Audit actual QQ16 pair and exclude selected-only post-pair bridge repair

research/qq16-paid-pair/COST.md

@@ -0,0 +1,144 @@
+# Literal pair invoice and bridge boundary
+
+This audit compares the selected gauge44 pair in replica 0 with literal
+`F_0; B_1_2` from the QQ16 packet. The selected calls are `calls[0]`
+(stage 0, forward) and `calls[4]` (stage 1, backward), not the first two
+serialized calls. The intervening calls concern other replicas. The next
+same-replica ordinary bridge is execution event 6, source event 3, after
+the pair-1 idle transition. It applies, at every one of the 112 ports,
+
+```
+X_0 += X_1
+Y_1 -= Y_0.
+```
+
+Here the subscripts name global pair banks. Each bank has 112 ports.
+Thus both legs and the selected A/T ports remain part of the caller's
+observation. The inherited signed bank semantics are bound to the pinned
+`global_k.py` controller by AST checks: its pure four-pair recipe matches
+all eighteen preterminal events of the selected recipe, and the signed
+bridge branch performs exactly the two updates above. Source `exchange`
+and selected `terminal(pairs=4)` remain explicitly different terminal
+conventions. Hosted execution event 6/source event 3 matches recipe event
+2. This authenticates inherited signed controller semantics; it does not
+execute the physical complex bridge. The bridge has 224 complex ADDs, or 448 real-component
+additions. The conservative typed dyadic ADD compiler used for the local
+scalar invoices instead charges 1,792 real primitives. These are different
+accounting conventions, explicitly separated in the receipt. The unchanged
+bridge appears on both sides of a valid replacement comparison.
+
+## Local words
+
+All counts below come from the literal event lists. `center` events are
+counted as native children as well as `climb` and `native_line` events.
+In particular, the selected word's eight rank-eight centre calls must not
+be omitted from its baseline.
+
+| Component | Selected forward | Selected backward | QQ16 F0 | QQ16 B1_2 |
+|---|---:|---:|---:|---:|
+| Native child calls | 3,494 | 3,494 | 3,494 | 3,510 |
+| Sum of native child ranks | 5,586 | 5,586 | 5,530 | 5,546 |
+| Scalar macros | 19,693 | 19,693 | 21,025 | 21,401 |
+| Private CLEAR macros | 16 | 16 | 32 | 32 |
+| Scalar real-primitive allowance | 198,766 | 198,814 | 211,210 | 214,986 |
+
+The receipt names each of the sixteen candidate removed centre events
+by orientation, source index, rank and canonical event digest, and each
+introduced line by its flattened QQ16 index, rank and digest. These are
+candidate substitutions, not accepted deletions. Owner-map counts come
+from literal `cover_map` events; native-wrapper counts come from the left
+and right recipes of each literal `native_line` event.
+
+The QQ16 pair removes sixteen rank-eight centre children and introduces
+thirty-two rank-one children. Its native call count increases by 16 while
+its rank sum decreases by 96. It also adds 3,040 scalar macros, including
+32 CLEARs, and 28,616 scalar real primitives. The QQ wrappers contain 528
+explicit owner maps and 64 native line wrapper maps. The invariant rank
+32 of the data-map discrepancy is not one of these native resource counts.
+
+The retained #4015 decoder word independently counts as 224 rank-one
+native lines, 448 native wrapper maps, 352 owner maps, 2,816 ADDs, 416
+CLEARs, and scalar allowance 26,944. It uses the 224 owned output fields;
+its zero additional query-field count does not mean zero output storage
+or zero private workspace. Its predecessor's private bound, physical
+allocation and rounding obligations are not re-proved here.
+
+Even the raw component subtotal QQ16 plus this decoder, less the selected
+pair, has **240 additional native calls, native rank sum +128, and scalar
+allowance +55,560**. This is not yet a replacement budget: the decoder
+recovers the original pair inputs, whereas the selected caller requires
+its selected pair outputs at the next bridge. A connection from those
+recovered inputs to that endpoint, with native frames, owner paths and
+stock, remains to be supplied. Restoring an input does not erase the
