) the selection tag is bijective with the returned NF in its frame, so a pop-at-Ret discipline restores coherence exactly for non-discarding terms: erasure costs coherence, Landauer surfacing as decoherence. Open crux: a forward+backward-deterministic pop schedule. Contract implication flagged: witness 7 should specify the linear NOT, or needs amendment via the review thread. Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
…ter, v1 fork Review verdict: DO NOT BUILD ON v0. machine.md updated honestly: universal forcing theorem retracted (LookFull(e, r) counterdesign gives injective, permanent, branch-independent lookup tags; the selector tag is also locally recoverable given a code decoder — the gap is provenance, not information loss); linearity conjecture retracted both directions (lb. b I I is linear but non-injective so the consumed bit must persist; generic pop-at-Ret merges N@1 with I@0 — the NF-tag bijection is code-relative and p is not a basis coordinate). Correct notion: operational injectivity of the computed map + synchronized code-aware cleanup — Landauer attached to semantics, not syntax. The narrow theorem stands review-confirmed: permanent Look(i, e-hole) residue in this table decoheres selector H-NOT-H absent code-aware cleanup. Bennett compute-copy-uncompute with a protected result zipper and true reverse mode recorded as the sound scheduling shape; witness 7 stands, to be pinned to an exact NOT wire term. Defect register: A2 vs canonical trimming (vacuous binders erase), A1/A2 range collision, A6' domain/range collisions, E1 drops (S, m), D3/R3 stale-tag collision, one-bit freshness non-reversible. v1 = coloured landing modes + explicit reverse control. Machine-defining fork 7.3 left open for ratification: minimal tags/earned coherence (A) vs symmetric redundant tags (B) — different canonical Omega_qALC. Ledger updated. Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
…ur discipline, Try-boundary protocol Full rewrite around three commitments answering the v0 review. (1) No-erasure forward machine: trimming abolished, envs never dropped — A2 injective by env-head pop, vacuous binders carry dead entries as coherence cost removed only by reverse execution; the only forward discard is the branch-independent Drop(c) at delta-fire. (2) Rule-coloured configurations: via on every Run, Join(predecessor colour) charged every step — ranges pairwise disjoint by colour, backward determinism structural, clean delta fibres realized by colour disjointness (J_h+J_t=0 is D1 != D2). Conservative charging means raw interference iff identical control path + equal content tags (HH survives raw); the minimality theorem becomes a monotone program of via-transparency lemmas, each enlarging measured coherence, ratified like the machine itself. (3) Try-boundary protocol: forward-copy-reverse across delta boundaries, with reverse running THROUGH interior deltas as adjoint columns — no code-supplied inverse needed, input retained by the frame so injectivity of the computed function is never required. Post-protocol branches differ only in result register and global timing: interference iff branch paths are transition-count-equal, deriving the architecture's common-T clean compilation from machine structure. Witness 7 passes for the path-symmetric NOT' (hand trace); v1 predicts the selector NOT also passes under the protocol (measurable, not a gate). Fork (A) ratified and built in; v0 defect resolution map and updated checklist included. Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
