Independently verify the P104 forward caller and target bindings (no new κ)

research/p104-caller-independent-audit/CAPSULE-INVENTORY.json

@@ -0,0 +1,30 @@
+{
+  "schema_version": 1,
+  "files": {
+    "capsule/outputs/p104-caller-route-audit-20261025/CLOSEOUT.json": "cfaaf70e5e6efd120ba75e0fd39ce928d20d0b9dbb9b8afbe92f0c18fd0e878b",
+    "capsule/outputs/p104-caller-route-audit-20261025/check.py": "140fde975d37b8518a1332dc8cecedf571188d7db438629b1acc67c3b2571655",
+    "capsule/outputs/p104-caller-route-audit-v2-20261025/INPUTS.json": "e94c93d575f2dcce37bf3aedaeaf5f2200ccb4b14693dead84a7eeb5ad71a701",
+    "capsule/outputs/p104-caller-route-audit-v2-20261025/check.py": "0e5e0e3d56165437fc2821f41927e3f0cae838c60841a93e8b5cb46caeb7657c",
+    "capsule/outputs/p104-caller-route-audit-v2-20261025/sources/DATA-PHASES.json": "c5916ff312678f12aba403847643fbbc29f643acec60bdf836fa56cd3e8251fa",
+    "capsule/outputs/p104-caller-route-audit-v2-20261025/sources/native_inventory.py": "65539acaf45bdfe8c5aee6830f9c9b3d1107c442a60842ef803fdd7688de8f32",
+    "capsule/outputs/p104-contact-reassessment-20261025/RESULT-normal.json": "3c4d8db5a9695924027f5801901b55cefe0cdc5ea07450aaf396de7e5887e608",
+    "capsule/outputs/p104-contact-reassessment-20261025/sources/contact/RESULT.json": "b1dc682a4d9db5a726fbdf5c05c081f3c9ff9816829e5d8c900db1904a1f98c1",
+    "capsule/outputs/p104-contact-reassessment-20261025/sources/contact/SOURCES.json": "42a7f956bf450c15a92449e285bd2b355159cc99e78bc1d815a4b902221ab89f",
+    "capsule/outputs/p104-contact-reassessment-20261025/sources/contact/WORD.json.gz": "a1e416add9d2bb44fce923111ebc50513e5d845e56aabe214f2f0bc7d5f2b45d",
+    "capsule/outputs/p104-contact-reassessment-20261025/sources/predecessor/CONTROLLER.json": "a84f6bbd547379597464e56c420cc665f96bc43a9a1351eed4b7cd0fce3cc4b9",
+    "capsule/outputs/p104-contact-reassessment-20261025/sources/predecessor/caller.py": "300d40043953e4f6aecfde8f2d1004b8ad6dc5a79ec3c7666c9908c2466c91c8",
+    "capsule/outputs/p104-contact-reassessment-20261025/sources/predecessor/local.py": "878734dfd64c4b9d362d0ae476cc5c2a19fe93e48b7378763b23f3e0dd0a5a71",
+    "capsule/outputs/p104-native-contact-audit-20261025/RESULT-normal.json": "34299df26387b69f1a89cc52d52cae3add51f4e12c0715e0de38d42fab302c9f",
+    "capsule/outputs/p104-native-contact-audit-20261025/check.py": "43fd40847a0853c0ac511d58ce77ef822a5c6dd1b25f662f01cefaa3196a2f7d",
+    "capsule/outputs/p104-native-contact-audit-20261025/sources/clifford_exact.py": "2982ead07721e83358dc67c6f49c2b4c16a236a6bebdfd72f58221be1f5d0962",
+    "capsule/outputs/p104-native-contact-audit-20261025/sources/native_verify.py": "3049471905b47b385c88378a1a3be7a18b8858aadd035b8a6da2f4a03b470f38",
+    "capsule/swarm-price-plane-20261023T112711Z/coordination/run_guarded.py": "720153b7491acede2c66bdc2dfd3174653195fa3222ec3589d0671c406a5bad8",
+    "expected/caller.json.gz": "383509fce75e2d364138775a4fc96181c9c4cf2608eda0819a7c0c6236faa679",
+    "original-evidence/CLOSEOUT.json": "bf26c6d0b2c8a4ed2a81cc907723a21dbd5670bbac4e1674b710d6c76f60ea68",
+    "original-evidence/GUARD-RECEIPT.json": "c674c07970f972e40296dc67fc40daa6a50f0121d5bd9009beb50aaedd672426",
+    "original-evidence/PAIR-RECEIPT.json": "29983c28d00942cc9dfbd81ccb02d59f1e33e597157475acef0cc6fe2b45308e",
+    "original-evidence/REQUEST.json": "3282f112db2854a9e739491b956f8e623f92dadb5ccdbe99d48cdea3c641f319",
+    "original-evidence/TRACE-normal.jsonl": "ab9f0a43c487a49fee4c1c61dc1b23097aedd5fd3b73d93cea4231d4dcc8381c",
+    "original-evidence/TRACE-optimized.jsonl": "492aad6c8e7d843139af5891caf5b212a697e4138386eeadbf68b6b20cd19d12"
+  }
+}

research/p104-caller-independent-audit/HANDOFF.md

@@ -0,0 +1,33 @@
+# Seven-field handoff
+
+1. **Research discussion / PR / artifact:** this package and its PR conversation
+   are canonical for the forward caller/target check. PR5548 covers the earlier
+   contact/native/ceiling layer; no new κ or complete parent admission.
+2. **Exact commit:** base32890bc5de034ab8fc93ac88ead7279a6a416d04;
+   source18968423711fd902aaee1099979497d55603e1dc;
+   predecessor e2a6cce2f9db5a2c5f9714f4b9e5292dd3a10581.
+   Publication commit contains this package. Original checker/output and all
+   inputs are authenticated by CAPSULE-INVENTORY.json and original-evidence.
+3. **Established:** source-bound265-axis forward contacts/block coefficients,
+   full free-Laurent endpoint, native target adapters/normalizers, main aliases,
+   disjoint helper ownership, positive-query/idle recipes and unchanged invoice:
+   69,912 paid jobs and12,814 productive fields. Full original scientific output
+   matches in both modes. Scoped fault controls reject; no economic gain.
+4. **Uncertain:** untouched native paid maps/all-input/all-width foundation,
+   complete-caller raw inverse transpose, owned erasure, uniform precision/tape,
+   storage, compiler/outer conjunction and human/formal acceptance. The finite
+   ordinary-pass census is not funded recursive work. Five assumptions remain.
+5. **Smallest useful next task:** review target input aliases and cost premises,
+   then select one missing native/dual admission obligation with actual source,
+   phases, restoration and cost. A broader.0033 mechanism needs its own complete
+   optimistic screen and physical falsifier; unchanged-profile repacking cannot
+   lift the preceding scoped repeated-H5 ceiling.
+6. **Reproduction / runtime:** Python3.12/stdlib:
+   python3 -B scripts/research_package.py --package research/p104-caller-independent-audit.
+   Original serial pair100.9167seconds/82.35MiB; two registered commands with
+   bounded70second scientific children. Exact frozen bytes, no network/pending
+   Git object at runtime. Publication/hosted validation is reported separately.
+7. **Credit:** Ali Romman, Gareth Davidson, David Arias, Cinna Davis,
+   Jacob Alan Sussman, zxv, Cao Yifan and retained native/Clifford/geometry/
+   tensor/compiler/analytic authors. New independent audit and publication for
+   Chris R/Clear1nsight with substantial OpenAI Codex assistance; Apache-2.0.

research/p104-caller-independent-audit/PROOF.md

@@ -0,0 +1,112 @@
+# Physical caller binding: implication and remaining premises
+
+This document describes the frozen checker, not an execution result. Its
+claims become finite validation evidence only if the paired execution passes.
+The retained all-input local native ABI and the correctness of unchanged paid
+native maps remain premises. This is not full parent admission or a new κ.
+
+## Exact spectral convention
+
+For binary vector a, the positive native Gaussian G_a has Fourier eigenvalue
+i^(a·z mod2). A list of signed axes is represented by a Z4 phase polynomial.
+The parity expansion modulo4 has only linear and quadratic terms; a coefficient
+2 contributes a translation and no quadratic term. The checker retains the
+constant phase separately. All Gaussians, binary address permutations and
+translations in this spectral calculation have eigenphase1 at zero frequency.
