WIP: native nonlinear processing and second-front source audit by amaybaum · Pull Request #910 · amaybaum/incompleteness

amaybaum · 2026-10-24T05:52:45Z

Native instruments: nonlinear processing, second-front route, q10 readouts and the whole-V read

Head: 3b5176d27aa4377a3354f29b5045375f0579d76b (pre-landing commits 57dede93cf2e10281e99262ce508321f09fa15af, bdfa054ed44acead2c8b1b6736c3f1d149b57352, fe24cb12e2fcdd1ec9a6e6bee92025fa11bfcda1 and 3b5176d27aa4377a3354f29b5045375f0579d76b on top of 47fe49ac).

One new directory, verification/programmes/oi-qm/reconstruction/native-y-continuation-20261024/: 430 files (429 in its top-level manifest), all added; nothing else in the repository changes. Scientific files are unchanged since 47fe49ac.

Results (each for its own construction; no OI-wide no-go, no full OI-to-QM equivalence)

Native nonlinear processing (native_gate_v13/). In seven small original-time families, actual fields contain nonlinear products of the retained old sign with actual pool operands. All 525 tested radius-two passive terminal pointers, and their complete same-site prefixes, fail to compute the acceptance flag. No huge-L or distributed-processing exclusion.
Actual-scale second front (native_route_v13/). An added original-time source E=(0,-78,78) restores 32 compatible common raw X pairs across the 24 accepted contexts; no jointly affine E/N controller selects one.
Fixed q10 readouts (native_decode_v14/). Over all 254 nonconstant readouts and all 2^36 jointly affine E/N laws, 78 readouts decode s on all 48 accepted rows; q10.A = s under 1456 laws. s is not singled out: for each affine function f of the known seeds, some law gives q10.A = x XOR f(k). q10 is a private cell under Native A, not a permitted read.
One tick into the whole-V read (native_vread_v15/). A hybrid test in the native_route_v13 geometry, neither contract adopted. NEGATIVE_FOR_CONSTRUCTION: two current-slice key cells of q9 are free and unseen by every permitted record through L+2, and V is a permutation of each, so no fixed readout of V equals s for every complement value. Conditional on the declared ready-cone, source and bank inventory being complete.

Each of these four packages keeps its plan (committed before computation), result and a separate-agent internal review. These reviews are not commissioned independent programme review, external validation or formal certification.

Replay (from the repository root)

D=verification/programmes/oi-qm/reconstruction/native-y-continuation-20261024
python3 -I -B $D/replay.py            # 60 groups, each normal and -O, byte-compared
python3 -I -B $D/replay.py --latest   # the 7 v13/v14/v15 groups, with the v13 and v14 reviews

At 57dede93 (Python 3.13.16, g++ 13.3): all 60 groups pass normally and with -O, the manifests verify, and the v15 mutant campaign reproduces mutants/OUTPUT.txt (32 of 32 caught). CHECKPOINT-CHECKS.json records this. Rerun at the head 3b5176d2 from an export of that commit: all 60 groups pass normally and with -O, plus the checkpoint replay, and every manifest verifies (top level 429/429, 50 nested). An independent pre-landing review accepts 3b5176d2, including a §A.42 audit: every commit and blob id the directory cites resolves, and the four off-main commits it names are provenance only.

Release gate, run locally at bdfa054e and on its scratch merge with main c91bf5dd: 20 of 21 steps pass in both (the later commits change only README text, a JSON string and manifests). lean-axioms fails only because no lake toolchain is available locally; the PR touches no Lean. CI's Mathlib bridge job runs the gate with the toolchain.

Related work and provenance

Architecture and finite approximation: States and dynamics: exactness, sourcing and finite-approximation audit (research record) #911. Complete-record composition: Composition and access: finite-depth composition contract, logical acceptance circuit and complete-record laws (research record) #912. Their results are not established in this directory. Wider open gates: OI reconstruction: remaining foundations, generalization and A2 gates #913.
The programme synthesis these tests start from is 2d555f17 on Research (WIP): the missing component for F2-reg, and measurement as an added condition #885's preserved branch research/missing-component-20261020 (closed without merge), cited as provenance only.
Off-repository archives (the combined handoff and the earlier native-Y handoff) are cited as provenance; parallel_v3's geometry and herald inputs are retained data whose generator is not in the repository (README.md).