…st identifiers) Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Review verdict: scaffold holds (no-erasure A2 repair, Drop branch-independence, color+Join killing the v0 cross-rule collisions, clean fibres compatible with colors), Try-boundary protocol CONFIRMED-BROKEN: CP basis-copies a superposed result, CNOT(|+>|0>) = Bell, (H x I)Bell != |0>|+>, (H x H)Bell = Bell — the copy entangles with history, uncomputation cannot restore the entry configuration, and HH itself breaks (outer argument h 0-hat is not a literal). The input-retained Bennett argument fails as applied: Try retains syntax, not the branch input as a register. Sections 5-6 retained as the record of a productive failed construction. Registered: six residual v1-local defects, the conservative-Join contract drift (transparency set vs architecture amendment), and three unadjudicated v2 routes (classical-restricted Bennett + compiled-gadget witness; token-style no-copy transport; raw machine + code-side coherence). Do not build src/qalc on v1. Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
… updated token.md: exploration brief for the v2 token-transport machine — why tokens (IAM bideterminism may dissolve the injectivity war at the root; term read-only, no copies anywhere), the eight inherited guardrails from four review rounds (no basis-copying superpositions, global injectivity, typed sectors, timing physics, clean fibres, conservativity, witness battery, fork A), the six design questions (token state space, h-fibre on superposed positions, readback/NF story, exponential-context vs Look tags, synchronization, where Omega_qALC lives), and the required architecture section-8 amendment (token machine promoted from parked alternative to THE machine — thread ratification before drafting). machine.md 9.4 records the route choice; STATUS carries the v1 outcome and reordered docket. Written for a fresh-context continuation. Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
… compaction Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Architecture §8 pivot to IAM-lineage token transport, ratified through thread qalc-architecture (AMEND-then-RATIFY) with consequential edits: init as root query token over invocation sectors, isometric halted factorization NF x garbage x control x tick, observational strategy clause, generator-duplication constraint, delta-rules as fibre amplitude specs, negative witness as battery item 8, Danos-Regnier in the lineage. token.md grows from brief to working design: exact lambda-IAM table (bideterminism verified independently by the fresh feasibility thread qalc-token-machine, verdict promising-with-hard-open-problems), the delta-gadget sketch (probe protocol, unitary blocks on arrival states, balanced virtual answers, no-store consistency), eight-point isometry checklist, lambda-JAM exclusion, and the three-program formal kernel as the v2 gate. STATUS docket reordered; ledger carries both consults. Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
AGENTS.md fourth-pillar paragraph carries the token design and the qalc-token-machine thread. token.md drops its brief-era sections (design questions, applied amendment, v1 carry-over) and stands as guardrails / substrate / sketch / kernel-gate (renumbered 1-4). machine.md re-headed as historical record. Review-event narration trimmed from the architecture in favor of ledger pointers; cross-references updated throughout. Ledger carries the history. Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
kernel.md: full transition table over the lambda-IAM substrate (call, recall, arrive/fire, retrace, anshead/vb2/vvar, root, errors, ticks), step-indexed HH trace, results and findings registers. HH cancels to mass 1 on 0-hat at the outer fire; H-NOT-H passes the coherence bar through NOT'; the negative witness stays orthogonal in residue and time; column-Gram orthonormal on all five reachable bases. Findings: C1 refuted (re-interrogation is real; recall = ticket-replay, no store); selector NOT measured decoherent by one-step desync; error residues must freeze complete control (third recurrence). token.md and STATUS synced; ledger carries the method and iterations. Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Output alphabet {0-hat, 1-hat, I}: the negative witness's I output is
a valid NF, both branches Halt(I) with orthogonal residues and times
(the real witness-8 assertion; v1.0 misclassified it as error).
RunDone/Halt split into separate U steps, timing convention pinned.
Gram strengthened to structural amplitude-blind BFS (cancelled
targets included; bases 53-135, zero stuck, zero defects).