+An exterior scalar coefficient remains outside every digit-column tensor power.
+
+On each actual port q, the nine vectors q,b1,...,b8 from the pinned source form
+an orthonormal basis. The stage route swaps b_i with the corresponding fresh
+axis e_i and fixes q. It is the product of commuting binary transvections
+v_i=b_i+e_i, each of even norm. The emitted recipe specifies these v_i. The
+checker reconstructs every column of its 265-dimensional matrix, checks the
+inverse on every basis vector, and checks orthogonality on its entire changed
+support against all265 columns. The unchanged complement is the identity.
+
+For each contact J=X_t P_R, the complete forward phase identity is checked,
+as is the inverse J^-1=P_R X_t=X_(Rt) P_R. Four contacts are emitted per full
+site/port. More than checking charts, the physical composite is checked:
+
+    J_exit F_local J_entry.
+
+For a diagonal coefficient this has phase R(D0)+2(t_exit+R(t_entry)); for the
+lower cross coefficient replace D0 by D0 Q^-1. Each equals the intended Ds or
+Ds Q^-1. This binds the actual entry/exit implementations to the free-Laurent
+caller. It retains all-input local independence from productive helpers.
+
+## Free-Laurent caller and canonical input aliases
+
+The literal RIGHT boundary program, physical chronology and LEFT program run
+over Q[Q^±1,D0^±1,...,D32^±1]. No Q^4=1 reduction is used to obtain the
+diagonal. Signed-swap negatives are ordinary scalar coefficients. The complete
+32-by32 matrix must equal the recorded diagonal with coefficient exactly1.
+
+For every final main field the emitted P_O is an orthogonal base9 change and
+an explicit permutation of all265 coordinates. Its first r columns are the
+positive Gaussian axes. Both complete input and inverse output actions are
+checked on every basis vector. The input alias P_O is placed before RIGHT;
+the final X_q correction when needed and P_O^-1 are placed after LEFT. Thus
+the independent input is not renamed for free only at the output. Every main
+field now has the literal standard rank-r Gaussian job on its own arbitrary
+input. The large canonicalizer is counted separately from a seventeen-axis
+local contact.
+
+## Helpers, scratch and paid jobs
+
+Each full site receives its own527 fresh arbitrary helper inputs. The local
+public735-label injection is emitted for every site. Global helper IDs are
+exactly3328,...,12813; main IDs are0,...,3327. Owned scratch12814 is not a
+productive input or target. Eight nonoverlapping source-owned lifetimes can
+reuse that scratch, as can chronological external queries; peak usage is1.
+The already checked complete local chronology supplies the lifetime premise.
+
+For every helper endpoint U=Phi_final Phi_initial^-1, the retained compiler
+lowers U=M_left G_r M_right. The literal native gate circuit is checked using
+signed Pauli images and an independently executed exact dense column at input0.
+All compared circuits consist of verified Clifford gate types. Equality of
+all signed Pauli conjugations implies that their quotient commutes with the
+full matrix algebra, hence is a scalar. The nonzero exact column fixes that
+scalar, including its absolute phase. The compiler's internal assertions are
+not relied on for these external comparisons, which also run under Python -O.
+
+The checker also tests M_left^-1 U M_right^-1=G_r. The right inverse acts on
+that fresh helper before its local call, and the left inverse after the call.
+This use of arbitrary helper inputs relies on the local all-input ABI: main
+outputs have no residual helper contamination and helper jobs are autonomous.
+Tensor spectators are identities, not additional productive jobs. Different
+port boundary routes can therefore coexist with these autonomous helpers.
+The208 intrinsic main adapters are checked by the same native method.
+
+Each thin query emits owned initialization, copy, the q-basis inverse,
+one positive rank-one Gaussian, the forward basis, X_q, add and owned clear.
+The source is untouched. Positive Q followed by X_q equals Q^-1 exactly;
+omitting X_q fails the spectral identity. Every whole idle emits the complete
+265-axis basis inverse, one positive Gaussian, forward basis and any required
+translation. Both bank occurrences are charged. No thin head interpretation
+replaces the complete inventory.