+selected pair's output computation from the ledger. If those 224 outputs
+must later be returned zero, a separate terminal contract would add their
+cleanup after legitimate use; this audit does not assume that contract.
+
+Only after endpoint compatibility and comparable child tariffs are proved
+could these counts enter a screen such as
+
+```
+16 C8 > 256 C1 + 55,560 S + conversion + repair + cleanup + movement + precision.
+```
+
+Here C8 and C1 would need common-unit tariffs for the removed rank-eight
+and introduced rank-one children, and S is the tariff for a scalar real
+primitive. Native wrappers must be included exactly once in those tariffs
+or charged separately; cover maps need their own tariff. This is a
+conditional bookkeeping expression, not a proved inequality, a global
+lower bound, or a claim that one decoder per pair is the actual caller
+incidence. Without those contracts the exact count ledger is the result.
+
+## The selected-only endpoint repair does not survive owner dependence
+
+The constant-owner, constant-payload test makes the QQ16-versus-selected
+pair difference appear supported only on A union T. That observation
+suggests a repair touching those sixteen ports. The independent replay
+here tests this exact restricted proposal using a larger legal input
+space: payload is constant, but values depend on the full cover owner.
+It evaluates the literal F0 word and every B1_2 prefix event, preserving
+each actual `out(d)=in(d*g)` owner map from the packet.
+
+For input X at port 0 equal to one on owner I and zero on other owners,
+with all other inputs zero, the QQ16 source Y at ordinary port 1 has value
+**-1/2 at owner I**. The selected pair's corresponding Y is zero.
+Consequently, the next bridge produces +1/2 in pair-1 Y at that ordinary
+port under QQ16, versus zero under the selected pair. A postpair repair confined
+to the sixteen selected output ports, preserving ordinary fields and the
+pair-1 target inputs of this bridge, cannot change this observation.
+The witness is admissible under the declared independent-cover module
+interface. Its reachability from a particular outer integer-multiplication
+input has not been proved. The failure concerns all-permitted-state
+postpair selected-only repairs; it is not a universal impossibility on
+an arbitrary restricted reachable-input subspace.
+All 96 ordinary Y rows have a nonzero owner-dependent residual for the
+symbolic input X at port 0. If every owner map is intentionally collapsed
+to the identity, all 96 residuals vanish. This control identifies the
+precise premise lost by the constant-owner test.
+
+The source-Y comparison does not assume a full nonconstant formula for
+the selected pair. The independent sparse scalar replay checks all 712
+local input columns of both backward cores: each Y endpoint is the same
+Y input, with no other-input dependence, on the constant-payload character.
+The selected forward word at stage 0 is the same-port unit shear.
+At a provider invocation indexed by local owner d, port p enters and exits
+at physical owner d*rho_p; conjugating a same-port identity by that route
+leaves it an identity. Constant payload also makes every payload phase
+and translation trivial. Thus the source-Y endpoint survives the actual
+owner lift on this restricted test space. No analogous all-character or
+full X-endpoint statement is inferred from it.
+
+The implemented test uses full 73-column matrices, checks that its witness
+owners are orthogonal, and retains the exact rational coefficient and
+row digest. It does not import the primary pair verifier. The package's
+main verifier authenticates the common pinned inputs before calling this
+audit.
+
+## Budget status
+
+The selected and QQ16 local invoices are exact literal counts. The failed
+selected-only repair and unchanged full bridge are also explicit. A
+successful general correction, its native realizations, physical ownership
+and release, recursive precision, head movement, bit cost, and actual
+source-caller incidence are not established. A proposed invocation fraction
+is not a funded budget. No global multiplication constant changes here.