Output-density report added. Architecture: witness 7 pinned to the
exact NOT' wire term with the selector as a measured distinct
coherence class; stale leftmost-outermost bullet replaced.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Independent reimplementation reproduced all five results (substrate, fibres, HH, H-NOT'-H balance, negative witness HOLD), then broke the replay protocol: p* = \h.\t.((((h 0^) h) h) 0^) creates nested same-kind tickets and recall's b'-erasure aliases two reachable re-entry states (norm 3/2 at step 89, verified against both implementations; 3-coin regression shows the cross-time column collision). Re-entry determinacy refuted; verdict and repair directions registered in kernel.md par.9; token.md obligations and STATUS docket updated: reversible ticket cleaning is the docket head, p* and 3-coin are mandatory regressions. Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
recall no longer erases its discriminator: Run gains a transport-inert replay stack RS that only recall pushes and nothing reads; terminal entry freezes it into the garbage factor. The p* collision pairs land on RS-distinct targets. All seven programs pass (five witnesses bit-unchanged — coherent code never creates an R-frame; p* norm 1 with the first coin mandatorily decohered; 3-coin clean) and the structural Gram is zero-defect on all seven bases (p* 458 states). Re-review dispatched with the refute-this challenge: find a program where recalled branches may legitimately merge. kernel.md par.9.5 registers the repair. Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
The p* repair is independently confirmed; the gate stays open on two new findings. q = \h.\t. h ((((h 0^) E) N) 0^) (E = eta-expanded id, N = selector NOT) transports the coin injectively through selection (H;id;H) and refutes mandatory decoherence — replaced by the transparency criterion: a replay frame is cleanable iff its bit is a function of the live core, cleaned inside the delta block with provenance-matched instance-indexed frames. Beneath it, an alpha/ gamma transport totality failure: q's 335-state graph has three stuck states (verified bit-exactly). kernel.md par.9.6 registers the verdict; v1.3 gate set (transport totality, instance frames, first transparency/pop rule, q at mass 1 on 0-hat); q joins the mandatory regressions. STATUS and token.md synced. Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
…he time register Transport totality restored on the whole battery via the replay rule (consumed-ticket re-seeks rederive their selection from the instance's frame, read nondestructively; derived from the literal visit-3 trace). Instance identity = the invoking occurrence's logged position, already at the log head on every gate-leaf entry; tickets and frames carry it. Transparency/pop implemented as certified fire-boundary cleanup, Gram-verified, with a negative control. q-family halts frame-free. Mass 1 on 0-hat is blocked by a new standing finding: branch arrival offsets are odd and even-pad-invariant — pattern-encoded slots are time-free, step-encoded selection skews the branch clock by the transported bit. v1.2 basis drift owned (double-call subgraphs; p-star support 8 was bug-inflated, correct 4). Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
…sion corrections FAIL as formalization gate / PASS for the replay repair; all numbers independently reproduced; no odd pad in 156 variants, composed selections stay odd; port-polarity coloring supplied as the parity theorem route. Corrections applied: buried same-instance frames and missing-instance entries now typed errors, TOP fallback removed, dup copy-discrimination probe added (battery bit-identical). v1.4 gate: encoded fibres, certificate soundness beyond Gram, canonical certificates, the coloring proof. Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
…s; exact ring The coloring is a theorem: phi = depth + dir + weighted marks, uniform flip everywhere except fire (defect 1 − w(erased lp)); 1,518 edges verified zero-violation; branch-offset formula matches all measured offsets; the review's even-pad search becomes a corollary. Refined coherence condition: equal erased weights (HNH carries w=1 on both branches). New standing object: the weight-conservation conjecture — gate-free readback conserves w ≡ slot, gate-mediated routing is the parity-free class. Certificates canonical via discovery fixpoint (HAND==AUTO x11); validity strictly beyond Gram catches the negative control. Amplitudes exact Q[sqrt2]. Encoded-fibre amendment drafted. Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
…cture §7 Codex ratification round: construction sound; strengthened language applied verbatim — arrival-lp erasure subsumed (F_a-trivial case), seven-item per-boundary exhibition, source-fibre disjointness, pop as coherent decoding not deletion (negative-witness fence explicit), computability discipline (total certification over a proved over-approximation, conservative fallback, U never undefined). Gate item 2 discharged; six of eight v1.4 items closed. Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
…ncoded fire Review verdict split on the dynamics/proof line: theorem PASS-WITH-CORRECTIONS (gauge uniqueness, interior-fire parity law), table FAIL on C = λb.λx.λy. b y y — same-slot arrivals with distinct lps collided through the v1-lineage fire's unsound erasure. v1.5 fire: conservative decode by default (D(l) on the inert ks stack, uniform φ-flip), certified erasure under the corrected fibre condition. Twelve programs total/clean; HNH's coherence now earned through its certified boundary; Ccoll decoheres Gram-clean; conservation conjecture refined to mark-free ancestry with the reviewer's exhaustive small-term enumeration behind it. Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
…fragment Lemma A (attribution ledger, t = k+b+1 mod 2) + the S11.1 coloring give w = exit slot on mark-free fire-free readback; transfer corollary names every escape route (interior-fire charge, VB pattern birth, mark capture). Invariant-level mechanization: 41,272 closed terms, 55,727 surfacings, zero failures, lockstep conformance. Terminal chains, gauge-pinned uniqueness (4/256 = predicted orbit), discover_total (h-Omega conservative reject), guard-rule and (g,i)-aliasing sweeps all discharged. kernel.md S13 registers the PASS claim with typed fences, pending fresh-instance adversarial review. Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
…withdrawn The threadless sol-tier round confirmed Lemma A (independent reimplementation), the transfer closure (153k-argument search, zero mark-free escapes), the gauge orbit (15-parameter solve), and discover_total — and corrected the theorem's missing depth term (w = k + b + |pos'| - |pos|; deployed specializations unaffected). The PASS scope was refuted: Q = ((h 0) SEL SEL)(h 0) reaches buried-frame from init with err mass 1 under a valid certificate and clean Gram. Q registered as the thirteenth battery program; validate gains semantic_coverage; mu-kind fire collision closed typed; well-formed-state preservation registered as an obligation. PASS stands for the three-program gate and the twelve sectors; v1.7 = the interleaved re-seek. Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
…-claimed (audit-gated) Q's trace showed frames accumulate as duplicates: the frame is a RECORD keyed by instance, not a stack entry. recall idempotent (frame-conflict typed), replay deep keyed lookup, rs canonical, buried-frame dissolved. Twelve bit-identical; Q heals to pre-predicted physics; stress set (weave/hweave/qq) all predicted then measured. WF subtype: six invariants, reachable within WF over 9,314 states, both v1.6 collisions closed. Certified-erasure cargo conditions: gamma-free enforced (zero cert changes), alpha-cargo scoped as a compile-time obligation (HNH's coherence depends on it). Fresh-context independent audit gates the claim. Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
… W7; typed fragment Fresh audit #1 (sol-tier, threadless): physics CONFIRMED-SOUND, but two countermodels (WF fire collision; Pα post-erasure re-seek under an admitted certificate), an instance-key injectivity gap, a fuzzed duplicate fresh call via alpha-decode, and honesty defects. All reproduced locally before correction. v1.8 (kernel.md §15): certified erasure leaves branch-consistent decode records; the refire guard makes one-fire-per-instance a runtime machine invariant (v1.7's alpha-cargo compile-time obligation discharged and deleted); discover_total validates its fixpoint and refuses unsound certificates (Pα heals to {1/4, 3/4} via the conservative fallback); W7 fibre coherence + mechanized range disjointness; typecheck.py = the typed h-only fragment judgment (15/18; q itself untypable — qprime is its repair). The sixteen stay marginal-, support-, and basis-identical. Auditor's reproducer reruns clean; fuzz 250/250. PASS re-claim gated on fresh audit #2. Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
The solomonoff driver now computes Levin measure, certified Omega_speed brackets, Kt, depth0, and halter spectra under three time gauges (beta, transitions, honest) in the same 4..41 sweep, with unknowns charged on the upper side only at machine-recorded fuel-death floors. Machine records its counters at every capped exit; the ladder threads failed rung-2/rescue counters as telemetry. docs/classical/speed.md is the contract; data gains speed_floors.txt and the extended solomonoff outputs (existing sections byte-stable, zero unclocked halters, all canonical Kt/depth0 exact). Post-trim open mass: 5.926e-12 (beta), 5.113e-17 (trans), 5.576e-12 (honest) — the 1/t charge suppresses the unknown frontier by ~10^3 against Omega under beta. Codex-reviewed on thread blc-speed-prior; full bar green in both feature shapes. Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
speed_floors.txt holds one line per open program — unknowns and unclocked halters, distinguished by the kind column — not unknowns only. Codex editorial nit from the blc-speed-prior close-out. Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Two Codex review rounds on thread qalc-rust-pillar: nine ranked findings folded, then three specification fixes; no architectural blocker. Key decisions: Amp canonical-invariant wrapper over the existing quantum::scalar::Dw; PyReprKey scoped to the two state-identity sorts with a separate typed wire format; complete-carrier column fixtures as the primary differential pin with SHA-256 column commitments (hoisting the ckpt.rs core); PinnedSector boundary until phase-4 admission; dual-evaluator cross-oracle verified per-step on all twenty programs. Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
The ratified sketch's phase 0 (docs/quantum-algebraic/rust-pillar.md):
- src/hash.rs: the ckpt SHA-256 core hoisted into the library (pure
code motion; sha256_16 stays in cli::ckpt; vectors move along).