Landing

Land as a merge commit (no squash or rebase). The directory's own commit ids, and the nine WIP: native nonlinear processing and second-front source audit #910 commits that Composition and access: finite-depth composition contract, logical acceptance circuit and complete-record laws (research record) #912 pins (f24a57a7 and others), resolve on main only that way.
Land before States and dynamics: exactness, sourcing and finite-approximation audit (research record) #911 and Composition and access: finite-depth composition contract, logical acceptance circuit and complete-record laws (research record) #912. Composition and access: finite-depth composition contract, logical acceptance circuit and complete-record laws (research record) #912's native-coupling-v5/PROVENANCE.json pins INTERFACE-CONTRACT.md at 1324d9a8 with the hash it has at 05397c79; Composition and access: finite-depth composition contract, logical acceptance circuit and complete-record laws (research record) #912 corrects that pin in its own landing.

🤖 Generated with Claude Code

https://claude.ai/code/session_013T6pfKuv5mi3MXtKjH2tZb

Add scoped continuation sources, combined report, independent review, and 13-group replay under the OI-QM reconstruction programme. Keep integrated native reset and broader transport review open. Establish forward checkpoint cadence; no merge, premise adoption, or native-round certification.

Retain all background terms via an exact unique-path permutation proof. Check the actual free-coordinate geometry, dense-background controls and incompatibility of a blanket cold extension. Independent review pending; qualify parity reasoning to the no-wrap cone on the odd torus.

Check every reserved-bank slack and isolate the six-source active-cone problem. Preserve all unknown inputs and old frontier randomness. The isolated M2 front cancels; populated-cone output and joint law remain open.

Accept the free-pivot permutation result and conditional six-source geometry within scope. Preserve author bytes and add independent controls. All 15 replay groups pass normally and with -O. Six-source output and complete joint record/carrier law remain unevaluated; no reset or certification.

…rogramme integration record Native instrument (design/native-instrument/): the independent review of 455eb09 (review/, verdict: accept with corrections) and its corrections: - the degree-12 flag is this threshold sampler's; any exact sampler needs degree >= 2 and lower-degree flags suffice; - the native part is the single read's transport and re-preparation at tick 2; the rest is register-level or supplied; - Q5 is 1024 runs (512 input assignments x 2 backgrounds); - Q6(c) holds for this layout; the permuted layout reverses R and Z; - the checker generates the Q6(a)/(b) verdict text from measured values and tests closure in Q6(b) by per-setting conditional-law comparison; its regenerated output changes only those lines, normal and -O identical; - PREREG's precedence over the code rests on its recorded timestamp. One reviewer output line not printed by its script is reproduced by review/coordinator_weight20.py. INTEGRATION.md: the programme index for #885. It names the checkpoints since 455eb09 by bundle sha256 or commit (#910 at 1c8c9e5), separates review levels, lists what is supplied and the open obligations, and sets the division of work with #910. PLAN.md's Status points to it. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01Y87wExth9RoUb3dtbumhTb

…eduction Retain origin/M2 interactions and actual selector parities. Reproduce all128 native and256 diagnostic cases; opposite-Y raw supports overlap at fixed X. Keep complete source, truth table, period proof and bounded controls. All17 replay groups pass; independent review pending.

…view reports retained The focused review of 1907963 (integration-reviews/INTEGRATION-REVIEW- 1907963.md) accepts it with corrections: - B1: the open Y obligation names the sourced outgoing encoding E(x,y,r,z) = ((x,1,y) at q_Y, (x,1,r) at q_R, (0,0,z) at q_Z); - the index rows carry their sources' caveats: the selector premises and the review-recommended factorization test, the one-time-pad reading of the native selection, the XOR-mask scope of 9/10 and the 7/30 permutation bound, the payload condition of the shell collapse, the native-X candidate's pending proof audit, the radius-185 audit's internal review and scope, #910's reviewer classed as internal, the reviewed commit of the native instrument; the legend defines external; - obligation 1 notes the internal proof audits; - native instrument: the README states PREREG's precedence as resting on its timestamp and places the re-preparation registers at the outgoing receivers; the REPORT marks review-derived values; the review's two replay outputs it cites are added. integration-reviews/ holds the independent review reports the index cites (selector, native selector, read-reset) and the review of the index. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01Y87wExth9RoUb3dtbumhTb

Independent signed-coordinate dipole, huge-integer controls, sparse native propagation and upper strip reproduce all retained slices and truth tables. Accept only this prepared endpoint obstruction; broader designs remain open.

Exact depth1..12 scan finds q7=(0,u,y), with raw diagnostic D=p from qY. Keep q2.A-only reuse distinction explicit. Initial correction under study; no clean full socket or reset claimed at this checkpoint.

