Skip to content

Dev - #3

Merged
a9lim merged 8 commits into
mainfrom
dev
Aug 4, 2026
Merged

Dev#3
a9lim merged 8 commits into
mainfrom
dev

Conversation

@a9lim

@a9lim a9lim commented Aug 4, 2026

Copy link
Copy Markdown
Owner

No description provided.

a9lim and others added 8 commits August 3, 2026 15:38
qeval grows run_traced (per-leaf root-to-leaf Effect paths: New/H/T/Cnot
with qubit+epoch, Meas with outcome; beta-stuttering erased by
construction). New bin qselfint: quote + check harness, full 4..=24
population — quantum effect-trace 19,014 verified / 34 unresolved skips /
0 mismatches (independent qvm endpoint cross-check per run), pure-NF
19,014/34/0, witness45 reproduced, poisoned-seed-env canary. Five pinned
tests. DESIGN-QBLC obligation 2 records the measured constant: tight
within the intL protocol, optimality open, bisimulation stays the proof.

Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
…ly n=53

New bin qradical: lambda^5-idiom slice ({h,meas,new,t} all-mentioned
filter, count cross-checked exactly against Codex's independent DP at 53),
exact Z[omega] per-size aggregation with sqrt2-coefficient extraction,
per-program mass-conservation assert, FATEDIV accounting (per-program
Sigma-halt-mass irrationality), optional [beta] [trans] args.
enumerate.rs grows split_tasks_at (seeded-root task splitting; the
lambda^5 slice runs under a fixed 10-bit prefix) + coverage test.
qvm pins P53 = first_fate_divergent_nondyadic_witness_at_53.

Measured 46..53: per-size sqrt2-coefficients exactly 0 through 52, then
-1/4 at 53 — fatediv = 1, the unique program bit-exactly P53; unknowns
at 53 (752) are beta-insensitive at 8x/64x budgets.

Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
… next

Three ledger entries for the day: E_q = intL I (176 bits) measured and
effect-trace verified; the swap-involution broken by Codex's P53; the
radical-aggregate census landing the idiom-sector threshold at exactly
n=53 with P53 the sole fate-divergent survivor. AGENTS docket updated:
phase-2 primitive-taint over the non-lambda^5 complement is the single
remaining blocker on the full-population claim.

Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Spec form of qBLC proof obligation 2 (Codex skeleton, rounds 2-3 on
thread qblc-selfint). World-indexed generated relation R_k with runtime
diagonals and the full synchronized frame grammar; binary Meas
transitions; divergence-sensitive clauses; case ledger with C-DESCEND
as the opening rule; L1 extensional over implementing closures (the VAR
closure carries a dead wire suffix); pair vs cons' separated; L3 as a
full weak-head/ArgView decomposition; L4 readback collapse closing the
gap to the measured bit-exact pure-NF identity; base theorem
signature-free with a compatibility rule so the pure-NF corollary
actually follows. All obligations open; measured evidence indexed.
Round-3 ratification conditional on six corrections, all incorporated.

AGENTS docket: pickup order recorded (bisim L1/L2 proof lane first,
then phase-2 taint; complete taint design in thread qblc-omega-
witnesses round 2).

Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
…e protocols

By-type layout picked by a9: docs/ takes DESIGN-BLC, DESIGN-QBLC,
SPEC-BISIM, LEDGER (names unchanged); data/ takes the canonical
outputs with the 41-bound suffix dropped (range lives in-file and in
AGENTS; n=42 regenerates in place with zero path churn); scripts/
encodes the standing protocols (spot-check, census-regen with kill
subtraction + identity check, recert-kills at 4x + certlean + lake,
solomonoff-regen) — CI's spot-check job now calls the script.
Sub-labs under tools/ deliberately intact; src/ left flat (module
regroup = public-API break on the published crate). Ledger entries
before 2026-08-04 keep historical paths. Verified: fmt, clippy -D,
cargo test --release (17/17 green), spot-check 4..32 bit-identical,
recert smoke 25/25 byte-identical, regen smoke reproduces the
canonical frontier slice exactly, lake build Certs green.

Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
README picks up the two newest qBLC headlines (E_q = intL I = 176
bits; idiom-sector dyadicity threshold at exactly n=53) and a
current-only roadmap. DESIGN-QBLC's header no longer says S1 is
cleared to build (S1-S3 core landed) and the irrationality open
question records the measured n=53 layer plus the phase-2 route.
DESIGN-BLC drops the crate-name question resolved by publishing as
blam. AGENTS: qBLC docket compressed from staging narrative to
current-state form (engines / measured state / next), live state
dated 2026-08-04, spot-check bullet points at the script, gaslamp
thread list completed. Ledger entry extended; historical entries
untouched.

Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Arms release.yml to publish on the next main merge.

Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Install -> library tour -> drivers -> verification, with the results
list collapsed into one Selected-results paragraph pointing at docs/
and data/ via absolute links (the crate tarball ships neither). The
two new usage snippets are committed as runnable examples
(examples/normalize.rs, examples/bell.rs — the Bell program is 41
bits, the census's entanglement threshold, amplitudes exact) so the
CI clippy/test bar keeps the README code honest.

Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
@a9lim
a9lim merged commit 156589d into main Aug 4, 2026
14 checks passed
a9lim added a commit that referenced this pull request Aug 9, 2026
…red literally; W healed at {halt1:1}

Fresh audit #3 found the v1.10 fire arm computed the P/Q spectator
split and discarded it (KD from all of RS, rs'=()), yielding a real
isometry countermodel (retained-Q columns collapsed to inner product
1) and explaining W: the registered placement limitation measured
the defect, not the certificate language. Under the literal register
transition W is fully certified — four boundaries, inner popping the
inner coin's two instances, outer frames retained as spectators —
computing its hand ideal {halt1: 1}. Staged-uncomputation
inexpressibility RETRACTED; the consult prediction ledgered wrong
last round was right about the design.

v1.11: split wired through; all twenty certificates re-frozen as
exact dicts (most boundaries pop nothing; q family/B/W pop exactly
the interfering instances); W9 live-ticket key uniqueness (the
duplicate-ticket algebra row); validate(None) contract gap closed;
two new permanent regressions; six honesty corrections (B/W
attribution, 20/17 fragment, transposed conservation counts,
semantic_coverage doc, audit counts, identity-sweep retraction).
All predictions written first, all six held. PASS re-claim gated
on fresh audit #4.

Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
a9lim added a commit that referenced this pull request Aug 10, 2026
… language

Audit #15 (FAIL) confirmed the v1.22 deltas and forcings, then
found the first raw-subtype unitarity break since audit #3:
omitting empty KD bundles made storage histories non-prefix-free
(a fresh decode and a carried-in record collided on identical
targets at norm 2), a BULLET in the log passed every invariant
and collided via bt1, the lambda-IAM slice equation was
unchecked, open terms crashed binder_path inside WF, and the t
boundary raised from WF states. v1.23: the certified fire emits
its bundle unconditionally — ('KD', ()) when empty — restoring
prefix-freeness (section 7.4 restated); t-unimplemented typed
scope fence; W0 gains the log sort, the slice equation with a
bound-finding binder walk, and the closed-term conjunct;
regressions eighteen and nineteen gated. The written-first hinge
prediction held: suite, certify, typecheck, polarity
byte-identical across a change touching 39 of 58 reachable
certified fires — no reachable interference crosses fire
histories. Three regression mechanizations updated and two
fixture lps corrected, all disclosed, theorems unchanged. Fresh
audit #16 launched; the PASS re-claim stays gated on it.

Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant