Reduce complete native-history replay time by 23.67% (no κ)

research/native-history-setup-cache/BENCHMARK.json

@@ -0,0 +1,189 @@
+{
+  "errors": 0,
+  "executable": "/Users/nick/git/integer-mult-bounds/.venv/bin/python3.12",
+  "optimized": false,
+  "platform": "macOS-27.0.1-arm64-arm-64bit",
+  "python": "3.12.15 (main, Oct 14 2026, 00:55:37) [Clang 23.1.3 ]",
+  "samples": [
+    {
+      "mode": "baseline",
+      "output_sha256": "31116d29f77eaa5732cc76eb0c716c808f063c195e2aa886f4993ea33f51d29d",
+      "returncode": 0,
+      "seconds": 1.246776457875967,
+      "statistics": {
+        "cache": {
+          "currsize": 0,
+          "hits": 0,
+          "maxsize": 256,
+          "misses": 0
+        },
+        "mode": "baseline"
+      },
+      "trial": 0
+    },
+    {
+      "mode": "cached",
+      "output_sha256": "31116d29f77eaa5732cc76eb0c716c808f063c195e2aa886f4993ea33f51d29d",
+      "returncode": 0,
+      "seconds": 0.9610304171219468,
+      "statistics": {
+        "cache": {
+          "currsize": 58,
+          "hits": 158,
+          "maxsize": 256,
+          "misses": 58
+        },
+        "mode": "cached"
+      },
+      "trial": 0
+    },
+    {
+      "mode": "baseline",
+      "output_sha256": "31116d29f77eaa5732cc76eb0c716c808f063c195e2aa886f4993ea33f51d29d",
+      "returncode": 0,
+      "seconds": 1.281701625091955,
+      "statistics": {
+        "cache": {
+          "currsize": 0,
+          "hits": 0,
+          "maxsize": 256,
+          "misses": 0
+        },
+        "mode": "baseline"
+      },
+      "trial": 1
+    },
+    {
+      "mode": "cached",
+      "output_sha256": "31116d29f77eaa5732cc76eb0c716c808f063c195e2aa886f4993ea33f51d29d",
+      "returncode": 0,
+      "seconds": 1.0426137920003384,
+      "statistics": {
+        "cache": {
+          "currsize": 58,
+          "hits": 158,
+          "maxsize": 256,
+          "misses": 58
+        },
+        "mode": "cached"
+      },
+      "trial": 1
+    },
+    {
+      "mode": "baseline",
+      "output_sha256": "31116d29f77eaa5732cc76eb0c716c808f063c195e2aa886f4993ea33f51d29d",
+      "returncode": 0,
+      "seconds": 1.304742000065744,
+      "statistics": {
+        "cache": {
+          "currsize": 0,
+          "hits": 0,
+          "maxsize": 256,
+          "misses": 0
+        },
+        "mode": "baseline"
+      },
+      "trial": 2
+    },
+    {
+      "mode": "cached",
+      "output_sha256": "31116d29f77eaa5732cc76eb0c716c808f063c195e2aa886f4993ea33f51d29d",
+      "returncode": 0,
+      "seconds": 1.0262446249835193,
+      "statistics": {
+        "cache": {
+          "currsize": 58,
+          "hits": 158,
+          "maxsize": 256,
+          "misses": 58
+        },
+        "mode": "cached"
+      },
+      "trial": 2
+    },
+    {
+      "mode": "baseline",
+      "output_sha256": "31116d29f77eaa5732cc76eb0c716c808f063c195e2aa886f4993ea33f51d29d",
+      "returncode": 0,
+      "seconds": 1.2796051658224314,
+      "statistics": {
+        "cache": {
+          "currsize": 0,
+          "hits": 0,
+          "maxsize": 256,
+          "misses": 0
+        },
+        "mode": "baseline"
+      },
+      "trial": 3
+    },
+    {
+      "mode": "cached",
+      "output_sha256": "31116d29f77eaa5732cc76eb0c716c808f063c195e2aa886f4993ea33f51d29d",
+      "returncode": 0,
+      "seconds": 0.9767264169640839,
+      "statistics": {
+        "cache": {
+          "currsize": 58,
+          "hits": 158,
+          "maxsize": 256,
+          "misses": 58
+        },
+        "mode": "cached"
+      },
+      "trial": 3
+    },
+    {
+      "mode": "baseline",
+      "output_sha256": "31116d29f77eaa5732cc76eb0c716c808f063c195e2aa886f4993ea33f51d29d",
+      "returncode": 0,
+      "seconds": 1.2514010828454047,
+      "statistics": {
+        "cache": {
+          "currsize": 0,
+          "hits": 0,
+          "maxsize": 256,
+          "misses": 0
+        },