A full nonlinear upper/source permutation selects initial (u,0,u), giving native output(0,0,y) atR+7 without copying unknown input. Retain qY diagnostic D=p and explicit q2.A-only failure. Delayed access and readiness remain supplied; no reset or general reusable late input socket is claimed.

…andoff Independent signed-strip/source audit accepts seed-only q7 correction within scope. All21 checker groups replay normally/-O with recorded outputs matched. Preserve arbitrary-input and q2 diagnostic limits, private prefix access, initial readiness and old-record resources. Reusable reset remains open.

…ting Retain native clean q7 result. Add exact full-cluster computation, bounded controls, explicit preparation scope and parallel equivalence obligation map. Separate-author native review pending; states and composition reviews linked.

…andoff Separate-author computation accepts actual q7 interface and one-source one-tick obstruction. Packaged author/reviewer normal and optimized replay passes. Link reviewed states/dynamics and composition work on #911/#912. Keep native reset, joint-history certification and full equivalence open.

…ndoff Packaged reset_v8 author/reviewer normal and optimized replay passes. Preserve preceding 23 groups without claiming a new full replay. Record reviewed stopping-source and native-X findings; reset and full history remain open.

…oute

Retain distinct author/reviewer scopes and the separately assessed ninth-tick conditional corollary. Packaged five-group normal/optimized replay passes; preserve prior 25 groups without claiming a new full run. Keep simultaneous calibrated reset, full boundary closure and A1 open.

amaybaum · 2026-10-24T15:47:30Z

Writer handoff (§A.43). On 2026-10-24 the owner transferred writer ownership of this PR and its branch draft/native-y-research-checkpoints-20261024 to the coordinating session session_013T6pfKuv5mi3MXtKjH2tZb.

At the handoff the branch head is 94a4e6ccf18845af6c2d37cd797da1082f575d65.

It stays as it is. No existing checkpoint is replaced, rewritten or force-pushed.
Earlier authors' reviewed results and immutable commits are preserved.

Previous writers: please make no further pushes to this branch. Before every push, the new writer fetch-checks the remote head and stops if it has moved.

New checkpoints are pushed only after an independent review has accepted them.

Generated by Claude Code

…nt sources Exhaustive exact decision over all 254 nonconstant q10 readouts and all 2^36 jointly affine known-seed E,N laws on the retained native_route_v13 table. 78 readouts succeed (7464 laws); the single field q10.A equals s under 1456 laws. Two independent solvers agree on every solution set; planted, brute-force, raw-pair reproduction and sign-erased controls pass. Post-hoc structure: no law with E identically 000; the other q10 fields carry s in every q10.A law. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_013T6pfKuv5mi3MXtKjH2tZb

…nd handoff Adds native_decode_v14 to replay.py --latest and the manifest (349 files), and corrects the handoff, obligation map, README and checkpoint record in place: fixed redundant q10 readouts are decided, q10.A carries s under jointly affine E/N, separate review pending. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_013T6pfKuv5mi3MXtKjH2tZb

Text-only correction of native_decode_v14/REPORT.md, CURRENT-HANDOFF.md, EQUIVALENCE-OBLIGATIONS.md and CHECKPOINT-CHECKS.json after the separate review of cd9390a: - 8 of the 14 nonconstant degree-one readouts succeed; the six failures (q10.B, q10.D, A XOR B XOR D and complements) are listed; - B and D carry s jointly; B alone is s-free in 8 laws (review); - s is not singled out: for each affine f of the known seeds, an f-dependent law gives q10.A = x XOR f(k) (review's computation); - q10 is a private cell under Native A, not a permitted read; no read, post-read carrier law or reusable interface is modelled; - "no affine law with E identically 000" replaces the second-front claim; - the sign-erased control holds by construction; - the stale review and plan-status lines are corrected. checks.py, result.json, PLAN.md and PROVENANCE.json are unchanged; both SHA256SUMS files are regenerated. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_013T6pfKuv5mi3MXtKjH2tZb

…view CURRENT-HANDOFF.md, CHECKPOINT-CHECKS.json and EQUIVALENCE-OBLIGATIONS.md now attribute the x XOR f(k) result and "B alone s-free in 8 laws" to the separate review; REPORT.md states the review is not yet retained and drops a repeated sentence; the handoff's review line is restated. Text only; checks.py, result.json, PLAN.md and PROVENANCE.json unchanged; both SHA256SUMS regenerated. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_013T6pfKuv5mi3MXtKjH2tZb

amaybaum · 2026-10-24T19:47:07Z

Checkpoint native_decode_v14, head 1c15797a2b9fa46fe7c30fc8d9da16826a1c2c7c. Fast-forward from 94a4e6cc; the owner authorized publishing it.

