From 2825d145ef2ea2297e31553c08a04f57a1be3925 Mon Sep 17 00:00:00 2001 From: =?UTF-8?q?Val=20Alexander=20=F0=9F=91=91?= Date: Thu, 12 Nov 2026 04:23:40 -0600 Subject: [PATCH 1/4] docs: record CF-8 retention evidence and protected delivery --- docs/reviews/2026-11-12-cf8-retention.json | 880 ++++++++++++++++++ docs/reviews/2026-11-12-cf8-retention.md | 48 + ...-11-12-local-model-refresh-preservation.md | 6 + docs/status.md | 8 +- 4 files changed, 939 insertions(+), 3 deletions(-) create mode 100644 docs/reviews/2026-11-12-cf8-retention.json create mode 100644 docs/reviews/2026-11-12-cf8-retention.md diff --git a/docs/reviews/2026-11-12-cf8-retention.json b/docs/reviews/2026-11-12-cf8-retention.json new file mode 100644 index 00000000..e2010d12 --- /dev/null +++ b/docs/reviews/2026-11-12-cf8-retention.json @@ -0,0 +1,880 @@ +{ + "at": "2026-11-12T09:40:39.871112+00:00", + "currentSource": "44fb41eecf718cf6648b7c8ee95cab1e85e0e072", + "historicalSource": "89c98f280292b13fd9b1c7af1bf2fe65cd2fa0d1", + "groups": { + "historical": { + "history": [ + { + "replicate": 1, + "host": { + "cpu": "Apple M5 Max", + "cores": 18, + "memoryBytes": 137438953472, + "release": "27.2.0", + "arch": "arm64", + "node": "v24.22.0", + "loadAverage": [ + 9.4541015625, + 11.06494140625, + 11.78466796875 + ] + }, + "native": { + "appVersion": "0.2.0", + "electron": "44.4.5", + "chromium": "152.0.7977.130" + }, + "rendererForcedDetachedBeforeAfterBytes": [ + 17336928, + 15012276 + ], + "rendererRetainedEndpointDeltaBytes": -2324652, + "mainNaturalHeapBeforeAfterBytes": [ + 32592720, + 29485112 + ], + "mainNaturalPrivateBeforeAfterKiB": [ + 84449, + 105025 + ], + "beforeDOM": { + "documents": 1, + "nodes": 317, + "jsEventListeners": 269 + }, + "afterDOM": { + "documents": 1, + "nodes": 317, + "jsEventListeners": 230 + }, + "measuredOpens": 63, + "warmupOpens": 3, + "naturalSampledRendererHeapRangeBytes": [ + 30501696, + 116625604 + ], + "savedBytes": [ + 2994024, + 2994024, + 2994024 + ], + "sourceReceipt": { + "file": "cf8-historical-natural-89c-replicate-1.json", + "sha256": "c302735a90fa3b74a7b0487cec2a456816b1110fc70575d15e8139d74bbe747c", + "startedAt": "2026-11-12T09:34:32.690Z", + "finishedAt": "2026-11-12T09:34:46.338Z", + "state": "observations-complete", + "appClose": "returned" + } + }, + { + "replicate": 2, + "host": { + "cpu": "Apple M5 Max", + "cores": 18, + "memoryBytes": 137438953472, + "release": "27.2.0", + "arch": "arm64", + "node": "v24.22.0", + "loadAverage": [ + 10.6201171875, + 11.25439453125, + 11.83984375 + ] + }, + "native": { + "appVersion": "0.2.0", + "electron": "44.4.5", + "chromium": "152.0.7977.130" + }, + "rendererForcedDetachedBeforeAfterBytes": [ + 14349376, + 15011772 + ], + "rendererRetainedEndpointDeltaBytes": 662396, + "mainNaturalHeapBeforeAfterBytes": [ + 32720296, + 29708988 + ], + "mainNaturalPrivateBeforeAfterKiB": [ + 84385, + 105105 + ], + "beforeDOM": { + "documents": 1, + "nodes": 312, + "jsEventListeners": 230 + }, + "afterDOM": { + "documents": 1, + "nodes": 317, + "jsEventListeners": 230 + }, + "measuredOpens": 63, + "warmupOpens": 3, + "naturalSampledRendererHeapRangeBytes": [ + 27623008, + 117432380 + ], + "savedBytes": [ + 2994024, + 2994024, + 2994024 + ], + "sourceReceipt": { + "file": "cf8-historical-natural-89c-replicate-2.json", + "sha256": "93c5a394508ac04d5bf7cb2a741013318078e027f5193290a5e90886664a914f", + "startedAt": "2026-11-12T09:34:46.396Z", + "finishedAt": "2026-11-12T09:34:59.849Z", + "state": "observations-complete", + "appClose": "returned" + } + }, + { + "replicate": 3, + "host": { + "cpu": "Apple M5 Max", + "cores": 18, + "memoryBytes": 137438953472, + "release": "27.2.0", + "arch": "arm64", + "node": "v24.22.0", + "loadAverage": [ + 10.205078125, + 11.14111328125, + 11.78857421875 + ] + }, + "native": { + "appVersion": "0.2.0", + "electron": "44.4.5", + "chromium": "152.0.7977.130" + }, + "rendererForcedDetachedBeforeAfterBytes": [ + 14349524, + 15017540 + ], + "rendererRetainedEndpointDeltaBytes": 668016, + "mainNaturalHeapBeforeAfterBytes": [ + 32684652, + 34822572 + ], + "mainNaturalPrivateBeforeAfterKiB": [ + 84049, + 114337 + ], + "beforeDOM": { + "documents": 1, + "nodes": 312, + "jsEventListeners": 230 + }, + "afterDOM": { + "documents": 1, + "nodes": 317, + "jsEventListeners": 230 + }, + "measuredOpens": 63, + "warmupOpens": 3, + "naturalSampledRendererHeapRangeBytes": [ + 27657652, + 83517336 + ], + "savedBytes": [ + 2994024, + 2994024, + 2994024 + ], + "sourceReceipt": { + "file": "cf8-historical-natural-89c-replicate-3.json", + "sha256": "ca10b19772e3b4b4b7cb075d424bc64c0fe28d94344dec5863e814a296dbe52c", + "startedAt": "2026-11-12T09:34:59.900Z", + "finishedAt": "2026-11-12T09:35:15.153Z", + "state": "observations-complete", + "appClose": "returned" + } + } + ], + "stream": [ + { + "replicate": 1, + "host": { + "cpu": "Apple M5 Max", + "cores": 18, + "memoryBytes": 137438953472, + "release": "27.2.0", + "arch": "arm64", + "node": "v24.22.0", + "loadAverage": [ + 11.296875, + 11.34423828125, + 11.849609375 + ] + }, + "native": { + "appVersion": "0.2.0", + "electron": "44.4.5", + "chromium": "152.0.7977.130" + }, + "rendererForcedDetachedBeforeAfterBytes": [ + 8688596, + 76521192 + ], + "rendererRetainedEndpointDeltaBytes": 67832596, + "mainNaturalHeapBeforeAfterBytes": [ + 72123300, + 202668900 + ], + "mainNaturalPrivateBeforeAfterKiB": [ + 169825, + 294529 + ], + "beforeDOM": { + "documents": 1, + "nodes": 243, + "jsEventListeners": 220 + }, + "afterDOM": { + "documents": 1, + "nodes": 371, + "jsEventListeners": 241 + }, + "measuredStreams": 21, + "warmupStreams": 1, + "logicalMeasuredBytes": 66060288, + "observedWindowBytesMeasured": 1684719770, + "maxSampledOutputBytes": 3145728, + "fixtureCodingStarts": 22, + "fixtureCodingFinishes": 22, + "naturalSampledRendererHeapRangeBytes": [ + 37475912, + 249385848 + ], + "sourceReceipt": { + "file": "cf8-historical-log-89c-replicate-1.json", + "sha256": "994bcc6c1eb517852524da27f3bc194b56d2e59be1804c15395df72ad771217a", + "startedAt": "2026-11-12T09:35:15.240Z", + "finishedAt": "2026-11-12T09:35:51.356Z", + "state": "observations-complete", + "appClose": "returned" + } + }, + { + "replicate": 2, + "host": { + "cpu": "Apple M5 Max", + "cores": 18, + "memoryBytes": 137438953472, + "release": "27.2.0", + "arch": "arm64", + "node": "v24.22.0", + "loadAverage": [ + 11.66796875, + 11.443359375, + 11.86474609375 + ] + }, + "native": { + "appVersion": "0.2.0", + "electron": "44.4.5", + "chromium": "152.0.7977.130" + }, + "rendererForcedDetachedBeforeAfterBytes": [ + 5542560, + 73373420 + ], + "rendererRetainedEndpointDeltaBytes": 67830860, + "mainNaturalHeapBeforeAfterBytes": [ + 72134436, + 202845984 + ], + "mainNaturalPrivateBeforeAfterKiB": [ + 165073, + 291442 + ], + "beforeDOM": { + "documents": 1, + "nodes": 243, + "jsEventListeners": 220 + }, + "afterDOM": { + "documents": 1, + "nodes": 371, + "jsEventListeners": 241 + }, + "measuredStreams": 21, + "warmupStreams": 1, + "logicalMeasuredBytes": 66060288, + "observedWindowBytesMeasured": 1684719770, + "maxSampledOutputBytes": 3145728, + "fixtureCodingStarts": 22, + "fixtureCodingFinishes": 22, + "naturalSampledRendererHeapRangeBytes": [ + 37565804, + 214739488 + ], + "sourceReceipt": { + "file": "cf8-historical-log-89c-replicate-2.json", + "sha256": "788da1a4bf82119616be1eb878e697b241ab88c9a6367b8250978f026ba71ef7", + "startedAt": "2026-11-12T09:35:51.419Z", + "finishedAt": "2026-11-12T09:36:26.842Z", + "state": "observations-complete", + "appClose": "returned" + } + }, + { + "replicate": 3, + "host": { + "cpu": "Apple M5 Max", + "cores": 18, + "memoryBytes": 137438953472, + "release": "27.2.0", + "arch": "arm64", + "node": "v24.22.0", + "loadAverage": [ + 10.150390625, + 11.091796875, + 11.71826171875 + ] + }, + "native": { + "appVersion": "0.2.0", + "electron": "44.4.5", + "chromium": "152.0.7977.130" + }, + "rendererForcedDetachedBeforeAfterBytes": [ + 5549952, + 73340480 + ], + "rendererRetainedEndpointDeltaBytes": 67790528, + "mainNaturalHeapBeforeAfterBytes": [ + 72136404, + 202849580 + ], + "mainNaturalPrivateBeforeAfterKiB": [ + 165617, + 291506 + ], + "beforeDOM": { + "documents": 1, + "nodes": 243, + "jsEventListeners": 220 + }, + "afterDOM": { + "documents": 1, + "nodes": 371, + "jsEventListeners": 241 + }, + "measuredStreams": 21, + "warmupStreams": 1, + "logicalMeasuredBytes": 66060288, + "observedWindowBytesMeasured": 1684719770, + "maxSampledOutputBytes": 3145728, + "fixtureCodingStarts": 22, + "fixtureCodingFinishes": 22, + "naturalSampledRendererHeapRangeBytes": [ + 21742188, + 261993784 + ], + "sourceReceipt": { + "file": "cf8-historical-log-89c-replicate-3.json", + "sha256": "a051ae78ac2ea2a9533f31c7e35cfcc5fd02b6e21054968a3ca80d8ce4242877", + "startedAt": "2026-11-12T09:36:26.905Z", + "finishedAt": "2026-11-12T09:37:01.472Z", + "state": "observations-complete", + "appClose": "returned" + } + } + ] + }, + "current": { + "history": [ + { + "replicate": 1, + "host": { + "cpu": "Apple M5 Max", + "cores": 18, + "memoryBytes": 137438953472, + "release": "27.2.0", + "arch": "arm64", + "node": "v24.22.0", + "loadAverage": [ + 7.2900390625, + 8.02001953125, + 9.60498046875 + ] + }, + "native": { + "electron": "44.4.5", + "chromium": "152.0.7977.130" + }, + "rendererForcedDetachedBeforeAfterBytes": [ + 14952760, + 15791412 + ], + "rendererRetainedEndpointDeltaBytes": 838652, + "mainNaturalHeapBeforeAfterBytes": [ + 37707428, + 50931160 + ], + "mainNaturalPrivateBeforeAfterKiB": [ + 90113, + 120161 + ], + "beforeDOM": { + "documents": 1, + "nodes": 320, + "jsEventListeners": 269 + }, + "afterDOM": { + "documents": 1, + "nodes": 320, + "jsEventListeners": 230 + }, + "measuredOpens": 63, + "warmupOpens": 3, + "naturalSampledRendererHeapRangeBytes": [ + 27200848, + 156610040 + ], + "savedBytes": [ + 2994024, + 2994024, + 2994024 + ], + "sourceReceipt": { + "file": "cf8-natural-retention-44fb-replicate-1.json", + "sha256": "87c7bcdb935e7ed72ba21ca0537e8ab595023eefea7deb9adf5e5a7bfcc4263c", + "startedAt": "2026-11-12T08:41:00.503Z", + "finishedAt": "2026-11-12T08:41:16.900Z", + "state": "observations-complete", + "appClose": "returned" + } + }, + { + "replicate": 2, + "host": { + "cpu": "Apple M5 Max", + "cores": 18, + "memoryBytes": 137438953472, + "release": "27.2.0", + "arch": "arm64", + "node": "v24.22.0", + "loadAverage": [ + 7.84619140625, + 8.1015625, + 9.60498046875 + ] + }, + "native": { + "electron": "44.4.5", + "chromium": "152.0.7977.130" + }, + "rendererForcedDetachedBeforeAfterBytes": [ + 14850864, + 15787868 + ], + "rendererRetainedEndpointDeltaBytes": 937004, + "mainNaturalHeapBeforeAfterBytes": [ + 37581204, + 49643892 + ], + "mainNaturalPrivateBeforeAfterKiB": [ + 90193, + 119105 + ], + "beforeDOM": { + "documents": 1, + "nodes": 315, + "jsEventListeners": 230 + }, + "afterDOM": { + "documents": 1, + "nodes": 320, + "jsEventListeners": 230 + }, + "measuredOpens": 63, + "warmupOpens": 3, + "naturalSampledRendererHeapRangeBytes": [ + 26923300, + 99632588 + ], + "savedBytes": [ + 2994024, + 2994024, + 2994024 + ], + "sourceReceipt": { + "file": "cf8-natural-retention-44fb-replicate-2.json", + "sha256": "b6da6766c79ecb988a0568dd08895da207a4efb84989755d32e0cb35cd323e7d", + "startedAt": "2026-11-12T08:41:16.967Z", + "finishedAt": "2026-11-12T08:41:32.494Z", + "state": "observations-complete", + "appClose": "returned" + } + }, + { + "replicate": 3, + "host": { + "cpu": "Apple M5 Max", + "cores": 18, + "memoryBytes": 137438953472, + "release": "27.2.0", + "arch": "arm64", + "node": "v24.22.0", + "loadAverage": [ + 8.03369140625, + 8.12890625, + 9.58740234375 + ] + }, + "native": { + "electron": "44.4.5", + "chromium": "152.0.7977.130" + }, + "rendererForcedDetachedBeforeAfterBytes": [ + 14842900, + 15804820 + ], + "rendererRetainedEndpointDeltaBytes": 961920, + "mainNaturalHeapBeforeAfterBytes": [ + 37726740, + 49530228 + ], + "mainNaturalPrivateBeforeAfterKiB": [ + 90737, + 118529 + ], + "beforeDOM": { + "documents": 1, + "nodes": 315, + "jsEventListeners": 230 + }, + "afterDOM": { + "documents": 1, + "nodes": 320, + "jsEventListeners": 230 + }, + "measuredOpens": 63, + "warmupOpens": 3, + "naturalSampledRendererHeapRangeBytes": [ + 27649144, + 106447948 + ], + "savedBytes": [ + 2994024, + 2994024, + 2994024 + ], + "sourceReceipt": { + "file": "cf8-natural-retention-44fb-replicate-3.json", + "sha256": "afcee7b05ea34cecb8fc3aacf8a87544d34b37aa65bedd6e0505e50499038e62", + "startedAt": "2026-11-12T08:41:32.546Z", + "finishedAt": "2026-11-12T08:41:48.663Z", + "state": "observations-complete", + "appClose": "returned" + } + } + ], + "stream": [ + { + "replicate": 1, + "host": { + "cpu": "Apple M5 Max", + "cores": 18, + "memoryBytes": 137438953472, + "release": "27.2.0", + "arch": "arm64", + "node": "v24.22.0", + "loadAverage": [ + 10.048828125, + 9.97509765625, + 9.75 + ] + }, + "native": { + "electron": "44.4.5", + "chromium": "152.0.7977.130" + }, + "rendererForcedDetachedBeforeAfterBytes": [ + 6011144, + 9448940 + ], + "rendererRetainedEndpointDeltaBytes": 3437796, + "mainNaturalHeapBeforeAfterBytes": [ + 47736696, + 47230308 + ], + "mainNaturalPrivateBeforeAfterKiB": [ + 148001, + 162305 + ], + "beforeDOM": { + "documents": 1, + "nodes": 246, + "jsEventListeners": 220 + }, + "afterDOM": { + "documents": 1, + "nodes": 374, + "jsEventListeners": 241 + }, + "measuredStreams": 21, + "warmupStreams": 1, + "logicalMeasuredBytes": 66060288, + "observedWindowBytesMeasured": 13945579, + "maxSampledOutputBytes": 262144, + "fixtureCodingStarts": 22, + "fixtureCodingFinishes": 22, + "naturalSampledRendererHeapRangeBytes": [ + 10984832, + 29487420 + ], + "sourceReceipt": { + "file": "cf8-log-retention-44fb-replicate-1.json", + "sha256": "13eb1cdba0c42b3adbb5a5fc01a9b754b4a06237fe27cee5c6527860425d3b2a", + "startedAt": "2026-11-12T08:54:24.015Z", + "finishedAt": "2026-11-12T08:54:58.034Z", + "state": "observations-complete", + "appClose": "returned" + } + }, + { + "replicate": 2, + "host": { + "cpu": "Apple M5 Max", + "cores": 18, + "memoryBytes": 137438953472, + "release": "27.2.0", + "arch": "arm64", + "node": "v24.22.0", + "loadAverage": [ + 12.4990234375, + 10.5673828125, + 9.9716796875 + ] + }, + "native": { + "electron": "44.4.5", + "chromium": "152.0.7977.130" + }, + "rendererForcedDetachedBeforeAfterBytes": [ + 6012644, + 9735108 + ], + "rendererRetainedEndpointDeltaBytes": 3722464, + "mainNaturalHeapBeforeAfterBytes": [ + 47740312, + 58711396 + ], + "mainNaturalPrivateBeforeAfterKiB": [ + 147921, + 175153 + ], + "beforeDOM": { + "documents": 1, + "nodes": 246, + "jsEventListeners": 220 + }, + "afterDOM": { + "documents": 1, + "nodes": 374, + "jsEventListeners": 241 + }, + "measuredStreams": 21, + "warmupStreams": 1, + "logicalMeasuredBytes": 66060288, + "observedWindowBytesMeasured": 13945579, + "maxSampledOutputBytes": 262144, + "fixtureCodingStarts": 22, + "fixtureCodingFinishes": 22, + "naturalSampledRendererHeapRangeBytes": [ + 6837120, + 34374320 + ], + "sourceReceipt": { + "file": "cf8-log-retention-44fb-replicate-2.json", + "sha256": "6dc99960390fe74a31f10150ff97396f9b80eb7bf43b78ec333b19cf342fbe98", + "startedAt": "2026-11-12T08:54:58.102Z", + "finishedAt": "2026-11-12T08:55:35.009Z", + "state": "observations-complete", + "appClose": "returned" + } + }, + { + "replicate": 3, + "host": { + "cpu": "Apple M5 Max", + "cores": 18, + "memoryBytes": 137438953472, + "release": "27.2.0", + "arch": "arm64", + "node": "v24.22.0", + "loadAverage": [ + 14.150390625, + 11.15576171875, + 10.208984375 + ] + }, + "native": { + "electron": "44.4.5", + "chromium": "152.0.7977.130" + }, + "rendererForcedDetachedBeforeAfterBytes": [ + 6005948, + 9207732 + ], + "rendererRetainedEndpointDeltaBytes": 3201784, + "mainNaturalHeapBeforeAfterBytes": [ + 47111688, + 32241836 + ], + "mainNaturalPrivateBeforeAfterKiB": [ + 146289, + 148866 + ], + "beforeDOM": { + "documents": 1, + "nodes": 246, + "jsEventListeners": 220 + }, + "afterDOM": { + "documents": 1, + "nodes": 374, + "jsEventListeners": 241 + }, + "measuredStreams": 21, + "warmupStreams": 1, + "logicalMeasuredBytes": 66060288, + "observedWindowBytesMeasured": 13945579, + "maxSampledOutputBytes": 262144, + "fixtureCodingStarts": 22, + "fixtureCodingFinishes": 22, + "naturalSampledRendererHeapRangeBytes": [ + 9392824, + 29843580 + ], + "sourceReceipt": { + "file": "cf8-log-retention-44fb-replicate-3.json", + "sha256": "6dd03d46aa221e16476255c9600a81bfc8ad6b7c3d8f14dfc0c659fc77e287d2", + "startedAt": "2026-11-12T08:55:35.569Z", + "finishedAt": "2026-11-12T08:56:19.531Z", + "state": "observations-complete", + "appClose": "returned" + } + } + ] + } + }, + "limits": [ + "Descriptive sequential source comparison, not randomized or component-causal; timestamps differ but fixed shapes/counts match", + "Common current pinned dependencies/runtime, not historical shipped binary; verified old source distinct from parent driver Git HEAD", + "Historical-only shallow DOM Text.length observation proves full ASCII output; no ordinary latency comparison", + "Each streaming cycle adds legitimate history; uniform all-x data is not a generic log distribution", + "Main always naturalGC; renderer forced only matched detached UI phases; no exhaustive async settlement guarantee", + "These measurements alone do not establish full S5, arbitrary memory ceilings, leak freedom, native real-provider or human acceptance; see separate literal S5 reconciliation", + "Reported PIDabsence/appclose does not prove exhaustive process-tree closure" + ], + "schemaVersion": 1, + "kind": "CF8 named-scope measurement observations; full S5 assessment also checks delivered structural caps and child regressions", + "units": { + "rendererAndMainNodeMemory": "bytes", + "mainNativePrivate": "KiB", + "logicalAndWindowOutput": "bytes" + }, + "sourceReceiptDisclosure": "Per-replicate originals, archived drivers, manifests and first setup failures are retained privately. This public aggregate omits local executable/app/userData paths and process IDs; numeric observations are unchanged.", + "naturalDetach5s": { + "historical": [ + { + "replicate": 1, + "naturalDetachedAfter5s": { + "documents": 1, + "nodes": 77453, + "jsEventListeners": 6548 + }, + "naturalDetachedHeapAfter5s": 41895184 + }, + { + "replicate": 2, + "naturalDetachedAfter5s": { + "documents": 1, + "nodes": 16144, + "jsEventListeners": 4620 + }, + "naturalDetachedHeapAfter5s": 37617744 + }, + { + "replicate": 3, + "naturalDetachedAfter5s": { + "documents": 1, + "nodes": 16144, + "jsEventListeners": 4620 + }, + "naturalDetachedHeapAfter5s": 43352336 + } + ], + "current": [ + { + "replicate": 1, + "naturalDetachedAfter5s": { + "documents": 1, + "nodes": 68313, + "jsEventListeners": 17099 + }, + "naturalDetachedHeapAfter5s": 76279220 + }, + { + "replicate": 2, + "naturalDetachedAfter5s": { + "documents": 1, + "nodes": 51298, + "jsEventListeners": 11940 + }, + "naturalDetachedHeapAfter5s": 60749228 + }, + { + "replicate": 3, + "naturalDetachedAfter5s": { + "documents": 1, + "nodes": 51298, + "jsEventListeners": 11940 + }, + "naturalDetachedHeapAfter5s": 66270116 + } + ] + }, + "reviewProvenance": { + "originalComparisonSha256": "7f4ae658f579263f316df5324ab4665f97f474520a80ca189e3f628fc75fd587", + "terminalComparisonReviewSha256": "9668574a5e5fb81a565154f34cfd27739acab2d265222b927eddd027dd0f65ce", + "correctedReportSha256": "815369f6e46b8a8128467884971719e4afb64ea60c5eb3253b9530f081bb3ac9", + "followupReviewSha256": "149d0835868d227afc2f06ecca908370e3b966bb4ef2ae2908cd338df6ff9fe7" + }, + "s5Reconciliation": { + "literalS5Supported": true, + "main": "558b54f8fa0cb3679852c6ece2b07e23a3b10ffb", + "scope": "Named fixture measurement, observed retention and delivered structural/lifecycle bounds plus child regressions; no global heap/RSS guarantee", + "independentReviewSha256": "397b156f542b16a899075f98a461fde96833abe30579f0e2b13427d768210312", + "publicReceipt": "https://github.com/TypeSafeAI/modex/issues/281#issuecomment-6820945583", + "observedIssueChecked": 9, + "observedIssueTotal": 27, + "onlyCheckboxChanged": "S5", + "separateOpenGate": "Reproducible capture/performance acceptance, native/provider/device/human acceptance and remaining dependencies", + "retainedDataCaps": { + "previewBytes": 262144, + "pageBytes": 262144, + "perToolRawLogBytes": 10485760 + }, + "implementationPaths": [ + "apps/desktop/src/shared/evidence.ts", + "apps/desktop/src/main/engine/tool-logs.ts", + "apps/desktop/src/main/engine/backends/codex.ts", + "apps/desktop/src/main/engine/runner.ts" + ], + "queuePassedTests": [ + "apps/desktop/test/evidence.test.ts", + "apps/desktop/test/tool-logs.test.ts", + "apps/desktop/e2e/transcript-measure.spec.ts", + "apps/desktop/e2e/conversation-fidelity.spec.ts" + ] + } +} diff --git a/docs/reviews/2026-11-12-cf8-retention.md b/docs/reviews/2026-11-12-cf8-retention.md new file mode 100644 index 00000000..88cd1a1f --- /dev/null +++ b/docs/reviews/2026-11-12-cf8-retention.md @@ -0,0 +1,48 @@ +# CF-8 repeated history and log-memory