- blam::qalc::{term, mark, state, amp, wire}: the typed kernel-stratum
state grammar; PyReprKey scoped to the two state-identity sorts (RS
frames, KD keys) with byte-exact Python repr; Amp holding the
stored-canonical invariant over quantum::scalar::Dw with a strict
rejecting decoder; the v1 wire format with whole-file byte round-trip
and SHA-256 column commitments.
- qalc/export_rust_fixtures.py: stdlib-only exporter under the
dual-evaluator cross-oracle (Fraction/Dw full map equality asserted
at every transition of all twenty programs), deterministic
double-generation check, kernel dispatcher only (the Gate-2 shadow
exporter will be a separate process by design).
- tests/qalc/: complete tick-depth-2 carriers, exact unmerged columns
with commitments, trace digest chains, absorption finals, and the
repr golden corpus for the twenty-program suite (3.2 MB).
- tests/qalc_codec.rs: byte-exact round-trip, PyReprKey vs Python
reprs, commitment recomputation, canonical RS/KD/cert orders under
the Rust comparator, structural pins.
- qalc workflow: fixture regeneration byte-checked in CI.
Bar: both feature shapes green, clippy -D warnings clean, classical
spot-check 4..32 and qBLC per-size rows 4..32 bit-identical, exporter
--check byte-identical. STATUS/AGENTS/ledger updated.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
…entries) Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
src/qalc/kernel.rs ports the frozen reference's full transition surface (kernel.py step + dw_machine.py): the VB automaton, gate-leaf recall/replay/call with the v1.7-v1.24 guard set, the boundary fire arms (certified KD erasure, alpha decode with KA suppression, retain-whole), root arrivals, classical rules — plus exact edge coefficients, transactional-Capacity evolution in deterministic first-touch order, the complete-carrier BFS, and the native Gram check. Off-grammar states return typed Defects, never panics. Primary pin (tests/qalc_kernel.rs): every .qfx regenerates byte-identically from the Rust engine given only the program identity — carrier, exact unmerged columns, commitments, trace digest chains, finals. Plus the universal Done-tick row, zero Gram defects (basis = carrier size), exact T-phase finals, and the kernel.md §11 HH trace row for row (fixing that table's t=31 path typo, ffbbf -> ffbbaf, verified against the Python reference). The exporter grows the §6-promised Dw-only T0/T1/HTH0 family (fire-t1's ω is outside Q[√2], so they skip the Fraction cross-oracle and pin the reference's hand-computed finals): 23 programs across 24 fixture files. Full bar: 256 all-features + 220 plain tests, fmt/clippy clean, classical spot-check 4..32 and qBLC per-size rows 4..32 bit-identical, exporter --check byte-identical. Residual recorded in STATUS/sketch: guard arms unreachable in the 23 programs stay review-pinned until the phase-2 Gate-1 differential. Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
No transition-table drift found; four findings folded, each re-verified against the reference first: the key/bit traversals and the PyReprKey renderer are now explicit-work-stack iterative (the reference's v1.26 discipline — wire DEPTH_CAP bounds decoded input only, stepping nests past it), Epoch children are Arc-shared so recalls clone pointers as Python shares tuples, rs_insert implements the frozen stable sort of rs+(fr,) verbatim instead of assuming a canonical RS, and the wire decoder rejects duplicate certificate positions (Vec first-match must equal dict lookup). New pins: a 2^20 nesting depth test with iterative teardown, and a duplicate-cert rejection test. Coverage correction recorded in STATUS/ledger: rootval, alien-ticket, and refire are fixture-pinned after all; the review-only residual is exactly the error/guard arm family. All 23 fixtures still regenerate byte-identically. Bar: 258 all-features + 222 plain tests, fmt/clippy clean, classical spot-check 4..32 and qBLC per-size rows 4..32 bit-identical, exporter --check 24 files byte-identical. Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