The plan, 36213be6, was committed before the computation.
The author group is 7f843073; cd9390a0 integrated it.
Text-only scope corrections followed in 96b92f39 and 1c15797a.

Result. Scope: the retained two-front q10 table, under jointly affine known-seed sources E and N, on the 24 accepted contexts. The table was regenerated from the native rule by the separate review, digest b2f32466….

Readouts: 78 of the 254 fixed readouts of the q10 triple decode the old sign s, over 7464 laws.
Degree one: 8 of the 14 degree-one readouts succeed. The single field q10.A equals s under 1456 laws. q10.B, q10.D and A⊕B⊕D, with their complements, have no law.
s is not singled out. The separate review found this. For each affine function f of the known seeds, some law depending on f gives q10.A = x ⊕ f(seeds). So the table carries the old input x with a seed offset chosen by the law; s is the case f = u ⊕ a.
B and D jointly carry s in every q10.A law. B alone is s-free in 8 laws (separate review).

Not shown:

No permitted read. q10 is a private cell under Native A; the only read Native A allows is the single whole-q9 read at L+2.
No read operation, post-read carrier law or reusable interface is modelled. After q10.A, the remaining fields stay correlated with s given the seeds, which is an untested hazard.
Nothing about other layouts, observer access or OI–QM equivalence.

Review. One separate-engine skeptical review, then two text re-reviews.

The first review accepted the computation and blocked the text. The first re-review blocked on attribution. The final re-review accepted 1c15797a.
checks.py, result.json, PLAN.md and PROVENANCE.json are byte-identical to cd9390a0. The replays, normal and -O, match. SHA256SUMS checks 349/349 and 5/5.

Not yet on the branch. The review files are not yet retained here; a separate review-retention commit, itself reviewed, will add them under native_decode_v14/review/. This is internal review, not the commissioned independent proof review and not certification.

Generated by Claude Code

…nology Adds the three separate reviews of native_decode_v14 under native_decode_v14/review/: round1 (review of 7f84307, with its checker, helper scripts and outputs, including mutant runs), round2 (re-review of 96b92f3 with its outputs) and round3 (re-check of 1c15797). review/SHA256SUMS lists all 26 files; the directory SHA256SUMS now covers them as well. Redaction: round3/REPORT.md line 21 named the reviewer's absolute session scratchpad path; it now reads "`rev910b/` in the reviewer's session scratchpad". Original sha256 8e349b808f4e1bf67d81985262ec77d234148d3ccee40aa4cbd98576075ee6c4, retained copy sha256 90bdb52893e027b06a9e755da2af6675570017ea1b7ff5033ae7bc06c9e5db1f. No other content changed. The status lines that called this review pending or not yet retained now cite its location, and "independent review" in the v14 checkpoint scope reads "separate review". No result, number or scope changes. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_013T6pfKuv5mi3MXtKjH2tZb

…lines Addresses the focused review of 7db93c2. The three remaining status sites now state that the v14 review is retained: EQUIVALENCE-OBLIGATIONS.md's opening (every path has a separate-agent internal review), and the replay sentence in CURRENT-HANDOFF.md and CHECKPOINT-CHECKS.json ("the v14 fixed-readout author group", the wording README.md uses). The REPORT.md sentence that cites the review is shortened and one paragraph is reflowed. The round 1 mutant runs' stderr is retained beside their empty stdout, so each kill (RuntimeError: control failed: <label>, rc=1) is on record. Redaction: the absolute scratchpad prefix of the reviewer's export tree is removed from each traceback's file path, leaving the repository-relative path; nothing else changed. Original sha256 -> retained sha256: m1_contexts.err a1b8cc580f0dab8dc343b1e4447db15fd63c3609bab0b2ecd44b560e30f2618e -> 2f38e67a2c875412d926687f9731ecb94bd0e1ff282b95ce7d9227ee1084a363 m2_sign_drop_u.err 555b240d20e09fe506a28dbd1973fa638dac0ba517bb882f370a94fd8e57aacc -> b9dbeedf3e3c25d5b0a30b6ecc303a6d037a3127cde0d1ac4cea9e3cf05c156a m3_solverB_offbyone.err 1ab0633b60b178f52049f9f1efa34029a54f56aa8647b1a9b157ca0f53fdc33e -> d298a89b44c9c0e13b2b2fa7c5f085e5826b149455d5e840263b65c7bcfd8427 m4_table_corrupt.err 2acf64f8c33003a59252c82244572714cc8ce9e4f1e0319be70519e58088b673 -> f92d8229e9138a5220c6d89afb2a88e7b2fa6903201413219daf8dbc18bb4d3f m5_readout_drop_s0.err 455a8363d0e1b65dea7ef2739ea2733bee29cee26e6898372092c0f7071a8710 -> 37c686edd3fdc6934ba58978bb16091327d0f7301e0a4d55f38e754f2e8dfb42 review/SHA256SUMS (31 entries) and the directory SHA256SUMS are regenerated. No result, number or scope changes. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_013T6pfKuv5mi3MXtKjH2tZb

