Construct a two-real packed Tensor caller with finite staged conformance (no new κ)

research/real-packed-tensor-staged-caller/PACKAGE-PINS.json

@@ -0,0 +1,12 @@
+{
+  "source_archive_sha256": "5b683359c99644635d088864cdf9427c9913f01f80af8ef9b387537512bbf77b",
+  "source_manifest_sha256": "d1accba0ce83ac82a14da3111e91ed52b4e3cdb16ff77eeba8a848f3f94cfaee",
+  "evidence_manifest_sha256": "4d94e4986bb7b4bff72ff0de4136d03908c36020cd427ca621f5cabed853c633",
+  "candidate_sha256": "5dfcdfe63d9d03a89a00dd92e681ab0a0efc5bbc86aa7edae58f007baf752483",
+  "source_head": "e65f47db55b9a3779fbf7fe23e93a8a4c5eb3ce1",
+  "source_status_successor": "467445ad61472f25e8b4d6757f838aceecf4fdec",
+  "source_status_successor_delta": "README only, scientific code and inputs unchanged",
+  "production_recovery_admitted": false,
+  "saving_admitted": false,
+  "recursive_kappa_claim": false
+}

research/real-packed-tensor-staged-caller/README.md

@@ -0,0 +1,53 @@
+# Two-real packed outer caller: finite staged conformance
+
+**The actual observed integer is 0, equal to the reduced staged reference 0; the true product is 43113955509404415, so observed minus true is −43113955509404415.** One actual caller completed **two Tensor calls and three CRT calls**, comparing all 35 records at each declared stage, in 781.405280 seconds. This is finite staged conformance, not accurate integer recovery, production admission, a total finite speedup, or recursive κ improvement. The zero opposite result makes dynamic recovery-normalization discrimination degenerate.
+
+## Construction and evidence
+
+The new construction is the unchanged [packed_sandwich_DRAFT.py](evidence/real-packed-outer-consumer-20261023/packed_sandwich_DRAFT.py), SHA256 `5dfcdfe63d9d03a89a00dd92e681ab0a0efc5bbc86aa7edae58f007baf752483`. It preserves the published caller's constructor, private ownership, ingress, child protocol, recovery, carry, admission and consume methods. Both independent real streams are represented once, packed as u+i v, transformed once, unmixed with paid coordinate negation J, multiplied using exact widened raw splits and one truncation, then passed through the original opposite transform and recovery. Three CRT conversions remain.
+
+For H=(a+ib)/2^p and JH=(c+id)/2^p, the exact raw split operands are (a+c,b−d) and (b+d,c−a), on the p+1 grid. Their product is truncated once by p+2 to return to the p grid. On this p48 fixture the operand width is 51 and truncation shift is 50. The source R frequency dilation commutes with J; both S factors and the original recovery scale remain. The [source-bound contract](evidence/real-packed-outer-consumer-20261023/CONTRACT.md), [mathematical guard](evidence/mathematical-guard/real-packing-5095-fit.md) and [independent guard review](evidence/real-packing-independent-review-20261023/REVIEW.md) state the conditional production premises separately.
+
+The independent input streams are `[255,0,3,17]` and `[1,254,0,9]`, b8/q4, source (5,7), target (8,8), p48/P80, `require_recovery=False`. Raising precision at fixed reduced geometry does not supply production admission. The [actual report](evidence/real-packed-outer-consumer-20261023/FULL-ATTEMPT2-REPORT.md), [receipt with actual arrays](evidence/real-packed-outer-consumer-20261023/FULL-PACKED-ATTEMPT2-RESULT.json), [checkpoint](evidence/real-packed-outer-consumer-20261023/FULL-PACKED-ATTEMPT2-CHILD-CHECKPOINT.json), log and [independent post-run review](evidence/packed-caller-independent-gate-20261023/POSTRUN-REVIEW.md) retain exact bytes and limits. The forward has 28 nonzero records, the product 35, and the opposite zero. No oracle value overwrote actual output.
+
+Normal closure, borrowed/metadata values and heads, idle child owners, empty parent leases, pending-output rejection, paid consume, and later default production rejection passed. A separate failure before the first original CRT numerical operation passed its declared parent cleanup. This is not arbitrary active-child failure coverage. The first full run remains a **580.390962-second FAIL**, including interrupted-child scratch-lease and marker-erasure cleanup exceptions. Earlier harness setup failures, source leaves and actual method-leaf receipts remain preserved, rather than promoted into whole-caller evidence.
+
+Finite counters are 2 transforms, 3 conversions, 2 J-axis passes, 35 split records, 140 signed split operations and 140 parent-declared real products. The contract separately derives 30200 J source-loop moves, 7000 writes and 70 width-86 recovery products. These do not price every metadata, packing, storage, reset/clear and Tape transition. The original source's three-Tensor count is construction context, **not a matched benchmark**; this package reran no baseline. Captures, reference evaluation and checkpoint I/O are experiment overhead.
+
+## Quick replay and optional long reproduction
+
+From repository root, the registered quick hook is:
+
+```sh
+python3 -I -B -W error research/real-packed-tensor-staged-caller/verify.py