observations + +These controls compare verified pre-CF source `89c98f280292b13fd9b1c7af1bf2fe65cd2fa0d1` with current source `44fb41eecf718cf6648b7c8ee95cab1e85e0e072`. PR #503 changes tests and documentation, so these differences are not attributed to that PR or to a new repair. Both groups use the same installed pinned dependencies and Electron runtime; the baseline is a source export, not a historical shipped binary. + +The named host is Apple M5 Max, 18 cores, 128 GiB RAM, arm64, kernel release `27.2.0`, driver Node `v24.22.0`, Electron `44.4.5`. Load averages and exact observation times are retained per replicate in the [public aggregate](2026-11-12-cf8-retention.json). Groups ran sequentially, without randomization, with fixed shapes and varying timestamps. This is descriptive source evidence, not component-causal inference or an ordinary-user latency comparison. + +Each source has three fresh-app history replicates: three 200-turn/1,400-item history variants, three warm opens and 63 measured opens per app. Renderer GC is forced only at warm detached endpoints; intermediate renderer and all main observations use natural GC. Each source also has three fresh-app log replicates: one warm-up and 21 measured new-thread streams per app, each with 3 MiB logical output in 48 deltas. The final CLI event repeats aggregated output, so 3 MiB is not a wire-byte total. Every cycle positively checks thread/user identity, completion, opened output and native call counts. + +| Descriptive median of three replicates | Historical source | Current source | +| --- | ---: | ---: | +| History renderer warm detached before, bytes | 14,349,524 | 14,850,864 | +| History renderer warm detached after, bytes | 15,012,276 | 15,791,412 | +| History main natural heap after, bytes | 29,708,988 | 49,643,892 | +| History main natural private after, KiB | 105,105 | 119,105 | +| Log window bytes for 21 measured streams | 1,684,719,770 | 13,945,579 | +| Log maximum sampled output, bytes | 3,145,728 | 262,144 | +| Log renderer detached after forced GC, bytes | 73,373,420 | 9,448,940 | +| Log main natural heap after, bytes | 202,845,984 | 47,230,308 | +| Log main natural private after, KiB | 291,506 | 162,305 | + +Both log groups measured the same 63 MiB logical output per replicate. Historical sampled window traffic was 1,684,719,770 bytes in every replicate; current traffic was 13,945,579 bytes. Historical maximum main output and opened ASCII output were exactly 3,145,728 bytes: the expected negative control for the current 262,144-byte output limit. Current sampled output never exceeded that limit and its truncation label was verified. The historical-only shallow DOM text-length read verifies full output; that instrumentation difference is another reason not to compare ordinary latency. + +Log renderer endpoint deltas were +67,832,596 / +67,830,860 / +67,790,528 bytes historically and +3,437,796 / +3,722,464 / +3,201,784 bytes currently. These runs add 21 legitimate histories and sidebar entries; retained growth is not itself a leak. Uniform all-x payloads also do not represent diverse real logs. The lower current log-memory and IPC observations are consistent with the shipped preview path, but this experiment does not isolate a single implementation change. + +History renderer endpoint deltas were -2,324,652 / +662,396 / +668,016 bytes historically and +838,652 / +937,004 / +961,920 bytes currently. Current history endpoints include higher main natural-GC snapshots; no overall memory reduction is claimed. Sampled history-renderer heap ranged from 27,623,008 to 117,432,380 bytes historically and 26,923,300 to 156,610,040 bytes currently. These are sampled observations, not operation high-water values or ceilings. Main natural-GC timing prevents treating its endpoint difference as retained-memory growth. + +All 724 historical source blobs, 542 historical build files and 838 current build files stayed unchanged. All six historical cases passed on their first measurement execution, with the expected legacy output-limit violation. Twenty historical archive files were copied and byte-verified. Current-series archives and independent reviews remain separate; all sources and first setup failures are retained. Reported app close calls returned and reported PIDs were absent at audit, without an exhaustive process-tree claim. + +These measurements make no arbitrary memory-ceiling, leak-freedom, causal-repair, real-provider/native-platform or human-acceptance claim. The literal S5 assessment below also checks delivered structural caps and child regressions. The one- and five-second sampling points do not prove exhaustive asynchronous settlement. Per-replicate numeric observations, host/runtime versions, timestamps, original-receipt hashes, five-second natural-detach observations and independent-review hashes are in the [public aggregate](2026-11-12-cf8-retention.json). Original logs, full receipts, drivers, manifests, first setup failures and verified recovery archives remain in the private closeout evidence store. Public aggregation omits local executable/app/user-data paths and process IDs without changing numeric observations. + +The first historical build was started before source preparation had finished and failed on missing prepared files/dependency links. After that original preparation reached terminal success, the corrected build passed without source, assertion or timeout changes. An initial historical-driver construction used the wrong working directory and failed before app launch; it was corrected before the six-case measurement execution. Pre-execution review also strengthened the unexecuted legacy control to require exact full output. These setup and review corrections are distinct from measurement failures and product defects. + +The repeated drivers remain archived diagnostic evidence rather than canonical product tests. The current canonical measurement can be run from an owned checkout after the pinned install and build with `MODEX_MEASURE=/absolute/private/measurement.json pnpm --filter @modex/desktop exec playwright test e2e/transcript-measure.spec.ts --workers=1 --retries=0`. That command records its actual checkout/environment and exercises five fresh-process cold opens plus the streaming/output case; it does not reproduce the separate 63-open/21-stream retention series by itself. Portable repeated-series reproduction remains a follow-up for the separate reproducible capture/performance acceptance criterion; it is not an extra gate in S5. + +The existing 500 ms cold-open budget and 256 KiB output limit are unchanged. Historical [named-host measurements](2026-10-18-conversation-fidelity-acceptance.md#measurements-on-a-named-mac) and the [closeout qualification](2026-11-11-closeout-qualification.md#memory-and-output-observations) retain their actual source/runtime and first failures; the repeated-series observations above do not replace those latency receipts. The new comparison was also [recorded on #281](https://github.com/TypeSafeAI/modex/issues/281#issuecomment-6820368540). The later literal-clause audit supports only S5, as recorded below; full CF-8 acceptance remains open. + +## Literal S5 reconciliation + +An independent full-clause audit on 2026-11-12 found S5 supported at the named fixture scope and [recorded the evidence on #281](https://github.com/TypeSafeAI/modex/issues/281#issuecomment-6820945583). Only S5 was checked, taking the issue from eight to nine of 27 boxes. The measurement comparison alone is not the whole proof: this assessment also uses actual main `558b54f8` and its passed structural/lifecycle and child regression tests. + +| S5 clause | Evidence and boundary | +| --- | --- | +| Named long-transcript/log measurement and before/after comparison | Twelve historical/current fresh-app replicates, exact source/build guards, named runtime/host, positive workload identities, timestamps and load observations. Common current dependencies and sequential fixed-shape workloads remain disclosed. | +| Check memory/output bounds | The observed retention series is evaluated alongside 256 KiB retained transcript previews and paged reads, the 10 MiB per-tool raw-log cap, UTF-8 boundaries and writer cleanup. `shared/evidence.ts`, `main/engine/tool-logs.ts`, their unit suites and canonical `e2e/transcript-measure.spec.ts` are on actual main and passed its queue. Current series respected preview/IPC bounds; the full-output legacy control intentionally violated the current cap. | +| Check child-specific regressions | Actual main’s protected queue passed 2,478 units with 36 skips, 522 desktop cases with eight skips and seven site cases. Original logs positively include conversation identity/outcomes, Stop, typed questions/approvals/drafts, Markdown, evidence, journal/replay/client fallback and task-to-Notes regressions. This checks regressions without completing every child’s native/human acceptance. | +| Preserve failures and investigate causes without timeout relaxation | Original collection/source-label, build-sequencing and driver-path failures, expected legacy cap failures and their specific controls are retained. Assertions and timeouts were not weakened. Higher memory observations are not relabelled as a failed budget or a causal repair. | + +Transcript caches scale with retained histories. Incoming chunks and transient serialization allocations are not globally capped by the preview/raw-log constants. S5 requires checking the documented bounds and measurements; it does not require lower memory, a universal heap/RSS ceiling, leak freedom or a new native/human run. No further S5-only experiment is required by its literal text. Portable repeated-series reproduction, the reference/candidate visual inventory, native/provider/device/human acceptance and the remaining parent/child gates stay separate. The [aggregate](2026-11-12-cf8-retention.json) includes the criterion-audit hash and published receipt alongside unchanged numeric observations. diff --git a/docs/reviews/2026-11-12-local-model-refresh-preservation.md b/docs/reviews/2026-11-12-local-model-refresh-preservation.md index e0fdd2da..23d6a746 100644 --- a/docs/reviews/2026-11-12-local-model-refresh-preservation.md +++ b/docs/reviews/2026-11-12-local-model-refresh-preservation.md @@ -36,3 +36,9 @@ Direct inspection of the final ready and minimum-window Graphite model-search ca Hosted iPhone job `131046081135` in run `43177772414` succeeded on attempt one against the same `156e6cd` checkout/tree. The log contains 44 distinct UI start/pass pairs: 43 ordinary passes and one deliberate expected screenshot-failure selftest, with zero unexpected failures or skips. The 88,346,257-byte test-results ZIP matches GitHub's SHA-256 digest `4106d48ce6b5dafea9fb3c9f2346e563689554655b2144398f3060943b4b33d1` and passes CRC across 1,604 members. This is simulator evidence, not physical-device or human acceptance, and does not establish a TLS/shutdown cause or repair. Hosted macOS job `131046080786` in run `43177772414` succeeded on attempt one against checkout `156e6cd4a5ed93693731fb060bdb7b56fdd973b6`, tree `afc170eec41b712bec2130a39eac6e017c905739`, matching implementation source `16d4fa8`. Its original log reports 2,478 unit passes with 36 skips, 522 desktop E2E passes with eight skips and seven site passes, with zero failures, cancellations or observed retries. The E2E, runner-memory and coverage ZIPs match fresh GitHub artifact digests and source provenance and pass CRC. The E2E ZIP contains 176 members; both new case receipts report passed state and returned app close at the actual checkout. Hidden last-run files are absent from the hosted ZIP, so no hosted last-run JSON result is inferred. + +## Actual protected delivery + +PR [#503](https://github.com/TypeSafeAI/modex/pull/503) merged on 2026-11-12 at 09:58:08 UTC as signed commit `558b54f8fa0cb3679852c6ece2b07e23a3b10ffb`. Independent API and Git reads verified that its tree `3ba66a41d3e6eb3801de50649f9bfa718d362a5f` exactly matches reviewed source `44fb41eecf718cf6648b7c8ee95cab1e85e0e072` and is on main. The [protected queue run](https://github.com/TypeSafeAI/modex/actions/runs/43189470394) passed on attempt 1: 2,478 unit passes with 36 skips, 522 desktop passes with eight skips and seven site passes, zero failures/cancellations or observed retries. The queue iPhone and coverage-upload jobs were intentionally skipped for `merge_group`; exact-head source qualification remains separately recorded in the PR. + +The original queue log has SHA-256 `fdbd92c40c953f06ed965d41603d8776e62abfc6061e2d0514fd5e5c2dfd2956`. E2E artifact `13933066786` has SHA-256 `75208bce729bb37fc342fda716462b97a2646795fe5cdc1d4a4f7b85ce612d03`; memory artifact `13932777388` has SHA-256 `68cdfff8414f6d1366ff218ed56397539214cac188956f57d744ed32a9156728`. Both fresh GitHub digests, workflow identities and ZIP CRCs were verified. Both canonical receipts passed at the actual merged checkout with returned app close and zero sampled coding counters. Hosted hidden last-run files were absent, so no hidden-record result is inferred. An independent terminal/delivery audit found no findings. Full #381 S2/AC3, native compatibility/qualification controls, paired provenance, human acceptance, runtime admission and release remain separate requirements. diff --git a/docs/status.md b/docs/status.md index 78c74ee5..1a5f9fda 100644 --- a/docs/status.md +++ b/docs/status.md @@ -110,7 +110,7 @@ before relying on a tick. | [#273](https://github.com/TypeSafeAI/modex/issues/273) CF tracker | 9/15 | CF-3, CF-7 and CF-8 are open; their code is on `main`. Fresh full checks name candidate `b643617`, which predates #450, #453 and #463 | Val decides whether to keep `b643617` or pin a new candidate | | [#276](https://github.com/TypeSafeAI/modex/issues/276) CF-3 | 11/13 | Draft, attachment, YOLO and question-focus regressions plus approval recovery coverage (#487, #490, #495). Human D/P rows live on #281 | Val runs D1–D5 | | [#280](https://github.com/TypeSafeAI/modex/issues/280) CF-7 | 11/14 | Steps 1, 2 and 4: model-proposed relations, file hashes and selection/notes context are not built; step-level linking was retired | Val reviews the two saved Notes and records steps 1, 2 and 4 as deferred or retired | -| [#281](https://github.com/TypeSafeAI/modex/issues/281) CF-8 | 8/27 | Keyboard A1–A8, VoiceOver B1–B7, CF-3 D1–D5, P1–P4 and P6; memory measurement; ChatGPT multi-agent restriction check; staged release recommendation; list of remaining visual differences. C1–C5, V1–V3 and P5 are deferred | Val chooses the candidate, then runs A1–A8 and D1–D5 | +| [#281](https://github.com/TypeSafeAI/modex/issues/281) CF-8 | 9/27 | Keyboard A1–A8, VoiceOver B1–B7, CF-3 D1–D5, P1–P4 and P6; ChatGPT multi-agent restriction check; staged release recommendation; list of remaining visual differences. C1–C5, V1–V3 and P5 are deferred | Retain the [S5 clause receipt](reviews/2026-11-12-cf8-retention.md#literal-s5-reconciliation); complete separate reproducibility/visual gates and Val’s candidate-bound A1–A8 and D1–D5 checks | | [#319](https://github.com/TypeSafeAI/modex/issues/319) CR tracker | 2/10 | CR-3 to CR-8 acceptance; the R1–R4 comparison is retired (2026-11-04) but its boxes still name it | Val's [one session](reviews/2026-11-02-v0.5.0-val-session.md) | | [#322](https://github.com/TypeSafeAI/modex/issues/322) CR-3 | 10/14 | Original-reference comparison (AC1; retired 2026-11-04, box still names it), keyboard/VoiceOver (AC4); clearance beside CR-6's inline section is measured on `main` (#491), but these boxes also need the person-run checks | Val runs the [CR-8 VoiceOver checklist](reviews/2026-10-22-cr8-voiceover/checklist.md) | | [#323](https://github.com/TypeSafeAI/modex/issues/323) CR-4 | 10/14 | Original R1/R2 comparison, VoiceOver; clearance beside CR-6's inline section is measured on `main` (#491), but these boxes also need the person-run checks. Avatars are settled as logins only (decision 1, 2026-11-04) | Benchmark on real pull requests replaces the retired comparison; the comparison boxes stay unticked | @@ -122,13 +122,15 @@ before relying on a tick. | [#378](https://github.com/TypeSafeAI/modex/issues/378) LM-2 | 12/13 | #498 delivered the cold-load explanation and token/export regressions; #499 delivered manual-ID visibility and the complete metadata IPC fixture matrix. Packaged secure-storage acceptance remains open | Complete secure-storage acceptance and re-audit every criterion; existing ticks do not replace current evidence. [Receipt](reviews/2026-11-12-lm2-manual-visibility.md) | | [#379](https://github.com/TypeSafeAI/modex/issues/379) LM-3 | 12/13 | AC1's real native read/edit/check and AC4's regression/configuration receipts now bind to actual main. AC6 still has native redirect and managed-policy gaps; simultaneous isolation claims retain their tested fixture scope | Resolve AC6 without weakening destination binding; product admission stays separate. [Receipt](https://github.com/TypeSafeAI/modex/issues/379#issuecomment-6811858808) | | [#380](https://github.com/TypeSafeAI/modex/issues/380) LM-5 | 6/13 | S5 shared capacity/cancellable waiting/explicit status and AC3/AC5 are reconciled on protected main. Redirect privacy, action-safe compaction, genuine OOM and older-client presentation remain open | Complete remaining engineering/native gates; preserve #475's owner disposition. [Delivered S5 evidence](https://github.com/TypeSafeAI/modex/issues/380#issuecomment-6815410455) | -| [#381](https://github.com/TypeSafeAI/modex/issues/381) LM-4 | 2/13 | S4 separates inference/workspace/harness labels and AC6 preserves existing cloud/workspace behavior on main. Search is delivered by #502; native compatibility-test/qualification UI and phone inference provenance are not implemented | Complete missing implementation, delayed-refresh/state/keyboard coverage and human/device gates; product admission remains separate. [Criterion evidence](https://github.com/TypeSafeAI/modex/issues/381#issuecomment-6813640283) | +| [#381](https://github.com/TypeSafeAI/modex/issues/381) LM-4 | 2/13 | S4 separates inference/workspace/harness labels and AC6 preserves existing cloud/workspace behavior on main. Search is delivered by #502 and the two canonical preservation races by #503; full S2/AC3, native compatibility-test/qualification UI and phone inference provenance remain incomplete | Complete missing implementation, delayed-refresh/state/keyboard coverage and human/device gates; product admission remains separate. [Criterion evidence](https://github.com/TypeSafeAI/modex/issues/381#issuecomment-6813640283) | | [#382](https://github.com/TypeSafeAI/modex/issues/382) LM-6 | 0/11 | Bounded native bundled llama-server read/edit/check, Stop and two-worker restart receipts exist; the full runtime matrix is incomplete | Qualify remaining runtime/resource cases without generalizing one named combination | | [#383](https://github.com/TypeSafeAI/modex/issues/383) LM-7 | 0/15 | [Evidence map](reviews/2026-11-03-lm7-evidence-map.md) reconciles actual #498/#499 protected delivery and separates fixtures, native models, network-denied and human evidence; privacy/compaction/resource gaps remain | Complete producer criteria, integrated golden/adversarial workflow and runtime/human gates | | [#384](https://github.com/TypeSafeAI/modex/issues/384) LM-8 | 0/11 | Optional; an outside contributor's scope question is unanswered | A maintainer answers the [scope question](https://github.com/TypeSafeAI/modex/issues/384#issuecomment-6390714922) | ## In flight +- **Local-model refresh preservation** (#381). Protected #503 delivered signed `558b54f8`, exact reviewed tree `3ba66a4`. Its queue passed 2,478 units with 36 skips, 522 desktop E2E cases with eight skips and seven site cases. Canonical cases preserve two thread drafts, staged image bytes, search/edit focus and selected text through delayed metadata, and keep an old revision-bound target after endpoint replacement. Full S2/AC3 and qualification/provenance/admission/human requirements remain open. [Delivery and limits](reviews/2026-11-12-local-model-refresh-preservation.md#actual-protected-delivery). +- **CF-8 repeated source observations** (#281). Three fresh-app history and log replicates per source add 189 measured opens and 63 measured streams on each source. Public per-replicate numbers retain named host/runtime, timestamps and original receipt hashes. The comparison is descriptive, with shared current dependencies and sequential fixed-shape workloads; it does not qualify a historical shipped binary. The full S5 clause audit also checks delivered payload/log caps, cleanup and child regressions; only S5 is now supported, at nine of 27 issue boxes. Separate reproducibility, native and human acceptance remain open. [Measurements and limits](reviews/2026-11-12-cf8-retention.md). - **Local-service queue progress** (#380). A follow-up from merged #499 adds one waiting notice and a neutral continuation