Preregisters one native-tick test before any computation: whether the unchanged native update from q10 at L+1 into q9 at L+2 makes q10.A's content available at Native A's declared single whole-V read, under the native_route_v13 source contract, and what the permitted public record then reveals. Decision rule, controls K1-K6, invariant table and scope. No access rule, preparation change or premise is adopted. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_013T6pfKuv5mi3MXtKjH2tZb

…onstruction) Executes the plan of ead2c38. The unchanged native update from q10 at L+1 into Native A's declared whole-V read at L+2, in the native_route_v13 source geometry, is keyed by free complement cells: the -x key (-(L+2),S,77), Native A's F9 cell, and the +y key (0,S+(L+2),77), both unseen by any permitted record through L+2, permute V for every fixed remainder. No fixed readout of V delivers s for every complement value; under a uniform key V carries no information about q10.A. A third free key (0,-79,77) reaches public port 0 at distance 341. Delivery needs a preparation change (pinning the keys) or an access change (reading their q9 neighbors at L+1); neither is adopted. Controls K1-K6 pass; 21 of 21 mutants are caught on their own labels; normal and -O outputs are byte-identical. Post-plan changes (parity in the key lemma, a second background family, extra anchors) are recorded in PROVENANCE.json and REPORT.md. Directory manifest regenerated. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_013T6pfKuv5mi3MXtKjH2tZb

Addresses the separate review of 2c43d61 (accept with nits): - REPORT Result: privacy and independence conditional on a uniform free key; under the contract alone only (i) holds. - Verdict generator: correct key counts and plurals; port-record inference as a third route for the -y key; port-coupling caveat only without a uniform unseen key; no redundant "nonconstant"; q10 at L+1 not listed as a permitted record; previous-slice keys added to (iii). - K5: anchors compare (status, reason); native_route_v13/checks.py pinned and its geometry() lists read as literals and compared; ready margins recorded (-x, +y, -y at margin 2). - K3: countercontrol for every n; evaluators compared on all 8 lanes; previous-slice keys checked in 144 further worlds. - K2: dependency-set propagation through the rule, compared with the cone formula. - Previous-slice keys classified: -x, +y, -y free; +x, -z pinned; +z in the previous q10 ball. - 28 mutants (adds the reviewer's K5_E_moved), all caught. NEGATIVE_FOR_CONSTRUCTION is unchanged. Manifests regenerated. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_013T6pfKuv5mi3MXtKjH2tZb

Addresses the re-check of 74541cf (accept with nits): - R1: the generated verdict, Scope and not-established list state that the outcome is conditional on the declared ready-cone, source and bank inventory shared with native_route_v13 being complete (no record with a deadline beyond L+1), and that (ii) also needs a uniform free key. - R2: junction_v4's fourth selector (4T,P+S,0) and opposite reset pulse (0,S,-S-136) added to the source list with strict anchors; no key classification changes. - R3: the native_route_v13 geometry reader fails closed on any write to a tracked list outside the top-level statements it reads; eight mutated in-memory copies of the pinned file are refused. - R4: REPORT cites junction_v4's inherited independent-uniform complement prior for the hypothesis of (ii). 32 mutants, all caught. NEGATIVE_FOR_CONSTRUCTION is unchanged. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_013T6pfKuv5mi3MXtKjH2tZb

Wording only; no checker logic changes and the outcome is unchanged. - N1: the geometry reader is described as refusing the listed write forms on a tracked name, not every write; indirect writes are not detected and the hash pin of native_route_v13/checks.py is the guarantee (REPORT, PROVENANCE, checker docstring). - N2: the generated verdict says the junction_v4 prior supplies the uniform key where that inherited prior is retained; only verdict.statement in result.json changes. - N3: Scope paragraph re-wrapped. PROVENANCE lists each revision's result.json digest. mutants/OUTPUT.txt is unchanged. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_013T6pfKuv5mi3MXtKjH2tZb