+python3 scripts/research_package.py --package research/real-packed-tensor-staged-caller --check-only
+python3 scripts/research_package.py --package research/real-packed-tensor-staged-caller
+```
+
+It authenticates the 83-file archived source cohort and frozen evidence, then runs the unchanged independent post-run host checker in a temporary byte-identical layout. It uses host integer arithmetic for CRT, exact packing, split product, recovery and carry checks. It checks source/return byte-drift rejection and wrong-J/wrong-shift discrimination. **It does not import the numerical cohort, instantiate owners, execute Tensor/CRT calls, prepare coefficients or freshen the physical receipt.** It is also runnable in a Git-free exported package. Normal and optimized verification must retain the same checks.
+
+For an explicit separate long reproduction, first materialize a NEW directory:
+
+```sh
+python3 -I -B -W error research/real-packed-tensor-staged-caller/materialize_reproduction.py /tmp/packed-staged-NEW
+cd /tmp/packed-staged-NEW
+(set -C; env -i PATH=/usr/bin:/bin LANG=C.UTF-8 /usr/bin/timeout -k 10s 900s /usr/bin/python3 -I -B -W error run_full_packed_attempt2_DRAFT.py > FULL-PACKED-ATTEMPT2.log 2>&1)
+```
+
+This executes the exact frozen actual harness with internal 880-second alarm, external 900-second cap and 30-second heartbeats, the same diagnostic fixture and no catalogue producer. Never overwrite existing result/checkpoint/tmp/log paths, vary the fixture, extend the budget or automatically retry. The optional command is **not registered in quick CI and was not rerun for packaging**. The existing successful receipt came from Python3.12/Linux; runtime portability is not promised. The prior 600-second Linux source timeout and later 452.68-second Linux source success are retained as historical source context, not matched candidate timing.
+
+## Source-bound ceiling and remaining scope
+
+The retained [real/Hermitian source closure report](evidence/real-hermitian-closure-20261023/REPORT.md) follows actual Tensor→Core→Interior/BandedSolve→FftWord lineage. That Core already executes real and imaginary scalar passes. Merely pairing two independent real channels scales both reward and paid work together, so this source-bound normalized moment does not improve. Independently dirty helper states need their full inventory served or lose independent reward. This is a ceiling on the inspected source contract, not a universal impossibility theorem. The rational-unitary RawGaussian/Gaussian/Full reference chain is not a runtime-child adapter here. No new literature search was used for packaging.
+
+The four complete-caller numerical mutations (wrong opposite sign, inverse R, missing S, double gamma) remain unrun. Earlier method-leaf controls have their narrower supplied-bank scope. This package establishes no all-input or all-size theorem, production recovery guard satisfaction, total priced profit, selected supplier, compiler reserve or κ change. Current canonical base is a5df48c55ea806d7952b65acc9c76d878ff09b13; bounded monitor coverage found no competing finite packed caller through its reported 11:12:56/cursor5156 intake, not exhaustive novelty.
+
+## Open next questions
+
+The structured [TASK.json](TASK.json) retains three bounded questions: bind a production-admissible nonzero recovery cohort; price the complete finite wrapper against a matched admitted baseline; and repair active-child interruption cleanup in a separate source change. Each states its relevance, first step and blocker. No further numerical fixture is proposed by this submission.
+
+## Sources, licensing and credit
+
+Primary source is [JoyboyBrian's PR5095 caller at e65f47db](https://github.com/CrocSwap/integer-mult-bounds/tree/e65f47db55b9a3779fbf7fe23e93a8a4c5eb3ce1/research/physical-tensor-crt-sandwich-recovery). Successor467445ad changes only README; scientific bytes are unchanged. Tensor derives from [Jose Manuel / jmbarrancoml's PR3935 at f687fa17](https://github.com/CrocSwap/integer-mult-bounds/tree/f687fa171ef760fa98a4be1289717b47dd4ee0b3/research/physical-gaussian-tensor); CRT derives from Tekalign / tekuuu's PR4024 at e99297be; Mason Hale's prior dense rigorous recovery is credited by the source and not duplicated. Nested source manifests preserve the native, word, ring, routing and predecessor authors.
+
+[Inputs](inputs/SOURCE-COHORT.json) preserve all 83 consumed public files, including their LICENSE/NOTICE and authenticated nested archives, without editing predecessor packages. [SOURCES.json](SOURCES.json) pins base history and primary source links; [PACKAGE-PINS.json](PACKAGE-PINS.json) and [EVIDENCE-SHA256.json](EVIDENCE-SHA256.json) pin the retained bytes. New construction, packaging and checks are Apache-2.0 with substantial OpenAI Codex assistance. Source-specific original licenses/notices remain authoritative. The reviews are separate agent reviews, not independent human or formal-kernel acceptance.