research/qq16-paid-pair/EVIDENCE.json

@@ -0,0 +1,25 @@
+{
+  "schema_version":1,
+  "scientific_base":"bb4bccbf42844d2801487450727c2f63e10d1d0f",
+  "orchestrator":"root: boundary selection, integration, source authentication, checks and publication",
+  "shared_context":"All agents share task/source context. Separate implementations and assisted review are not independent human acceptance.",
+  "workers":[
+    {"role":"primary construction","agent":"query_schur","specialty":"exact linear maps and cover representation","inputs":["selected replay archive","QQ16 packet","actual PR4015 WORD","native character/cover-action conventions"],"files":["primary.py","PROOF.md"],"deliverable":"actual first-pair and bridge binding, restricted repair obstruction, decoder data-domain identities","blockers":"outer entrance reachability and general paid repair unestablished","reviewed_sha256":{"primary.py":"5a333cb851c7fd3205137dacebdae5d21b34d70577f40d9dc5f8f248d1212c10","PROOF.md":"4c33abb3d69cfbef21a5f60541b20c27f2d5076b4bb3e9e26f073273d8e1bdd6"}},
+    {"role":"supporting analysis","agent":"centre_budget","specialty":"literal compiler accounting and independent group-algebra replay","inputs":["same pinned archive/packet","retained decoder WORD","authenticated signed-controller source"],"files":["cost_audit.py","COST.md"],"deliverable":"operation-indexed candidate substitutions, native/scalar/map invoice, independent ordinary-owner witness and bridge provenance","code_reuse":"AST extracts only the authenticated pure inherited recipe constructor; no predecessor module imported; independent owner replay imports no primary implementation","blockers":"no comparable native tariff, physical bill or actual recursive incidence","reviewed_sha256":{"cost_audit.py":"24873f9d76be16ffbc6199217c7408556a98e84aa51fe47874ab4acb48a6160f","COST.md":"caafddad767e2e40c4561d9df161ded6ad9f18e86ca393ecd5feb0667eddd26b"}},
+    {"role":"adversarial reviewer","agent":"centre_copy_trace","files":["REVIEWS.json"],"deliverable":"read-only final source/domain/current-vs-recovered/complete-bill review with independent small probes","review_receipt_sha256":"33d0f400b152639297007167d5cc8f4223db8e34ee264f37fec5b7dd81b88f53","blockers":"none for the scoped claim; inherited supplier obligations remain"}
+  ],
+  "claim_ledger":[
+    {"claim":"constant common-cover data difference rank32; DQ=QD=I224 and DG differs rank32","mathematical_status":"exact finite specialization","evidence_provenance":"new literal source/retained decoder replay"},
+    {"claim":"ordinary-owner delta witness and selected-only support obstruction","mathematical_status":"scoped quantified argument over declared independent-cover interface, exact witness","evidence_provenance":"new source-bound operator derivation, two implementations and assisted adversarial probe"},
+    {"claim":"native constant-payload and frame behavior","mathematical_status":"inherited operator premise on invariant subspace","evidence_provenance":"pinned bridge convention and actual phase recipes; no arbitrary-payload certification"},
+    {"claim":"local counts and scalar allowances","mathematical_status":"exact literal counts under stated dyadic compiler tariff","evidence_provenance":"reproduced event census, indexed substitutions and recipe wrappers"},
+    {"claim":"owned224 outputs/privatepeak16/queryextra0","mathematical_status":"inherited conditional module stock contract","evidence_provenance":"retained PR4015 receipt; exclusive caller lease not reproduced"},
+    {"claim":"224 terminal CLEARs/448 scalar if erasure is required","mathematical_status":"conditional inherited terminal charge","evidence_provenance":"PR4047 stated contract; not imposed on every redesigned endpoint"},
+    {"claim":"complete saving, native cost, motion, widths/error, lease, recursive incidence and improved kappa","mathematical_status":"unknown or unestablished","evidence_provenance":"not supplied; unknown terms never silently zero"}
+  ],
+  "failed_routes":["Unrepaired full pair: current port0 bridge -1 versus12 on constant X0","Naive recovered Y used as current bridge source: -1 versus0 on X1","Post-pair selected16-only writes with other bridge inputs fixed: ordinary bridge0 versus1/2 on X0 delta_I"],
+  "review_disagreements_resolved":["calls[1] was not B1; actual call index4 bound","F0/B1 helper sets differ; only224 data inverted","target precompensation excluded by preserving pair1 bridge inputs","delta input is permitted module state; outer multiplication reachability not established","signed bridge anchored separately from native alignment review","first18 controller events match; terminal/exchange spelling differs and is retained"],
+  "selected_frontier_changed":false,
+  "scientific_status":"scoped reusable negative result; no paid replacement",
+  "publication_status":"tracked outside frozen scientific evidence"
+}