+
+## Ordinary accounting is separate
+
+The finite abstract pass census retains the source's local scalar and route
+allowances. It additionally charges1000 passes per external query, per idle
+field and per pair of main input/output canonicalizers, plus a conservative
+bank-operation allowance and every even-Q correction. A general seventeen-axis
+binary map and translation use at most17^2+17 elementary address passes.
+A265-coordinate permutation uses at most264 swaps, and an invertible9-by9
+binary map uses at most81 elementary operations; a main pair plus translation
+therefore fits2(264+81)+9<1000. Source local helper monomials remain within
+the retained per-map allowance. These are finite address/phase/scalar passes
+under the retained per-pass and tape contract, not recursive rank charges.
+
+This does not newly establish uniform bit precision, complete tape routing or
+storage, materialized compiler reserve, outer transfer, all remaining native
+maps, human mathematical review or proof-kernel acceptance. Those remain
+separate admission gates. A passing result closes this one caller binding
+layer with unchanged paid and productive profiles. It cannot itself increase
+the raw crossing or fund any new κ.
+
+Credit Ali Romman/#5544/#5545; Gareth Davidson/#2305; David Arias/#2152;
+Cinna Davis/#2219 and controller work; native phase, Clifford, tensor and
+compiler predecessors including Jacob Alan Sussman, zxv and Cao Yifan. New
+audit and this implication for Chris R/Clear1nsight with OpenAI Codex
+assistance. Apache-2.0.

research/p104-caller-independent-audit/README.md

@@ -0,0 +1,117 @@
+# Independent P104 forward caller and target-binding audit
+
+This package publishes a reproducible, source-bound check of one missing layer
+in Ali Romman's [PR5545](https://github.com/CrocSwap/integer-mult-bounds/pull/5545):
+the actual265-axis forward caller routes and independent productive targets.
+The complete recursive invoice is unchanged. **No new construction, κ gain,
+global ceiling, selected update or full parent admission is claimed.**
+
+Canonical base:32890bc5de034ab8fc93ac88ead7279a6a416d04. Reviewed source:
+18968423711fd902aaee1099979497d55603e1dc; predecessor PR5544:
+e2a6cce2f9db5a2c5f9714f4b9e5292dd3a10581. The preceding independent
+contact/native/ceiling result is [PR5548](https://github.com/CrocSwap/integer-mult-bounds/pull/5548),
+head b038f9d9b5cf0ce92184da909e57a3be476fc812. Its hosted result is separate.
+Only its dense native interpreter needed by this layer is reused; this package
+does not rerun the old contact/ceiling science or copy that complete package.
+
+## What is checked
+
+| Object | Coverage |
+|---|---|
+| Actual265-axis routes |7,488 entry/exit contacts at18 full sites ×104 ports; orthogonal inverses, complete spectral translations and composed local block coefficients |
+| Literal complete caller |All32 banks over a free Laurent ring in Q,D0,...,D32, with RIGHT/events/LEFT order, exterior signs and exact expected diagonal |
+| Main targets |3,328 arbitrary inputs, explicit265-axis input aliases before RIGHT and output inverse maps after LEFT, final translation corrections and positive ranks |
+| Native target binding |527 helper endpoint normalizers and208 intrinsic main adapters;288 distinct signed-Pauli/independent dense-phase comparisons |
+| Independent fields |9,486 fresh helpers,18 injective735-public-label bindings; scratch12814 never counted as productive |
+| Paid callers |3,120 whole-idle jobs and1,560 positive-child query recipes with inverse-Q correction and fixed owned cleanup template |
+| Complete invoice |69,912 paid jobs and12,814 productive fields, exactly equal to the source histograms |
+
+Helper profile per local call:51H4+36H5+24H7+122H8+294H9. Ten scoped faults
+reject, plus an exact negative-Laurent-power cancellation check. The complete
+5,990,419-byte output is stored losslessly in `expected/caller.json.gz`.
+Its decompressed SHA256 is
+`b51a7d19c10faea7d569a48e9536db8ba1e7418f8dd8dfb25646cc87cc9c4009`.
+Compression bytes are immutable capsule data, not a cross-environment science
+comparison; replay compares the complete decompressed original output.
+
+The separate ordinary census is127,292,680 abstract address/phase/scalar passes
+under the retained per-pass/tape contract, below source raw allowance184,647,552.