research/real-packed-tensor-staged-caller/SOURCES.json

@@ -0,0 +1,30 @@
+{
+  "schema_version": 1,
+  "base_commit": "a5df48c55ea806d7952b65acc9c76d878ff09b13",
+  "files": [
+    {
+      "path": "certificates/selected-result.json",
+      "usage": "provenance",
+      "git_blob": "c35482cd9035843b15a634cc8276a7e1e40e8395",
+      "sha256": "6fe0a099b89aa2aed796a13c3a21dafc63f3388ebf397464e44e0451893b9375"
+    },
+    {
+      "path": "docs/research/agent-task-catalog.json",
+      "usage": "provenance",
+      "git_blob": "f126dd2e2dad2b669e75d1d0d0bd05c600725602",
+      "sha256": "f7a48bfa2659b88378282a8eee7d9fe52cdf505594124c4d1686b58510fffd72"
+    }
+  ],
+  "retained_source_cohort": {
+    "path": "inputs/pr5095-consumed-source.tar.gz",
+    "sha256": "5b683359c99644635d088864cdf9427c9913f01f80af8ef9b387537512bbf77b",
+    "manifest": "inputs/SOURCE-COHORT.json",
+    "file_count": 83,
+    "origin": "Exact consumed PR5095 e65 public source/input cohort; nested source licenses and notices preserved.",
+    "primary_git_urls": [
+      "https://github.com/CrocSwap/integer-mult-bounds/tree/e65f47db55b9a3779fbf7fe23e93a8a4c5eb3ce1/research/physical-tensor-crt-sandwich-recovery",
+      "https://github.com/CrocSwap/integer-mult-bounds/tree/f687fa171ef760fa98a4be1289717b47dd4ee0b3/research/physical-gaussian-tensor",
+      "https://github.com/CrocSwap/integer-mult-bounds/tree/e99297be6749518a1b5b85cd129ecaf204d26c63/research/crt-address-streams"
+    ]
+  }
+}