…15 result Adds the separate review of native_vread_v15 under its review/: round1/ (review of 2c43d61, with the reviewer's independent verification script, the mutant generator and their outputs) and round2/ (re-checks of 74541cf, 96bfc56 and ca7d06d, with the independent inventory script, the mutant outputs, and each geometry-reader test as diffs of its changed checker and changed native_route_v13/checks.py). review/README.md maps the rounds and lists what was not kept (copies and replays byte-identical to committed files, generated mutants, unchanged copies of other repository files). review/SHA256SUMS lists all 41 files. The round 2 report is kept as REPORT.md.txt because it repeats a heading, which the duplicate check rejects in a .md file. Redaction: in the two injection-test diffs the reviewer's absolute marker-file path is replaced by MARKER; nothing else changed. Generated diff sha256 -> retained: round2/asttest/route_injection_ifexp/native_route_v13.checks.py.diff a1ff690fcdb132650e8c78cd4657bf81fea1eba7b24990813fc0268761969d26 -> ffece62bd66e1ca826d1e13561072afc17bf8336c92a7f1d36e2b4c819c318c2 round2/asttest/route_injection_call/native_route_v13.checks.py.diff 57da9222603390e04581c9b681d048dbd4b3252ea0d539c0e0e0583beff27192 -> b2e259c2a1e4081d88a2c390c149758917338edd769bf93a26e5903aff74e73e CURRENT-HANDOFF.md adds the v15 result in its scope (NEGATIVE_FOR_CONSTRUCTION for the one-tick route into the whole-V read; independence under a uniform free key; conditional on the declared inventory being complete; port-0 key recovery untested; no OI-wide no-go) and names it in the next-gate paragraph. EQUIVALENCE-OBLIGATIONS.md adds it to the #910 row and the author list; CHECKPOINT-CHECKS.json adds four native_vread_v15 keys. The directory SHA256SUMS is regenerated. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_013T6pfKuv5mi3MXtKjH2tZb

Addresses the review of 9cbeee5. review/ adds the three mutants the reviewer wrote by hand as unified diffs against checks.py at 2c43d61 (round1/indep/mutwork/), since the review's control and receiver claims rest on them and the kept generator does not rebuild them. It drops round 2's mut_out.txt and mut3.txt, identical to mutants/OUTPUT.txt at 74541cf and 96bfc56, and rev_mutants3.txt, identical to round2/rev_mutants_out.txt. review/README.md states this, maps the reviewer's paths, and notes that the two redacted injection tests pin the unredacted route file. review/SHA256SUMS still lists 41 files. Status wording: the EQUIVALENCE #910 row now says one tick gives no fixed readout equal to s, two current-slice key cells being free and unseen, in a hybrid test with neither contract adopted, conditional on the declared inventory, and drops port-0 recovery as a next test, since it would matter only together with a change determining the unseen -x and +y keys (the handoff says so). The handoff section adds the current-slice qualifier, the free previous-slice keys, "necessary, not sufficient", and the untested other reads or ports and carrier law; the checkpoint scope key adds the hybrid framing and "no OI-wide no-go". README.md and the handoff opening no longer count the native tests; the obligations list adds the v15 plan, the review-input sentence and the exclusion paragraph cover v15. Edited paragraphs are reflowed; the directory SHA256SUMS is regenerated. CHECKPOINT-CHECKS manifest_files_verified records the last replay-driver run and is left as it stands. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_013T6pfKuv5mi3MXtKjH2tZb

Addresses the re-check of 96aac0d. The checkpoint scope key, the EQUIVALENCE #910 row and exclusion paragraph and the handoff next-gate sentence now say "no fixed readout ... equal to s for every complement value", as REPORT (i) does. The review-input sentence says the v13 and earlier native reviews pin their six-file author input and the v14 review checks the manifest and plan hashes. review/README.md names strict_plus_E_moved's reported label and lists round 1's in/sg.json, a copy of parallel_v3/history_control/SOURCE-GEOMETRY.json, as not kept. Both SHA256SUMS are regenerated; edited paragraphs are reflowed. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_013T6pfKuv5mi3MXtKjH2tZb