+        "mode": "baseline"
+      },
+      "trial": 4
+    },
+    {
+      "mode": "cached",
+      "output_sha256": "31116d29f77eaa5732cc76eb0c716c808f063c195e2aa886f4993ea33f51d29d",
+      "returncode": 0,
+      "seconds": 0.966591291828081,
+      "statistics": {
+        "cache": {
+          "currsize": 58,
+          "hits": 158,
+          "maxsize": 256,
+          "misses": 58
+        },
+        "mode": "cached"
+      },
+      "trial": 4
+    }
+  ],
+  "schema": "native-history-setup-benchmark/v1",
+  "source_sha256": {
+    "SOURCES.json": "a8665633efa339149c3e1b936b0770a8f1de1a914538ed79c85e01f3f8a2e6c7",
+    "benchmark.py": "97bc142fe47b2911dae2f7695cae9cbb106bf3235a09dd9d66cea9ded6238f78",
+    "run.py": "78a2b5038354facf910908af2d9a533d759d444b79edd5ffce1122916cf77606"
+  },
+  "summary": {
+    "baseline": {
+      "maximum_seconds": 1.304742000065744,
+      "median_seconds": 1.2796051658224314,
+      "minimum_seconds": 1.246776457875967
+    },
+    "cached": {
+      "maximum_seconds": 1.0426137920003384,
+      "median_seconds": 0.9767264169640839,
+      "minimum_seconds": 0.9610304171219468
+    },
+    "fraction_less_time": 0.23669703510745088
+  },
+  "timed_region": "fresh interpreter, authentication and complete direct caller"
+}

research/native-history-setup-cache/README.md

@@ -0,0 +1,133 @@
+# Reuse native-map setup in the complete full-history caller
+
+The complete PR5447 direct-history caller takes **23.67% less wall time** with
+an invocation-owned cache of exact native-map construction. Five alternating
+fresh-process runs per side give medians of 1.279605 s and 0.976726 s. Baseline
+range is 1.246776 to 1.304742 s; cached range is 0.961030 to 1.042614 s. All ten
+complete executions succeed with identical scientific stdout bytes.
+[BENCHMARK.json](BENCHMARK.json) is generated by the retained harness and binds
+its source hashes.
+
+Claim type: scoped implementation-preservation proof, finite conformance and
+local performance evidence. This is a newly posed runtime question on the
+unchanged predecessor caller. No new kernel theorem, kappa or supplier admission
+is claimed.
+
+## Change and preservation argument
+
+The complete caller initializes three separate native registries for its
+provider-backward, source-backward and source-forward cases. They request 216
+native constructions but use only 58 distinct binary maps. The adapter reuses
+158 exact results. It neither caches history rows nor shares native registries.
+
+The pinned `monomial.native(columns)` function is deterministic. It reads nine
+binary map columns and returns one immutable 1,024-byte signed permutation.
+The key snapshots all nine columns, so equal keys have equal outputs. Each miss
+calls the original function. Each hit returns the bytes that function would
+construct. Mutating an input list cannot change a stored tuple key; callers
+cannot mutate returned bytes. Exceptions are not cached. Capacity is 256 entries,
+and the cache is cleared on every invocation exit, including exceptional exits.
+
+Every `intern` call therefore receives the original bytes in the original order.
+Operator IDs, products, scalar rows, histories and final output remain the same.
+This implication does not depend on a small-width sample. The adapter
+authenticates all seven external inputs before installing reuse and runs the
+original `verify_kernel.py` unchanged. Its three complete cases, literal
+permutation checks, equations, frozen-result comparison and four original
+controls still execute.
+
+## Evidence and measurement
+
+The new verifier compares original and cached construction on 300 distinct
+invertible nine-dimensional unitriangular maps. It checks full-key reuse,
+256-entry capacity, eviction, mutable lists, restored contents and independent
+cache owners. Six malformed inputs reject at the boundary. Two singular-map
+executions retain the original exception and leave no cached value.