notice through the existing host-event path, plus an abort check at capacity handoff. Three canonical regressions failed before repair; 39 focused cases now pass, including actual runner/store persistence before dispatch and cancellation with no continuation or extra child. Full local build/typechecks, 2,478 units with 36 skips, 515 desktop E2E cases with 11 skips and seven site cases passed. Protected #500 delivered exact tree `b354cf7` at signed main `4f1d80c`; its queue passed 2,478 units with 36 skips, 518 desktop E2E cases with eight skips and seven site cases. S5 is reconciled at six of 13 issue boxes; product admission and remaining acceptance stay open. [Evidence and limits](reviews/2026-11-12-local-queue-status.md). - **Manual local-model IDs after metadata failure** (#378, #381). Protected #499 merged at `0f43631dec22db5ca9390e35613a139ae61bff07`, tree `17dfbd74e186da1fb7ecb8759170595ffe4685d5`. Its queue passed 2,474 units with 36 skips, 518 desktop E2E cases with eight skips and seven site cases. The source-head iPhone suite separately passed 44 executions, including its intentional screenshot-failure selftest; the queue iPhone job skipped. Manual visibility and IPC regressions are delivered; full issue/native/human acceptance remains open. [Evidence and limits](reviews/2026-11-12-lm2-manual-visibility.md). @@ -254,7 +256,7 @@ that fixed each. ## Next -- **Local-model refresh preservation** (#381), owned `model-refresh-preservation` candidate: promote two reviewed scripted-fixture characterizations into canonical coverage for cross-thread image/raw-draft preservation, edit-form focus, exact-request stale completion and unchanged old revision-bound targets. Both targeted cases pass with before/after selection checks. Build/typecheck, 2,478 units with 36 skips, 519 desktop E2E cases with 11 skips and seven site cases pass; final artifacts and terminal results are independently verified. Draft #503 has signed source `16d4fa8`; source-head hosted macOS, iPhone and native-platform jobs pass. Final-head checks and protected delivery are tracked in #503. [Plan](superpowers/plans/2026-11-12-local-model-refresh-preservation.md); [evidence and limits](reviews/2026-11-12-local-model-refresh-preservation.md). +- **CF-8 reproducibility and visual acceptance** (#281): make the archived repeated-series reproduction portable within the canonical measurement path and reconcile reference/candidate captures and remaining visual differences. The [reviewed comparison and S5 clause audit](reviews/2026-11-12-cf8-retention.md) check named measurements, delivered retained-data caps and child regressions without claiming overall memory reduction, a universal heap ceiling or leak freedom. S5 is supported; separate reproducibility and human/native/device acceptance remain open. Roughly in the order they earn their place; none is scheduled. From 743c2b71d69382194d4a5858ae254b4c004db10b Mon Sep 17 00:00:00 2001 From: =?UTF-8?q?Val=20Alexander=20=F0=9F=91=91?= Date: Thu, 12 Nov 2026 04:55:21 -0600 Subject: [PATCH 2/4] docs: reconcile delivered model-menu and refresh criteria --- ...26-11-12-local-model-refresh-preservation.md | 17 +++++++++++++++-- docs/status.md | 4 ++-- 2 files changed, 17 insertions(+), 4 deletions(-) diff --git a/docs/reviews/2026-11-12-local-model-refresh-preservation.md b/docs/reviews/2026-11-12-local-model-refresh-preservation.md index 23d6a746..9662b1e7 100644 --- a/docs/reviews/2026-11-12-local-model-refresh-preservation.md +++ b/docs/reviews/2026-11-12-local-model-refresh-preservation.md @@ -2,7 +2,7 @@ ## Candidate and scope -The `model-refresh-preservation` candidate starts from protected main `0e4e2126737874bb6604e9dece6334aa789f7c34`, including #502's model search and #501's Code Review decision record. This change adds two canonical Electron cases and an observer fixture; production application source is unchanged. It covers delayed metadata preservation under #381, without claiming the complete S2 or AC3 clauses, real-model qualification, ordinary product admission, paired-client provenance or human acceptance. +The `model-refresh-preservation` candidate starts from protected main `0e4e2126737874bb6604e9dece6334aa789f7c34`, including #502's model search and #501's Code Review decision record. This change adds two canonical Electron cases and an observer fixture; production application source is unchanged. This implementation receipt covers delayed metadata preservation under #381; the two cases alone did not claim the complete S2 or AC3 clauses. The later literal-clause reconciliation below combines them with the delivered search/manual-ID/status evidence. Real-model qualification, ordinary product admission, paired-client provenance and human acceptance stay separate. The fixture requires a disposable `modex-e2e-home-*` directory under the system temporary directory, scripted E2E mode and a mock default backend. It reuses the existing local-model workflow bootstrap. Discovery and coding observers call the original methods with their original receivers and return their results. The tests use bounded local metadata servers and request no coding turns. @@ -41,4 +41,17 @@ Hosted macOS job `131046080786` in run `43177772414` succeeded on attempt one ag PR [#503](https://github.com/TypeSafeAI/modex/pull/503) merged on 2026-11-12 at 09:58:08 UTC as signed commit `558b54f8fa0cb3679852c6ece2b07e23a3b10ffb`. Independent API and Git reads verified that its tree `3ba66a41d3e6eb3801de50649f9bfa718d362a5f` exactly matches reviewed source `44fb41eecf718cf6648b7c8ee95cab1e85e0e072` and is on main. The [protected queue run](https://github.com/TypeSafeAI/modex/actions/runs/43189470394) passed on attempt 1: 2,478 unit passes with 36 skips, 522 desktop passes with eight skips and seven site passes, zero failures/cancellations or observed retries. The queue iPhone and coverage-upload jobs were intentionally skipped for `merge_group`; exact-head source qualification remains separately recorded in the PR. -The original queue log has SHA-256 `fdbd92c40c953f06ed965d41603d8776e62abfc6061e2d0514fd5e5c2dfd2956`. E2E artifact `13933066786` has SHA-256 `75208bce729bb37fc342fda716462b97a2646795fe5cdc1d4a4f7b85ce612d03`; memory artifact `13932777388` has SHA-256 `68cdfff8414f6d1366ff218ed56397539214cac188956f57d744ed32a9156728`. Both fresh GitHub digests, workflow identities and ZIP CRCs were verified. Both canonical receipts passed at the actual merged checkout with returned app close and zero sampled coding counters. Hosted hidden last-run files were absent, so no hidden-record result is inferred. An independent terminal/delivery audit found no findings. Full #381 S2/AC3, native compatibility/qualification controls, paired provenance, human acceptance, runtime admission and release remain separate requirements. +The original queue log has SHA-256 `fdbd92c40c953f06ed965d41603d8776e62abfc6061e2d0514fd5e5c2dfd2956`. E2E artifact `13933066786` has SHA-256 `75208bce729bb37fc342fda716462b97a2646795fe5cdc1d4a4f7b85ce612d03`; memory artifact `13932777388` has SHA-256 `68cdfff8414f6d1366ff218ed56397539214cac188956f57d744ed32a9156728`. Both fresh GitHub digests, workflow identities and ZIP CRCs were verified. Both canonical receipts passed at the actual merged checkout with returned app close and zero sampled coding counters. Hosted hidden last-run files were absent, so no hidden-record result is inferred. An independent terminal/delivery audit found no findings. Actual delivery alone did not advance S2/AC3; the later clause audit below provides that reconciliation. Native compatibility/qualification controls, paired provenance, human acceptance, runtime admission and release remain separate requirements. + +## Literal S2 and AC3 reconciliation + +A later independent full-clause audit on 2026-11-12 checked actual main `558b54f8`, the production wiring, #499’s manual-ID/IPC coverage, #502’s production picker coverage and #503’s queue receipts. It found both S2 and AC3 supported at their complete UI/race scope; [only those two boxes were checked](https://github.com/TypeSafeAI/modex/issues/381#issuecomment-6821489229), taking #381 from two to four of 13. The independent clause-ledger SHA-256 is `2ad4118f4a68b125d109c234ca03d9fec18c6edc2fb39e4f4c2ec5967030fd55`. + +| Clause | Delivered evidence | +| --- | --- | +| Grouping/search/exact/custom IDs | Production-main/IPC `model-search.spec.ts` and `local-models.spec.ts`, plus manual-visibility units, exercise connection groups, original opaque model IDs, case-insensitive search, no-match/query editing and explicit IDs after metadata failure without admitting execution; unreadable-credential and rejected-IPC cases instead prevent descriptor reconstruction. | +| Delayed-refresh preservation | The two canonical preservation cases check exact thread drafts, PNG bytes, query/composer selections and focus, unsaved connection edits and custom IDs. The production unadmitted-picker case separately preserves query/focus/raw draft during delayed metadata. | +| Reachability versus qualification | Settings and the menu report metadata Answered/Unavailable separately from unverified coding capability. A metadata-ready group still receives the normal product-selection refusal; it does not become a coding-ready badge or persisted target. | +| Thread/profile/model/endpoint identity | The old exact request completes stale after endpoint revision replacement, while the existing thread keeps its old model/revision/raw draft and shows unavailable-server provenance. Generation/revision/request identity checks and explicit-only thread target mutation prevent metadata from silently rebinding it. Production discovery IPC separately exercises old-request/new-endpoint handling. | + +This accepts the literal UI/race requirements, not production admission or native inference. The synthetic admitted target setup, cancellation-before-stale-completion limit and preserved range-setup failure above remain unchanged. S1/S3’s user-started native compatibility test and recorded qualification states, the full options/state/keyboard/minimum-window matrix, paired provenance, actual producer integration and human/native/device acceptance remain open under their own full clauses. diff --git a/docs/status.md b/docs/status.md index 1a5f9fda..711a79bb 100644 --- a/docs/status.md +++ b/docs/status.md @@ -122,14 +122,14 @@ before relying on a tick. | [#378](https://github.com/TypeSafeAI/modex/issues/378) LM-2 | 12/13 | #498 delivered the