…directory The status surfaces (README.md, CURRENT-HANDOFF.md, EQUIVALENCE-OBLIGATIONS.md, CHECKPOINT-CHECKS.json) no longer state #911's or #912's results or cite their commits; they link to those pull requests. The #885 synthesis commit and later draft are named as provenance on #885's preserved branch. Dated openers and dispatch/merge narration are removed; every scope condition is kept. Scientific files are unchanged. replay.py --latest adds native_decode_v14's review checker and native_vread_v15 (60 groups in all). README.md gives the replay commands from the repository root, the A1/A2 label note and the off-repository provenance of the handoff archives and the parallel_v3 input data. native_decode_v14/review: drop the two exit-status/timing files of the author runs (rc=0 and the run time, both stated in round1/REPORT.md; the outputs are retained) and the reviewer's replay log (its six output lines regenerate byte-identically at 47fe49a with replay.py unchanged since the reviewed commit; the dropped line is rc and run time). The mutant .err files the review's kill claims rest on are kept. A review README maps the reviewer's working directory to round1/. Manifests regenerated. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_013T6pfKuv5mi3MXtKjH2tZb

…rded CHECKPOINT-CHECKS.json records the full replay of 57dede9: all 60 groups pass normally and with -O against their recorded outputs (exit status 0, Python 3.13.16, g++ 13.3.0), with the manifests and reviewed input hashes; and the native_vread_v15 mutant campaign reproduces mutants/OUTPUT.txt byte for byte (32 of 32 caught). Manifest regenerated. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_013T6pfKuv5mi3MXtKjH2tZb

…ance notes README no longer says every package keeps a plan: the packages from reset_v7/ on carry one, and the five earlier packages (continuation/, parallel_v3/, junction_v4/, six_source_v5/, outward_v6/) carry none. Every package has a separate-agent review; continuation/'s is in parallel_v3/review/. It also states that --latest runs the v13 and v14 reviews, records the 180 s per-checker limit, and names which handoff file each parallel_v3 input comes from. CHECKPOINT-CHECKS.json gives the working directory of its replay commands. native_decode_v14/review/README.md records the round-3 redaction in place of a not-kept note. Manifests regenerated. No checker, result, plan, provenance or report changes; the full_replay record of 57dede9 is unaffected. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_013T6pfKuv5mi3MXtKjH2tZb

shared_v12/ is the interface contract, not a result package, and carries neither a review nor a plan. Manifest regenerated. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_013T6pfKuv5mi3MXtKjH2tZb

Landing audit of bafb38d (text-only findings): - F1: the handoff no longer restates #912's or #910's results; it links native construction (#910), composition (#912) and wider gates (#913) and says their results are not established here. - F2/F7: the stale replay and process paragraph is replaced by a package index with the replay statement (every checker reproduces its result.json byte for byte, normally and with -O; mutant campaigns are recorded per package). - F3: a2_depth_v7's REPORT.md restatement at dc21028 and 49e1a11 is noted as outside the retained review rounds, checker and numbers unchanged. - F6: the 2d555f1 snapshot on research/missing-component-20261020 is recorded as provenance; the claims rest on Lean at de6fe29 (on main) and their checkers. - F8: dated opener and process line removed; missing spaces fixed. No checker, result, report or manifest changes. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_013T6pfKuv5mi3MXtKjH2tZb

… handoff for landing README becomes the package index: each package's plan and review coverage, root-relative replay commands for every checker group, the mutant campaign and the manifests, and the provenance of the inputs. It records that native-coupling-v5/PROVENANCE.json pairs shared_interface_commit 1324d9a with the digest of shared_v12/INTERFACE-CONTRACT.md at 05397c7 (1a0b34eb; the file at 1324d9a is 95cd4d20), and that the contract the package rests on is the 05397c7 file. The provenance file is pinned by its review and is left as it is. #885's 2d555f1 and the off-repository native-X archive are stated as provenance. CURRENT-HANDOFF drops the dated opener, the restatement of #910 and #911 results and the process sentences, fixes missing spaces, and points to the README for the package list and replay. SHA256SUMS takes README's new digest. No checker, result, plan, report, review or package manifest changes. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_013T6pfKuv5mi3MXtKjH2tZb

amaybaum added 4 commits October 24, 2026 08:52

WIP: specify compatible initial parity extension with retained M2

4f24077

Check every reserved-bank slack and isolate the six-source active-cone problem. Preserve all unknown inputs and old frontier randomness. The isolated M2 front cancels; populated-cone output and joint law remain open.

amaybaum changed the title ~~WIP: native-Y research checkpoints and reproducible handoff~~ WIP: native-Y junction obstruction and six-source continuation Oct 24, 2026

amaybaum mentioned this pull request Oct 24, 2026

Research (WIP): the missing component for F2-reg, and measurement as an added condition #885

Closed

amaybaum changed the title ~~WIP: native-Y junction obstruction and six-source continuation~~ WIP: exact six-source native-Y junction and checkpointed research Oct 24, 2026

amaybaum added 3 commits October 24, 2026 09:47

Record independent actual-scale six-source review

bf16540

Independent signed-coordinate dipole, huge-integer controls, sparse native propagation and upper strip reproduce all retained slices and truth tables. Accept only this prepared endpoint obstruction; broader designs remain open.