research/qq16-paid-pair/PENDING-SOURCE-HISTORY.json

@@ -0,0 +1,31 @@
+{
+  "status": "PASS_RETAINED_PENDING_SOURCE_HISTORY",
+  "method": "git show exacthead:originalpath compared byte-for-byte to retained data; SHA256 and blob verified",
+  "scope": "Local full-history provenance check. Registered replay authenticates retained bytes without fetching pending heads or executing predecessor code.",
+  "files": [
+    {
+      "pr": 4015,
+      "head": "0fff81d8997cef646538bcd4f60d7d455cf93b8e",
+      "path": "research/qq16-streaming-recovery/WORD.json",
+      "git_blob": "9a57a6f9c2fd8cdc4e5934408d78ef54e5f266b6",
+      "sha256": "272d8c1bd172c29044eb227ec6d67a882501250978234685e035158f4479aa58",
+      "retained_byte_equality": "PASS"
+    },
+    {
+      "pr": 4015,
+      "head": "0fff81d8997cef646538bcd4f60d7d455cf93b8e",
+      "path": "research/qq16-streaming-recovery/RESULT.json",
+      "git_blob": "e9b6b8ea6f5a7dc47d8547cd7ed255e3d5864f7d",
+      "sha256": "b4af55d3db3212e79f8f31108e048c5c2d4c32d520fc980ebaa66f47933ad622",
+      "retained_byte_equality": "PASS"
+    },
+    {
+      "pr": 4044,
+      "head": "a8c75dd7c964855cc9b03c210575be27c65467e4",
+      "path": "research/qq16-provider-binding/RESULT.json",
+      "git_blob": "71bc31b468c6956e61f7527bbded721f6398b238",
+      "sha256": "7c64555cf48956b4692c1c078c1231bfd67cd4fa063abaf395a3c6a532da3051",
+      "retained_byte_equality": "PASS"
+    }
+  ]
+}

research/qq16-paid-pair/PREDECESSORS.json

@@ -0,0 +1,33 @@
+{
+  "schema_version": 1,
+  "files": [
+    {
+      "pr": 4015,
+      "commit": "0fff81d8997cef646538bcd4f60d7d455cf93b8e",
+      "path": "research/qq16-streaming-recovery/WORD.json",
+      "local": "vendor/PR4015-WORD.json",
+      "git_blob": "9a57a6f9c2fd8cdc4e5934408d78ef54e5f266b6",
+      "sha256": "272d8c1bd172c29044eb227ec6d67a882501250978234685e035158f4479aa58",
+      "usage": "retainedscientificdata; noexecutableimport"
+    },
+    {
+      "pr": 4015,
+      "commit": "0fff81d8997cef646538bcd4f60d7d455cf93b8e",
+      "path": "research/qq16-streaming-recovery/RESULT.json",
+      "local": "vendor/PR4015-RESULT.json",
+      "git_blob": "e9b6b8ea6f5a7dc47d8547cd7ed255e3d5864f7d",
+      "sha256": "b4af55d3db3212e79f8f31108e048c5c2d4c32d520fc980ebaa66f47933ad622",
+      "usage": "retainedscientificdata; noexecutableimport"
+    },
+    {
+      "pr": 4044,
+      "commit": "a8c75dd7c964855cc9b03c210575be27c65467e4",
+      "path": "research/qq16-provider-binding/RESULT.json",
+      "local": "vendor/PR4044-RESULT.json",
+      "git_blob": "71bc31b468c6956e61f7527bbded721f6398b238",
+      "sha256": "7c64555cf48956b4692c1c078c1231bfd67cd4fa063abaf395a3c6a532da3051",
+      "usage": "retainedscientificdata; noexecutableimport"
+    }
+  ],
+  "credit": "PR4015 and4044 byChrisR/Clear1nsight withsubstantialOpenAICodexassistance; preservecreditedQQ16/native/hostinglineage"
+}