research/real-packed-tensor-staged-caller/TASK.json

@@ -0,0 +1,86 @@
+{
+  "schema_version": 1,
+  "task_id": "uniform-transfer-audit",
+  "priority_revision": 7,
+  "base_commit": "a5df48c55ea806d7952b65acc9c76d878ff09b13",
+  "original_card": {
+    "id": "uniform-transfer-audit",
+    "question": "Does one quantified step of the proposed #4100 all-size construction hold at its actual caller?",
+    "starting_points": [
+      "docs/research/research-family-status.md",
+      "certificates/selected-result.json",
+      "research/unconditional-batched-crt/README.md"
+    ],
+    "methods": [
+      "Prove one missing implication with explicit quantified hypotheses.",
+      "Find an exact counterexample at the claimed interface boundary."
+    ],
+    "acceptance_gate": "Identify the proposition and what would change downstream; do not equate a finite replay with a universal proof.",
+    "first_step": {
+      "command": [
+        "python3",
+        "-B",
+        "scripts/research_baseline.py"
+      ],
+      "expected": "PASS_RESEARCH_BASELINE_BINDINGS_ONLY",
+      "inspect": "Read PROOF.md and AUDIT.md in #4100. Select one distinct obligation: CRT total repair, native/free-algebra lifting, precision/conditioning, or complete fixed-tape cost. This is an author-submitted proof, not an accepted unconditional theorem.",
+      "scope": "Source-binding orientation only; no scientific replay, theorem acceptance or supplier admission. Budget any heavy replay separately."
+    },
+    "first_experiment": "Check live handoffs; name one proposition, its caller, quantified domain and dependency closure. Independently prove the implication or exhibit a precise gap/counterexample. Prioritize CRT.md total repair or COMPLEX.md free-algebra lift only if unclaimed. Stop at the chosen obligation; do not substitute finite samples or root arithmetic for an all-size proof, or launch the complete package by default.",
+    "bottleneck_link": "Finite supplier improvements are useful only under the retained transfer obligations.",
+    "allowed_variations": [
+      "One CRT repair, native lift, precision/conditioning or tape-cost proposition, with its exact caller."
+    ],
+    "scope_change_triggers": [
+      "Replacing a quantified obligation with another finite smoke test.",
+      "Treating archive intake, an author PASS receipt or a finite composed fixture as discharge of the universal hypotheses."
+    ],
+    "overlap_terms": [
+      "uniform transfer",
+      "all-size",
+      "precision",
+      "4100",
+      "unconditional-batched-crt"
+    ],
+    "existing_checks": [
+      {
+        "scope": "PR #4100: proposed unconditional construction and finite checks, without independent acceptance.",
+        "evidence": "research/unconditional-batched-crt/README.md",
+        "remaining": "Independent proof audit of one caller-bound all-size implication; all other premises remain explicit.",
+        "status": "author-checked",
+        "source_commit": "337d04939571a637aeab54a0f58f54fde811663a",
+        "selected_record_sha256": "bb2eaa58647b353f88d0aec2eafb9d30df1be53589231b23e12ceef3e4bad129"
+      }
+    ]
+  },
+  "scope_status": "changed",
+  "actual_scope": "Construct one two-real packed outer caller and preserve one actual two-Tensor reduced staged conformance receipt, plus quick independent host evidence replay.",
+  "scope_change_reason": "Explicit donor assignment: a finite construction side result on the actual PR5095 caller, rather than a complete all-size uniform transfer proof.",
+  "bottleneck_link": "Finite physical caller wiring is a prerequisite check; reduced conformance does not close production or recursive transfer.",
+  "budget_minutes": 60,
+  "stop_rule": "One frozen reduced fixture; no further numeric replay, no production/total-saving/kappa admission.",
+  "outcome": "completed",
+  "evidence": "README.md and evidence/real-packed-outer-consumer-20261023/FULL-ATTEMPT2-REPORT.md",
+  "author": "CrocSwap research contributors with OpenAI Codex assistance",
+  "ai_assistance": "Substantial OpenAI Codex assistance in construction, static review, host arithmetic review, packaging and checks.",
+  "open_questions": [
+    {
+      "question": "Can a production-admissible source cohort deliver nonzero accurate recovery under the same exact packed caller?",
+      "why_it_matters": "The current reduced opposite is zero and the output differs from the true product.",
+      "first_step": "Bind admitted geometry, coefficient precision and native premises before proposing a new bounded fixture.",
+      "blocked_by": "Source production parameter and full preparation-cost admission remain open."
+    },
+    {
+      "question": "Does the complete physical tariff favor this outer wrapper?",
+      "why_it_matters": "Two Tensor calls alone do not price packing, J, split, storage and cleanup or imply a matched speedup.",
+      "first_step": "Price every added finite operation against one matched admitted baseline without changing the fixture implicitly.",
+      "blocked_by": "No complete priced tariff or matched baseline is claimed."
+    },
+    {
+      "question": "Can active-child deadline cleanup be made complete?",
+      "why_it_matters": "The first full attempt exposed scratch-lease and marker cleanup exceptions.",
+      "first_step": "Review the inherited active-child ownership on the recorded timeout cut in a separately scoped source repair.",
+      "blocked_by": "This package preserves source and candidate bytes and does not repair interruption cleanup."
+    }
+  ]
+}