cold-load explanation and token/export regressions; #499 delivered manual-ID visibility and the complete metadata IPC fixture matrix. Packaged secure-storage acceptance remains open | Complete secure-storage acceptance and re-audit every criterion; existing ticks do not replace current evidence. [Receipt](reviews/2026-11-12-lm2-manual-visibility.md) | | [#379](https://github.com/TypeSafeAI/modex/issues/379) LM-3 | 12/13 | AC1's real native read/edit/check and AC4's regression/configuration receipts now bind to actual main. AC6 still has native redirect and managed-policy gaps; simultaneous isolation claims retain their tested fixture scope | Resolve AC6 without weakening destination binding; product admission stays separate. [Receipt](https://github.com/TypeSafeAI/modex/issues/379#issuecomment-6811858808) | | [#380](https://github.com/TypeSafeAI/modex/issues/380) LM-5 | 6/13 | S5 shared capacity/cancellable waiting/explicit status and AC3/AC5 are reconciled on protected main. Redirect privacy, action-safe compaction, genuine OOM and older-client presentation remain open | Complete remaining engineering/native gates; preserve #475's owner disposition. [Delivered S5 evidence](https://github.com/TypeSafeAI/modex/issues/380#issuecomment-6815410455) | -| [#381](https://github.com/TypeSafeAI/modex/issues/381) LM-4 | 2/13 | S4 separates inference/workspace/harness labels and AC6 preserves existing cloud/workspace behavior on main. Search is delivered by #502 and the two canonical preservation races by #503; full S2/AC3, native compatibility-test/qualification UI and phone inference provenance remain incomplete | Complete missing implementation, delayed-refresh/state/keyboard coverage and human/device gates; product admission remains separate. [Criterion evidence](https://github.com/TypeSafeAI/modex/issues/381#issuecomment-6813640283) | +| [#381](https://github.com/TypeSafeAI/modex/issues/381) LM-4 | 4/13 | S2/S4/AC3/AC6 are supported on main. Native compatibility-test/qualification UI, full options/state/keyboard/minimum-window coverage and phone inference provenance remain incomplete | Complete the remaining implementation and producer/human/device gates; product admission stays separate. [S2/AC3 clause receipt](reviews/2026-11-12-local-model-refresh-preservation.md#literal-s2-and-ac3-reconciliation) | | [#382](https://github.com/TypeSafeAI/modex/issues/382) LM-6 | 0/11 | Bounded native bundled llama-server read/edit/check, Stop and two-worker restart receipts exist; the full runtime matrix is incomplete | Qualify remaining runtime/resource cases without generalizing one named combination | | [#383](https://github.com/TypeSafeAI/modex/issues/383) LM-7 | 0/15 | [Evidence map](reviews/2026-11-03-lm7-evidence-map.md) reconciles actual #498/#499 protected delivery and separates fixtures, native models, network-denied and human evidence; privacy/compaction/resource gaps remain | Complete producer criteria, integrated golden/adversarial workflow and runtime/human gates | | [#384](https://github.com/TypeSafeAI/modex/issues/384) LM-8 | 0/11 | Optional; an outside contributor's scope question is unanswered | A maintainer answers the [scope question](https://github.com/TypeSafeAI/modex/issues/384#issuecomment-6390714922) | ## In flight -- **Local-model refresh preservation** (#381). Protected #503 delivered signed `558b54f8`, exact reviewed tree `3ba66a4`. Its queue passed 2,478 units with 36 skips, 522 desktop E2E cases with eight skips and seven site cases. Canonical cases preserve two thread drafts, staged image bytes, search/edit focus and selected text through delayed metadata, and keep an old revision-bound target after endpoint replacement. Full S2/AC3 and qualification/provenance/admission/human requirements remain open. [Delivery and limits](reviews/2026-11-12-local-model-refresh-preservation.md#actual-protected-delivery). +- **Local-model refresh preservation** (#381). Protected #503 delivered signed `558b54f8`, exact reviewed tree `3ba66a4`. Its queue passed 2,478 units with 36 skips, 522 desktop E2E cases with eight skips and seven site cases. Canonical cases preserve two thread drafts, staged image bytes, search/edit focus and selected text through delayed metadata, and keep an old revision-bound target after endpoint replacement. The full S2/AC3 UI/race audit is now reconciled at four of 13 boxes; qualification/provenance/admission/human requirements remain open. [Delivery and limits](reviews/2026-11-12-local-model-refresh-preservation.md#actual-protected-delivery). - **CF-8 repeated source observations** (#281). Three fresh-app history and log replicates per source add 189 measured opens and 63 measured streams on each source. Public per-replicate numbers retain named host/runtime, timestamps and original receipt hashes. The comparison is descriptive, with shared current dependencies and sequential fixed-shape workloads; it does not qualify a historical shipped binary. The full S5 clause audit also checks delivered payload/log caps, cleanup and child regressions; only S5 is now supported, at nine of 27 issue boxes. Separate reproducibility, native and human acceptance remain open. [Measurements and limits](reviews/2026-11-12-cf8-retention.md). - **Local-service queue progress** (#380). A follow-up from merged #499 adds one waiting notice and a neutral continuation notice through the existing host-event path, plus an abort check at capacity handoff. Three canonical regressions failed before repair; 39 focused cases now pass, including actual runner/store persistence before dispatch and cancellation with no continuation or extra child. Full local build/typechecks, 2,478 units with 36 skips, 515 desktop E2E cases with 11 skips and seven site cases passed. Protected #500 delivered exact tree `b354cf7` at signed main `4f1d80c`; its queue passed 2,478 units with 36 skips, 518 desktop E2E cases with eight skips and seven site cases. S5 is reconciled at six of 13 issue boxes; product admission and remaining acceptance stay open. [Evidence and limits](reviews/2026-11-12-local-queue-status.md). - **Manual local-model IDs after metadata failure** (#378, #381). Protected #499 merged at `0f43631dec22db5ca9390e35613a139ae61bff07`, tree `17dfbd74e186da1fb7ecb8759170595ffe4685d5`. Its queue passed 2,474 units with 36 skips, 518 desktop E2E cases with eight skips and seven site cases. The source-head iPhone suite separately passed 44 executions, including its intentional screenshot-failure selftest; the queue iPhone job skipped. Manual visibility and IPC regressions are delivered; full issue/native/human acceptance remains open. [Evidence and limits](reviews/2026-11-12-lm2-manual-visibility.md). From 84867006b1b45dc506dd53bae55a9be85c67eee5 Mon Sep 17 00:00:00 2001 From: =?UTF-8?q?Val=20Alexander=20=F0=9F=91=91?= Date: Thu, 12 Nov 2026 05:49:09 -0600 Subject: [PATCH 3/4] docs: accept LM-2 secure storage and LM-4 option safety --- ...026-11-12-lm2-secure-storage-acceptance.md | 25 +++++++++++++++++++ ...-11-12-local-model-refresh-preservation.md | 14 +++++++++++ docs/status.md | 9 ++++--- 3 files changed, 44 insertions(+), 4 deletions(-) create mode 100644 docs/reviews/2026-11-12-lm2-secure-storage-acceptance.md diff --git a/docs/reviews/2026-11-12-lm2-secure-storage-acceptance.md b/docs/reviews/2026-11-12-lm2-secure-storage-acceptance.md new file mode 100644 index 00000000..d6df5df9 --- /dev/null +++ b/docs/reviews/2026-11-12-lm2-secure-storage-acceptance.md @@ -0,0 +1,25 @@ +# LM-2 secure-storage acceptance and complete literal audit + +All seven implementation steps and six acceptance criteria are supported on signed protected main `5a580d4f98f5b1c62c5801db4400d5b533634082`. The 20 credential/discovery/identity/IPC/export source and test files in the independent audit are byte-identical to their reviewed versions; the intervening #505 runner changes retain the local pinning/no-cloud-fallback guards. This is a narrow consumer check, not a review of the entire credit-conservation feature. The accepted #377 contract dependency is satisfied. + +- S1: main persists explicit Ollama, LM Studio and Custom Responses profiles using the accepted connection/model schema and production Settings IPC; scope remains configured same-machine loopback. +- S2: trusted-sender and exact-schema checks, connection ownership/revision, canonical IPv4/IPv6 and repeated DNS/socket checks enforce main-owned endpoints. URL credentials, injection, proxy routing and redirects are refused; no renderer auth helper or shell is executed. +- S3: local tokens use their own opaque namespace in the shared encrypted SecretStore, presence-only replies and fail-closed credential resolution. The packaged native test below proves the real macOS cipher boundary; local-token tests prove namespace, routing and redaction consumers. +- S4: metadata has request/aggregate deadlines, cancellation, response/model bounds and stale-result protection. Exact/manual IDs survive metadata failures when credentials remain readable; unavailable credentials or rejected IPC cannot reconstruct descriptors. Runtime-exposed installed/loaded state stays separate from unknown. +- S5: generative, embedding and other advertised model kinds remain distinct; advertised tools do not qualify coding. Known cloud-routed models are excluded and unknown downstream inference locality remains unverified. +- S6: Settings and refresh use only fixed metadata routes, with no inference, install/download, model load/unload, directory scan or port enumeration. The cold-load explanation is separately delivered by #498. +- S7: revisions, tombstones, exact credential binding, active-run edit/delete refusal and retained snapshots prevent replacement/deletion from redirecting an existing run while preserving readable history. +- AC1: normal, empty, malformed, unavailable, authenticated and delayed fixtures pass through production main/IPC; recorded requests carry only bounded metadata, never project content or inference. +- AC2: endpoint, DNS/redirect/proxy and forged-IPC tests enforce the same production boundary. +- AC3: safe renderer descriptors omit tokens; URLs reject credentials; fixed diagnostics and reflection guards avoid secret echoes; pre-event/log/record/snapshot redaction covers split and truncation boundaries. The actual saved and reopened Task Note export retains positive marks while omitting tokens. Authenticated discovery sends exactly its own connection token; tokenless requests have no Authorization header, and