WIP: find seventh outward qY operand readout under actual giant delay

6db981c

Exact depth1..12 scan finds q7=(0,u,y), with raw diagnostic D=p from qY. Keep q2.A-only reuse distinction explicit. Initial correction under study; no clean full socket or reset claimed at this checkpoint.

amaybaum changed the title ~~WIP: exact six-source native-Y junction and checkpointed research~~ WIP: clean native Y readout and checkpointed research Oct 24, 2026

amaybaum changed the title ~~WIP: clean native Y readout and checkpointed research~~ WIP: independently reviewed clean native Y readout Oct 24, 2026

This was referenced Oct 24, 2026

States and dynamics: exactness, sourcing and finite-approximation audit (research record) #911

Merged

Composition and access: finite-depth composition contract, logical acceptance circuit and complete-record laws (research record) #912

Closed

amaybaum changed the title ~~WIP: independently reviewed clean native Y readout~~ WIP: clean native Y readout and actual reset-boundary audit Oct 24, 2026

amaybaum changed the title ~~WIP: clean native Y readout and actual reset-boundary audit~~ WIP: reviewed native reset boundary and parallel equivalence handoff Oct 24, 2026

amaybaum added 7 commits October 24, 2026 11:21

Retain workspace-assisted native Y routing probe and exact controls

e46d30f

Establish conditional seven-port noise law for workspace-assisted Y r…

96e7b62

…oute

Construct native masked-sign continuation from actual q8 boundary

328b74b

Retain separate history audit and ninth-tick geometry corollary

46d5882

Retain independent masked-route replay and conditional-history corollary

844c908

amaybaum changed the title ~~WIP: reviewed native reset boundary and parallel equivalence handoff~~ WIP: reviewed native Y masking and conditional public-history law Oct 24, 2026

amaybaum mentioned this pull request Oct 24, 2026

Release gate: citation-roles, support ids reachable from the commit under check (#915) #916

Merged

amaybaum and others added 4 commits October 24, 2026 15:51

amaybaum and others added 10 commits October 24, 2026 19:57

amaybaum mentioned this pull request Oct 25, 2026

OI-BH-1: audit Hawking information, Page curve, and gravity implications #917

Merged

amaybaum and others added 4 commits October 25, 2026 09:31

amaybaum mentioned this pull request Oct 25, 2026

OI-PE-1: pseudoentropy, deterministic randomness and observer access (research plan) #923

Closed

amaybaum marked this pull request as ready for review October 25, 2026 16:10

amaybaum merged commit facb46a into main Oct 25, 2026
26 checks passed

amaybaum deleted the draft/native-y-research-checkpoints-20261024 branch October 25, 2026 16:10

amaybaum mentioned this pull request Oct 26, 2026

NP-AVG-1: does the native acceptance predicate's negative depend on the pointwise background convention? #929

Open

amaybaum mentioned this pull request Nov 4, 2026

evidence(#912): land the review record of composition-access, byte-identical, as author-posted #1105

Merged

Navigation Menu

Repository navigation

WIP: native nonlinear processing and second-front source audit - #910

WIP: native nonlinear processing and second-front source audit#910
amaybaum merged 70 commits into
mainfrom
draft/native-y-research-checkpoints-20261024

amaybaum commented Oct 24, 2026 •

edited

Loading

Uh oh!

amaybaum commented Oct 24, 2026

Uh oh!

amaybaum commented Oct 24, 2026

Uh oh!

Uh oh!

Reviewers

Uh oh!

Assignees

Labels

Projects

Milestone

Development

Uh oh!

1 participant

Conversation

amaybaum commented Oct 24, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Native instruments: nonlinear processing, second-front route, q10 readouts and the whole-V read

Results (each for its own construction; no OI-wide no-go, no full OI-to-QM equivalence)

Replay (from the repository root)

Related work and provenance

Landing

Uh oh!

amaybaum commented Oct 24, 2026

Uh oh!

amaybaum commented Oct 24, 2026

Uh oh!

Uh oh!

Reviewers

Uh oh!

Assignees

Labels

Projects

Milestone

Development

Uh oh!

1 participant

amaybaum commented Oct 24, 2026 •

edited

Loading