+
+Both complete callers must agree byte for byte and match every field of the
+predecessor's frozen result. Their stdout SHA256 is
+`31116d29f77eaa5732cc76eb0c716c808f063c195e2aa886f4993ea33f51d29d`.
+The cached path must record exactly 58 misses and 158 hits. The verifier
+authenticates the generated measurement against its implementation, harness and
+manifest, and recomputes the measurement summary. Timing itself is not a CI gate.
+
+Local measurement uses CPython 3.12.15, standard library, macOS arm64 and 18
+logical CPUs. Initial load averages were 2.86, 2.41 and 1.74; other work could
+run concurrently. Alternating sides exposes both to that load. Their observed
+ranges are disjoint. Each subprocess includes startup, authentication and the
+entire direct caller inside the timed region. Baseline uses the same adapter
+without installing reuse. Neither side skips checks or uses a cross-process
+cache. All ten return codes are zero. Profiler runs are excluded from samples.
+
+The initial complete profile identifies 216 `monomial.native` calls; a separate
+cached profile has 58. Fraction arithmetic remains a larger cost after reuse.
+The repeated setup is only part of the complete caller. This explains a bounded
+end-to-end saving and does not predict a twofold speedup. The first improvement
+is packaged without further arithmetic optimization.
+
+## Reproduce
+
+From the repository root on the recorded machine:
+
+```sh
+/Users/nick/git/integer-mult-bounds/.venv/bin/python3.12 -W error -B research/native-history-setup-cache/verify.py
+/Users/nick/git/integer-mult-bounds/.venv/bin/python3.12 -W error -O -B research/native-history-setup-cache/verify.py
+/Users/nick/git/integer-mult-bounds/.venv/bin/python3.12 -B scripts/research_package.py --package research/native-history-setup-cache
+/Users/nick/git/integer-mult-bounds/.venv/bin/python3.12 -B scripts/check_research_commit.py --package research/native-history-setup-cache --check-source-history
+```
+
+Direct status is `PASS_NATIVE_HISTORY_SETUP_CACHE`: 300 maps, eight rejecting
+executions, two complete callers, 58 misses and 158 hits. Registered status is
+`PASS_REGISTERED_RESEARCH_PACKAGE`. Other machines may substitute CPython 3.12.
+
+Run the complete cached caller or generate a new measurement outside the checkout:
+
+```sh
+python3.12 -W error -B research/native-history-setup-cache/run.py --mode cached
+python3.12 -W error -B research/native-history-setup-cache/benchmark.py --output /absolute/scratch/new-benchmark.json
+```
+
+The harness refuses to overwrite its output. For a diagnostic profile, insert
+`-m cProfile -o /absolute/scratch/profile.pstats` before `run.py`. Profiler timings
+are not comparable to the retained uninstrumented samples.
+
+## Limits and code discipline
+
+Timing is local to this complete direct caller and environment. Other PR5447
+commands, selected cold replay, physical native admission, compiler closure,
+ownership, precision/tape and analytic recovery remain unrun or inherited.
+Independent human and formal proof-kernel review are unclaimed. No predecessor
+source, credit line or selected-frontier file changes.
+
+The adapter accepts nine integer columns in `[0,512)`; invertibility retains the
+original check. Retained payload is at most 256 times 1,024 bytes plus bounded
+keys and bookkeeping. Python allocation is a managed-runtime deviation from
+fixed allocation. Finite loops are bounded by 300 maps, 36 triangular positions,
+six malformed inputs and two singular executions. Benchmarking runs ten children
+with 120-second timeouts. New functions fit within 60 lines. Standard import
+loading and `runpy` load only the pinned modules, a bounded dynamic-loading
+deviation with no generated code. Explicit exceptions remain active under `-O`.
+Warning-strict normal and optimized replay supply the interpreted diagnostics.
+
+## Open next questions
+
+1. Can exact rational row arithmetic reduce the remaining complete caller cost?
+   Start with a new cached profile and preserve all scientific output. No such
+   implementation or measurement is supplied here.
+2. Should a future native compiler expose an invocation-owned operator factory?