+It is not a new machine-runtime/precision bound, recursive charge or κ reserve.
+
+## Proof and limits
+
+[PROOF.md](PROOF.md) states the spectral convention, actual block composition,
+input alias placement, native phase-anchor implication and finite pass census.
+[REVIEW.md](REVIEW.md) records coverage and deliberate limitations.
+
+Local all-input/all-width native semantics and untouched paid-map correctness
+remain inherited. Owned initialization/copy/use/clear is checked against the
+declared template; the owned-erasure framework assumption is not independently
+proved. Contact inverses are not a complete-caller raw inverse transpose.
+No new whole-caller raw inverse-transpose, uniform precision/tape/storage,
+compiler reserve or complete outer conjunction replay is supplied. The source
+five framework assumptions remain. Agent checks are not human mathematical
+acceptance or proof-kernel verification of the multiplication theorem.
+
+## Reproduction
+
+Python3.12, standard library, no network or pending Git object at runtime:
+
+```sh
+python3 -B scripts/research_package.py --package research/p104-caller-independent-audit --check-only
+python3 -B scripts/research_package.py --package research/p104-caller-independent-audit
+python3 -B scripts/check_research_commit.py --package research/p104-caller-independent-audit --check-source-history
+```
+
+Expected marker: `PASS_P104_FORWARD_CALLER_INDEPENDENT_AUDIT`.
+Registered normal and optimized commands each permit150seconds, with a70second
+scientific child. The original serial pair passed in100.9167seconds/82.35MiB,
+normal51.9168seconds and optimized48.3985seconds, within150seconds/512MiB.
+Mode equality is reproducibility, not a second independent implementation.
+Publication validation and hosted CI are reported separately in the PR.
+
+`capsule/` preserves the exact original checker and all authenticated inputs;
+`original-evidence/` preserves the original request, pair, guard and stage traces.
+The first checker hit512MiB before a result; its code and closeout are retained
+only as authenticated history. This revision bounds the phase cache at128,
+packs all34 exact exponents with overflow rejection and streams outputs. It
+retains full matrix coverage and never substitutes finite-order identities.
+
+The capsule contains pending-head bytes authenticated by its complete inventory
+and the reviewed source manifest, not a dependency on fetching that pending
+head. Canonical compiler/circuit/phase files are live inputs in SOURCES.json
+and VERIFY.json and must agree byte-for-byte with their capsule copies.
+Source-history checks cover the registered canonical pins; they do not imply
+that the pending construction has been merged or accepted.
+
+## Task and overlap
+
+Task `scalar-to-moment`, revision7, changed scope/partial: an explicit user
+assignment to validate the pending P104 caller. The card's new paid-profile
+gain gate is not claimed met. Current main guidance was reread; fresh bounded
+PR/discussion checks found no competing successor, but hidden/unpublished
+coverage is incomplete. The source author released scope in
+[Discussion5125](https://github.com/CrocSwap/integer-mult-bounds/discussions/5125#discussioncomment-19169220).
+
+## Open next questions
+
+Structured first steps and blockers are in [TASK.json](TASK.json). The next
+admission question is the untouched native foundation and the actual whole
+caller raw inverse transpose required by the compiler, followed separately by
+precision/tape and complete funding applicability. Review target aliasing and
+ordinary-cost premises before expanding. For κ≥.0033, nominate a genuinely
+broader paid-path mechanism with a complete optimistic bill and smallest
+physical falsifier. This unchanged invoice does not lift PR5548's scoped
+repeated-H5 raw ceiling(.003106074473,.003106074474).
+
+This completed layer stops after publication. It launches no further search,
+recurring monitor, merge or selected change. See [HANDOFF.md](HANDOFF.md).
+
+## Credit
+
+Ali Romman/#5544/#5545; Gareth Davidson/#2305, David Arias/#2152,
+Cinna Davis/#2219/controller; Jacob Alan Sussman, zxv, Cao Yifan and all
+retained native, Clifford, geometry, tensor, compiler/analytic contributors.
+Source notices remain intact. New independent checker, memory-bounded
+representation and publication for Chris R/Clear1nsight with substantial
+OpenAI Codex assistance. Apache-2.0.