research/real-packed-tensor-staged-caller/VERIFY.json

@@ -0,0 +1,21 @@
+{
+  "schema_version": 1,
+  "status": "ready",
+  "python": "3.12",
+  "requirements": [],
+  "inputs": [],
+  "commands": [
+    {
+      "argv": [
+        "{python}",
+        "-I",
+        "-B",
+        "-W",
+        "error",
+        "{package}/verify.py"
+      ],
+      "timeout_seconds": 60,
+      "checks": "Host-only frozen actual35-record chain, exact83-file source cohort, independent CRT/packing/product/recovery/carry arithmetic, lifecycle evidence binding and source/return/J/shift negative controls. No fresh numerical execution."
+    }
+  ]
+}

research/real-packed-tensor-staged-caller/evidence/canonical-guidance-check-1043.json

@@ -0,0 +1,97 @@
+{
+  "at": "2026-10-23 10:42:42 UTC",
+  "head": "a5df48c55ea806d7952b65acc9c76d878ff09b13",
+  "base": "20a1e31bc6009b9b3f4d084bbaadc6350e34fcba",
+  "checks": [
+    {
+      "status": "fulfilled",
+      "value": {
+        "path": "AGENTS.md",
+        "sha": "aea5573d1b65cef7d9e08ced46d869e75da8145b",
+        "expected": "aea5573d1b65cef7d9e08ced46d869e75da8145b",
+        "unchanged": true,
+        "error": false
+      }
+    },
+    {
+      "status": "fulfilled",
+      "value": {
+        "path": "CONTRIBUTING.md",
+        "sha": "fe54f23f19b082a9f8df3da9b2e205f7e230acf3",
+        "expected": "fe54f23f19b082a9f8df3da9b2e205f7e230acf3",
+        "unchanged": true,
+        "error": false
+      }
+    },
+    {
+      "status": "fulfilled",
+      "value": {
+        "path": "autoresearch.md",
+        "sha": "2fe84a7b30b1e26e60ce0ec4775894eb1831c5d6",
+        "expected": "2fe84a7b30b1e26e60ce0ec4775894eb1831c5d6",
+        "unchanged": true,
+        "error": false
+      }
+    },
+    {
+      "status": "fulfilled",
+      "value": {
+        "path": "docs/research/coordination.md",
+        "sha": "96db59be318181939d1e02b54ebba626b997d787",
+        "expected": "96db59be318181939d1e02b54ebba626b997d787",
+        "unchanged": true,
+        "error": false
+      }
+    },
+    {
+      "status": "fulfilled",
+      "value": {
+        "path": "docs/research/agent-procedure.md",
+        "sha": "27d7029ef3840b96db1654d2448b6e7b49a8ce52",
+        "expected": "27d7029ef3840b96db1654d2448b6e7b49a8ce52",
+        "unchanged": true,
+        "error": false
+      }
+    },
+    {
+      "status": "fulfilled",
+      "value": {
+        "path": "docs/research/current-priorities.md",
+        "sha": "a87818dafbca574053504ab07ceb7dd5eb805f77",
+        "expected": "a87818dafbca574053504ab07ceb7dd5eb805f77",
+        "unchanged": true,
+        "error": false
+      }
+    },
+    {
+      "status": "fulfilled",
+      "value": {
+        "path": "docs/research/research-family-status.md",
+        "sha": "743a1afcbc46c715d2035037797e2f81b32e58a5",
+        "expected": "743a1afcbc46c715d2035037797e2f81b32e58a5",
+        "unchanged": true,
+        "error": false
+      }
+    },
+    {
+      "status": "fulfilled",
+      "value": {
+        "path": "docs/research/agent-task-catalog.json",
+        "sha": "f126dd2e2dad2b669e75d1d0d0bd05c600725602",
+        "expected": "f126dd2e2dad2b669e75d1d0d0bd05c600725602",
+        "unchanged": true,
+        "error": false
+      }
+    },
+    {
+      "status": "fulfilled",
+      "value": {
+        "path": "certificates/selected-result.json",
+        "sha": "c35482cd9035843b15a634cc8276a7e1e40e8395",
+        "expected": "c35482cd9035843b15a634cc8276a7e1e40e8395",
+        "unchanged": true,
+        "error": false
+      }
+    }
+  ]
+}