+   First specify its exact key, lifetime and source invalidation on one consumer.
+   Shared API design and other caller replays remain open.
+
+Structured questions are in [TASK.json](TASK.json).
+
+## Credit and sources
+
+The adapter, preservation argument, conformance checks and measurements were
+produced by Nick Petalas. Memoization is classical; priority for it is unclaimed.
+
+Credit THU New Media Research Center / thu-nmrc PR5447 for the native algebra,
+complete caller, kernel proofs and controls; Mark / Matarener PR5427 for event
+and ownership semantics; Dillon Amburgey PR5402, Cao Yifan PR4610, Cinna Davis,
+zxv, David Ganz, Jacob Alan Sussman, Ali Romman and the retained
+native/query/compiler lineage. Their original files and credit are unchanged.
+Exact main-branch pins and hashes are in [SOURCES.json](SOURCES.json). Apache-2.0.

research/native-history-setup-cache/SOURCES.json

@@ -0,0 +1,60 @@
+{
+  "schema_version": 1,
+  "base_commit": "2485805a6d47f36ea83ddde6327d453ac82d44dc",
+  "files": [
+    {
+      "path": "certificates/selected-result.json",
+      "usage": "provenance",
+      "git_blob": "d59b5035e7b4cbaa9d688a1d32dd0d6eb79c466a",
+      "sha256": "9b8d0eea95b3bf0dfcb7938a732a6439b87a2685dec2a5ab9e525b3b34debbfa"
+    },
+    {
+      "path": "docs/research/agent-task-catalog.json",
+      "usage": "provenance",
+      "git_blob": "3404fbdd5e7529a0674eef26114b5071e4586cec",
+      "sha256": "ead6cdc9d0af6d58e8a49744d8ec8c7c15e13d2ba93c276c434bcf9b51b93a81"
+    },
+    {
+      "path": "research/kappa-round3-warp-20261022/ACTUAL-WORD.json.gz",
+      "usage": "input",
+      "git_blob": "46d929df83817b85a5387bcdb9745e0627d6366e",
+      "sha256": "7679daa4ec6a40f56fc7f0202379a6f454310e94945aba4f48beaaaafb838d05"
+    },
+    {
+      "path": "research/native-rank8-helper-provider/WORD.json.gz",
+      "usage": "input",
+      "git_blob": "f3111b47e89496da8ca0ee45b2e4b258c0634d8e",
+      "sha256": "7dc356001ed597ad6fc9fd3e0bc6ff7e3ec3a580a579a956827aa17fccfdd2dc"
+    },
+    {
+      "path": "research/physical-zqy-scalar-span/KERNEL-RESULT.json",
+      "usage": "input",
+      "git_blob": "f57f4accbc0d334db480d2b31db4185420654567",
+      "sha256": "0d56847d579aa195c6ee462a631a5ccdbfbb128dccc26503dbdc481963b6cab8"
+    },
+    {
+      "path": "research/physical-zqy-scalar-span/KERNEL.json",
+      "usage": "input",
+      "git_blob": "cd699656f9192ccf6d462a45617b5e56d6799dfb",
+      "sha256": "ad403cdd98975ae9a8033afff5a86f96d055e44c5be19c1f171387cf26cd3b81"
+    },
+    {
+      "path": "research/physical-zqy-scalar-span/monomial.py",
+      "usage": "input",
+      "git_blob": "1a4483808478808784fe6b3e0b88cdc959752a66",
+      "sha256": "87cdb1688b841891eccefdcaebe67b97a6c86b7af676bbbb667d1d5ee5d45d3e"
+    },
+    {
+      "path": "research/physical-zqy-scalar-span/native.py",
+      "usage": "input",
+      "git_blob": "0784aa84b7bd49454e57530be3ded522313a91f8",
+      "sha256": "1bad1ffe747c3977d1a6bc1cfa9faa4127e38cbf49d7fcf4ae8060fbc7f67d4e"
+    },
+    {
+      "path": "research/physical-zqy-scalar-span/verify_kernel.py",
+      "usage": "input",
+      "git_blob": "036d1f199f03a2a3e2792e33d28c6e5a7e4d26c2",
+      "sha256": "29af4fd8536aa10d8ed0459087744fd077113073ec7e3a682fde970593fd857d"
+    }
+  ]
+}