The differential port of readback.py: RB/RBL grammar growth (RBL is a log-alphabet member captured into lp slices, RS-frame instances, and KS by ordinary transport — measured on all 30 Gate-1 cores; bare RB stays tape-only), the composed state stratum in state.rs (NF trees with Arc-shared children, zipper, binder marks, the four scope residues, terminal garbage, typed error kinds and garbage families), additive wire records incl. a generic cprogram fixture section with probes, the full composed dispatcher in readback.rs (vlam/vvar, head-gate, head-neutral-gate, head, BA-alphabet b1-b4, enter, return, rootdone, kernel delegation through the BA adapters, typed totalization to the host-fault landing), the four predecessor inverses, and the composed evolvers (carrier/columns with per-edge inverse checks, evolve trace, native Gram). tests/qalc_composed.rs pins the structural half of the Gate-1 differential: all 30 cores close at the reference's exact per-core state/tick-cut counts (aggregate 7,507 states / 7,417 columns / 90 cut), the exact 28-rule inventory, per-edge predecessor inverses, zero Gram defects, and absorption — first differential contact, green on the first run. Byte-level column/trace/final pins via the composed fixture exporter are the remaining phase-2 exit work. Design deltas sparred with Codex on thread qalc-rust-pillar (counts and placement claims re-verified empirically here): composed fixtures will live under tests/qalc/composed/; totalization normalizes CPython exception classes to a stable fault category (identity rides on the retained source, not the class name); one deliberate divergence — TapeEntry::Rbl is kept, since kernel bt1 genuinely transports RBL log heads and tape entries never enter the sort domain. 261 all-features + 225 plain tests green, fmt/clippy clean; phase-1 fixtures byte-identical throughout. Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
qalc/export_composed_fixtures.py writes tests/qalc/composed/ — 30 cprogram fixtures (deterministic BFS carriers, exact unmerged columns with commitments, qalc-ctrace v1 digest chains, absorption finals, totalization/fallback probes) plus a corpus growing the PyReprKey pins with RBL-carrying frames and bundle keys. tests/qalc_composed.rs regenerates every file byte-identically from the Rust engine alone; the codec battery's widened enumeration round-trips the composed tree. Probes byte-pin the noncanonical-rootdone fallback (terminal inverse verified natively), kernel-stuck (egc), rooterr (egk), and host-fault (egf) landings; egi stays structurally unreachable, review-only. Fixtures byte-stable under two PYTHONHASHSEED values. Kernel exporter gains the rbl/rb wire arms (mirror of wire.rs) and directory-tolerant stale detection; kernel fixtures unchanged. Docs, STATUS docket, and ledger record the phase close and the two sparred decisions (TapeEntry::Rbl retention, host-fault normalization). Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
1 participant
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
No description provided.