a seeded cloud credential is never forwarded. The native proof composes with these separate consumer tests; the generic key test alone does not establish every surface. +- AC4: identity includes connection and revision; delayed edit/switch/delete responses cannot become current. #503's canonical endpoint-replacement case preserves the old thread target. +- AC5: explicit custom/manual identities remain available for compatibility testing; embedding/nontext/cloud exclusions cannot become coding agents. The downstream compatibility-test interface and execution admission belong to their own issues. +- AC6: exact request fixtures prove discovery makes no forbidden calls, and revision/history/active-run guards prevent silent rebinding. + +The protected #503 queue passed 2,478 unit cases with 36 skips, 522 desktop cases with eight skips and seven site cases, with zero failures or observed retries ([required macOS job](https://github.com/TypeSafeAI/modex/actions/runs/43189470394/job/131084824667)). Relevant cases and unchanged source were rechecked against current main. Production references: [shared encrypted store](https://github.com/TypeSafeAI/modex/blob/5a580d4f98f5b1c62c5801db4400d5b533634082/apps/desktop/src/main/engine/secrets.ts), [local connection namespace](https://github.com/TypeSafeAI/modex/blob/5a580d4f98f5b1c62c5801db4400d5b533634082/apps/desktop/src/main/engine/local-models/connections.ts), [token/record/export regressions](https://github.com/TypeSafeAI/modex/blob/5a580d4f98f5b1c62c5801db4400d5b533634082/apps/desktop/test/local-token-records.test.ts) and [canonical credential assertions](https://github.com/TypeSafeAI/modex/blob/5a580d4f98f5b1c62c5801db4400d5b533634082/apps/desktop/e2e/app.spec.ts). + +The private QA artifact is ad-hoc macOS arm64 Modex 0.4.2, built from docs head `2825d145ef2ea2297e31553c08a04f57a1be3925` with application inputs equal to protected `558b54f8`. ASAR SHA-256: `9f8f2334167e87706eb6a177f745a21e8fb2c10f30943617a5a5114d21e4ec66`. The first canonical invocation timed out clicking Settings before key entry; Val observed no Keychain prompt, and the cause remains unestablished. A separate read-only Settings diagnostic passed. One explicitly recorded second canonical invocation then passed the unchanged original assertions: packaged Keychain backend, synthetic key outside state.json, ciphertext without plaintext including after base64 decoding, 0600 file, masked/cleared input, retrieval after reopening and Clear. Result: one pass, zero skips and zero framework retries. The app closed, private canonical fixture directories were cleaned by the test, artifact hashes stayed unchanged and strict deep codesign verification returned zero. No model turn ran and the installed user app was not replaced or closed. + +External retained receipts: full 13-clause independent audit SHA-256 `72a20278a2274d42b748976b7add5066c955501251d8b2aaa7cc4eada49f2976`; native terminal independent review `02b30a3b37017452a204cc88c7ab550024c12ef0bb12222075b0f2ff6d7476ef`; native terminal audit `fdd6f9518457e692df9fae11b624826e4d5f5a00f84f9bea7e6285433fe4b628`. First-failure logs and both subsequent invocations remain retained. This closes only LM-2's literal metadata/credential requirements. It does not establish cold-start reliability, local coding compatibility, offline privacy, Windows/Linux native-store qualification, human accessibility, product activation or release/notarization. Parent #376 and the downstream issues remain open. + +The full public clause receipt is [issue #378 comment 6822708976](https://github.com/TypeSafeAI/modex/issues/378#issuecomment-6822708976). The issue was independently read back as completed with 13 of 13 boxes on 2026-11-12. This documentation records the observed acceptance; PR #507 has its own source and protected-queue checks before delivery. diff --git a/docs/reviews/2026-11-12-local-model-refresh-preservation.md b/docs/reviews/2026-11-12-local-model-refresh-preservation.md index 9662b1e7..840477e4 100644 --- a/docs/reviews/2026-11-12-local-model-refresh-preservation.md +++ b/docs/reviews/2026-11-12-local-model-refresh-preservation.md @@ -55,3 +55,17 @@ A later independent full-clause audit on 2026-11-12 checked actual main `558b54f | Thread/profile/model/endpoint identity | The old exact request completes stale after endpoint revision replacement, while the existing thread keeps its old model/revision/raw draft and shows unavailable-server provenance. Generation/revision/request identity checks and explicit-only thread target mutation prevent metadata from silently rebinding it. Production discovery IPC separately exercises old-request/new-endpoint handling. | This accepts the literal UI/race requirements, not production admission or native inference. The synthetic admitted target setup, cancellation-before-stale-completion limit and preserved range-setup failure above remain unchanged. S1/S3’s user-started native compatibility test and recorded qualification states, the full options/state/keyboard/minimum-window matrix, paired provenance, actual producer integration and human/native/device acceptance remain open under their own full clauses. + +## S5 literal safety acceptance + +The complete S5 safety requirement is supported on protected main `5a580d4f98f5b1c62c5801db4400d5b533634082`. Unsupported features remain restricted; this clause does not require positively qualifying image, reasoning, Plan or structured-output support. + +- Composer blocks local image submission before dispatch and preserves staged bytes. The independent main/runtime guards also reject incompatible images, Plan/pages/knowledge before spawn; #503's canonical queue case verifies exact staged PNG bytes and raw drafts survive refresh and thread switching. +- Existing PlusMenu disables Plan, Auto and backend changes with reasons; the local model menu omits reasoning controls. The native adapter strips unsupported effort/fast flags and disables unqualified image/web/subagent options. Main validates the effective configuration and refuses forged or mismatched target/option mutations. +- Local targets are explicitly pinned: local means Auto is false, reroutes interrupt rather than silently rebind, live target changes are refused and no cloud fallback can replace the local target. The intervening #505 quota/runner changes retain these guards; no broad #505 review is claimed. + +Evidence: [production Composer](https://github.com/TypeSafeAI/modex/blob/5a580d4f98f5b1c62c5801db4400d5b533634082/apps/desktop/src/renderer/components/Composer.tsx), [native option/refusal tests](https://github.com/TypeSafeAI/modex/blob/5a580d4f98f5b1c62c5801db4400d5b533634082/apps/desktop/test/local-model-execution.test.ts), [runner pinning/forgery tests](https://github.com/TypeSafeAI/modex/blob/5a580d4f98f5b1c62c5801db4400d5b533634082/apps/desktop/test/runner-local-models.test.ts), and [protected preservation queue](https://github.com/TypeSafeAI/modex/actions/runs/43189470394/job/131084824667). Independent remaining-clause audit SHA-256: `416d35f9c32eeb5dfbae4dbe827fa192c64abccf4856d9a45d6b6ae2f454e62e`. + +Only S5 advances, from four to five of 13 boxes. Compatibility-test/result UI, ordinary producer integration/admission, full qualification/state/theme/keyboard/minimum-window coverage and paired host/model provenance remain open. Fixture/native-adapter guards do not activate production local inference or qualify any actual model. + +The [public S5 receipt](https://github.com/TypeSafeAI/modex/issues/381#issuecomment-6822710861) was read back with only S5 changed, taking the open issue to five of 13 boxes. diff --git a/docs/status.md b/docs/status.md index 711a79bb..0a96942a 100644 --- a/docs/status.md +++ b/docs/status.md @@ -63,6 +63,8 @@ and [landing verification](reviews/2026-10-11-landing-clarity.md). ## On main, unreleased +- **LM-2 metadata and credential acceptance** (#378). The complete 13-clause audit and packaged macOS synthetic-key receipt support closure on signed main `5a580d4f`. The first Settings-click timeout remains unexplained and retained; the second canonical invocation passed the unchanged encryption, masking, retrieval and clearing assertions. No local inference admission, release or universal platform claim follows. [Acceptance and native limits](reviews/2026-11-12-lm2-secure-storage-acceptance.md). + - **Model-picker search** (#381, #502). Protected main `bf497f7` has connection/model filtering with exact IDs, retained query/focus through metadata completion, and keyboard return to the filter. Its signed merge exactly matches the reviewed/tested tree; the protected queue passed 2,478 units with 36 skips, 520 desktop E2E cases with eight skips and seven site cases. The separate source iPhone suite passed 43 ordinary cases and one expected-failure selftest; installer/human/admission limits remain. The full #381 preservation and qualification matrix stays open. [Delivery and limits](reviews/2026-11-12-model-search.md). The published application source is `6a97d23aaea315b29008890f25dde2d7ea07f1d0` (v0.4.1). v0.4.2 is in preparation with Code Review hidden ([review](reviews/2026-11-03-v0.4.2-release-review.md)). Later source deliveries remain unreleased: #462 records local-model context provenance; #461 enforces local-only auxiliary dispatch; #465 refuses known cloud-routed and non-generative models; #463 (`32f1cb4`) defers Markdown far from the view so long threads open fast. Main still does not construct `LocalExecution`; native qualification and product admission remain separate gates. #459, #460 and #464 (`3d72c6e`, the CR-3 acceptance re-run) are evidence/documentation deliveries after the release source. @@ -119,20 +121,19 @@ before relying on a tick. | [#326](https://github.com/TypeSafeAI/modex/issues/326) CR-7 | 13/14 | Live canary on a disposable repository (AC6) | Val names a disposable repository | | [#327](https://github.com/TypeSafeAI/modex/issues/327) CR-8 | 5/17 | Reference comparison (retired 2026-11-04, boxes still name it), VoiceOver and the live gates. Budgets and C76 landed in #490; re-check item 18 and AC5 once Val confirms the budget numbers | Val's [one session](reviews/2026-11-02-v0.5.0-val-session.md) | | [#376](https://github.com/TypeSafeAI/modex/issues/376) LM tracker | 1/14 | LM-2 to LM-7 core acceptance; LM-8 is optional and does not block the core release; `LocalExecution` is not constructed, so the product refuses local turns | Val answers the five questions in `docs/reviews/2026-11-03-lm5-val-decision.md` on branch `lm-ready` | -| [#378](https://github.com/TypeSafeAI/modex/issues/378) LM-2 | 12/13 | #498 delivered the cold-load explanation and token/export regressions; #499 delivered manual-ID visibility and the complete metadata IPC fixture matrix. Packaged secure-storage acceptance remains open | Complete secure-storage acceptance and re-audit every criterion; existing ticks do not replace current evidence. [Receipt](reviews/2026-11-12-lm2-manual-visibility.md) | | [#379](https://github.com/TypeSafeAI/modex/issues/379) LM-3 | 12/13 | AC1's real native read/edit/check and AC4's regression/configuration receipts now bind to actual main. AC6 still has native redirect and managed-policy gaps; simultaneous isolation claims retain their tested fixture scope | Resolve AC6 without weakening destination binding; product admission stays separate. [Receipt](https://github.com/TypeSafeAI/modex/issues/379#issuecomment-6811858808) | | [#380](https://github.com/TypeSafeAI/modex/issues/380) LM-5 | 6/13 | S5 shared capacity/cancellable waiting/explicit status and AC3/AC5 are reconciled on protected main. Redirect privacy, action-safe compaction, genuine OOM and older-client presentation remain open | Complete remaining engineering/native gates; preserve #475's owner disposition. [Delivered S5 evidence](https://github.com/TypeSafeAI/modex/issues/380#issuecomment-6815410455) | -| [#381](https://github.com/TypeSafeAI/modex/issues/381) LM-4 | 4/13 | S2/S4/AC3/AC6 are supported on main. Native compatibility-test/qualification UI, full options/state/keyboard/minimum-window coverage and phone inference provenance remain incomplete | Complete the remaining implementation and producer/human/device gates; product admission stays separate. [S2/AC3 clause receipt](reviews/2026-11-12-local-model-refresh-preservation.md#literal-s2-and-ac3-reconciliation) | +| [#381](https://github.com/TypeSafeAI/modex/issues/381) LM-4 | 5/13 | S2/S4/S5/AC3/AC6 are supported on main. Native compatibility-test/qualification UI, full state/keyboard/minimum-window coverage and phone inference provenance remain incomplete | Complete the remaining implementation and producer/human/device gates; product admission stays separate. [UI/race and S5 safety receipts](reviews/2026-11-12-local-model-refresh-preservation.md#literal-s2-and-ac3-reconciliation) | | [#382](https://github.com/TypeSafeAI/modex/issues/382) LM-6 | 0/11 | Bounded native bundled llama-server read/edit/check, Stop and two-worker restart receipts exist; the full runtime matrix is incomplete | Qualify remaining runtime/resource cases without generalizing one named combination | | [#383](https://github.com/TypeSafeAI/modex/issues/383) LM-7 | 0/15 | [Evidence map](reviews/2026-11-03-lm7-evidence-map.md) reconciles actual #498/#499 protected delivery and separates fixtures, native models, network-denied and human evidence; privacy/compaction/resource gaps remain | Complete producer criteria, integrated golden/adversarial workflow and runtime/human gates | | [#384](https://github.com/TypeSafeAI/modex/issues/384) LM-8 | 0/11 | Optional; an outside contributor's scope question is unanswered | A maintainer answers the [scope question](https://github.com/TypeSafeAI/modex/issues/384#issuecomment-6390714922) | ## In flight -- **Local-model refresh preservation** (#381). Protected #503 delivered signed `558b54f8`, exact reviewed tree `3ba66a4`. Its queue passed 2,478 units with 36 skips, 522 desktop E2E cases with eight skips and seven site cases. Canonical cases preserve two thread drafts, staged image bytes, search/edit focus and selected text through delayed metadata, and keep an old revision-bound target after endpoint replacement. The full S2/AC3 UI/race audit is now reconciled at four of 13 boxes; qualification/provenance/admission/human requirements remain open. [Delivery and limits](reviews/2026-11-12-local-model-refresh-preservation.md#actual-protected-delivery). +- **Local-model refresh preservation** (#381). Protected #503 delivered signed `558b54f8`, exact reviewed tree `3ba66a4`. Its queue passed 2,478 units with 36 skips, 522 desktop E2E cases with eight skips and seven site cases. Canonical cases preserve two thread drafts, staged image bytes, search/edit focus and selected text through delayed metadata, and keep an old revision-bound target after endpoint replacement. The full S2/AC3 UI/race audit and conservative S5 option/pinning guards are reconciled at five of 13 boxes; qualification/provenance/admission/human requirements remain open. [Delivery and limits](reviews/2026-11-12-local-model-refresh-preservation.md#actual-protected-delivery). - **CF-8 repeated source observations** (#281). Three fresh-app history and log replicates per source add 189 measured opens and 63 measured streams on each source. Public per-replicate numbers retain named host/runtime, timestamps and original receipt hashes. The comparison is descriptive, with shared current dependencies and sequential fixed-shape workloads; it does not qualify a historical shipped binary. The full S5 clause audit also checks delivered payload/log caps, cleanup and child regressions; only S5 is now supported, at nine of 27 issue boxes. Separate reproducibility, native and human acceptance remain open. [Measurements and limits](reviews/2026-11-12-cf8-retention.md). - **Local-service queue progress** (#380). A follow-up from merged #499 adds one waiting notice and a neutral continuation notice through the existing host-event path, plus an abort check at capacity handoff. Three canonical regressions failed before repair; 39 focused cases now pass, including actual runner/store persistence before dispatch and cancellation with no continuation or extra child. Full local build/typechecks, 2,478 units with 36 skips, 515 desktop E2E cases with 11 skips and seven site cases passed. Protected #500 delivered exact tree `b354cf7` at signed main `4f1d80c`; its queue passed 2,478 units with 36 skips, 518 desktop E2E cases with eight skips and seven site cases. S5 is reconciled at six of 13 issue boxes; product admission and remaining acceptance stay open. [Evidence and limits](reviews/2026-11-12-local-queue-status.md). -- **Manual local-model IDs after metadata failure** (#378, #381). Protected #499 merged at `0f43631dec22db5ca9390e35613a139ae61bff07`, tree `17dfbd74e186da1fb7ecb8759170595ffe4685d5`. Its queue passed 2,474 units with 36 skips, 518 desktop E2E cases with eight skips and seven site cases. The source-head iPhone suite separately passed 44 executions, including its intentional screenshot-failure selftest; the queue iPhone job skipped. Manual visibility and IPC regressions are delivered; full issue/native/human acceptance remains open. [Evidence and limits](reviews/2026-11-12-lm2-manual-visibility.md). +- **Manual local-model IDs after metadata failure** (#378, #381). Protected #499 merged at `0f43631dec22db5ca9390e35613a139ae61bff07`, tree `17dfbd74e186da1fb7ecb8759170595ffe4685d5`. Its queue passed 2,474 units with 36 skips, 518 desktop E2E cases with eight skips and seven site cases. The source-head iPhone suite separately passed 44 executions, including its intentional screenshot-failure selftest; the queue iPhone job skipped. Manual visibility and IPC regressions are delivered; #378’s metadata/credential scope is accepted, while full #381 and remaining native/human acceptance stay open. [Evidence and limits](reviews/2026-11-12-lm2-manual-visibility.md). - **Noble security pocket source** (#498). Published parent `fc593e0` fetched official archive indexes but failed `security.ubuntu.com` DNS before Linux package installation. The helper now retains all four pockets, including `noble-security`, on the official archive host with the same Ubuntu keyring, strict update failure, nine packages and acquisition/host limits. Official Release metadata confirms that security pocket's presence; it is not signature or runner-connectivity proof. Eight focused cases and frozen build/typecheck/full units passed (2,455 passes, 36 explicit skips, zero failures). Real signed acquisition and installed Linux acceptance remain for fresh exact-head hosted qualification. [Evidence and limits](reviews/2026-11-11-linux-security-pocket.md). From 3f5d386d1098ec0f46e9be1e57645e39e83d3bb9 Mon Sep 17 00:00:00 2001 From: =?UTF-8?q?Val=20Alexander=20=F0=9F=91=91?= Date: Thu, 12 Nov 2026 05:56:33 -0600 Subject: [PATCH 4/4] docs: reconcile accepted LM-2 tracker dependency --- docs/status.md | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-) diff --git a/docs/status.md b/docs/status.md index 0a96942a..b8bc2ce6 100644 --- a/docs/status.md +++ b/docs/status.md @@ -120,7 +120,7 @@ before relying on a tick. | [#325](https://github.com/TypeSafeAI/modex/issues/325) CR-6 | 13/14 | Authorized live native smoke (AC6) | Run the smoke once on a small open TypeSafeAI/modex PR (authorized 2026-11-04) | | [#326](https://github.com/TypeSafeAI/modex/issues/326) CR-7 | 13/14 | Live canary on a disposable repository (AC6) | Val names a disposable repository | | [#327](https://github.com/TypeSafeAI/modex/issues/327) CR-8 | 5/17 | Reference comparison (retired 2026-11-04, boxes still name it), VoiceOver and the live gates. Budgets and C76 landed in #490; re-check item 18 and AC5 once Val confirms the budget numbers | Val's [one session](reviews/2026-11-02-v0.5.0-val-session.md) | -| [#376](https://github.com/TypeSafeAI/modex/issues/376) LM tracker | 1/14 | LM-2 to LM-7 core acceptance; LM-8 is optional and does not block the core release; `LocalExecution` is not constructed, so the product refuses local turns | Val answers the five questions in `docs/reviews/2026-11-03-lm5-val-decision.md` on branch `lm-ready` | +| [#376](https://github.com/TypeSafeAI/modex/issues/376) LM tracker | 2/14 | LM-1/LM-2 accepted; LM-3 to LM-7 core acceptance; LM-8 is optional and does not block the core release; `LocalExecution` is not constructed, so the product refuses local turns | Val answers the five questions in `docs/reviews/2026-11-03-lm5-val-decision.md` on branch `lm-ready` | | [#379](https://github.com/TypeSafeAI/modex/issues/379) LM-3 | 12/13 | AC1's real native read/edit/check and AC4's regression/configuration receipts now bind to actual main. AC6 still has native redirect and managed-policy gaps; simultaneous isolation claims retain their tested fixture scope | Resolve AC6 without weakening destination binding; product admission stays separate. [Receipt](https://github.com/TypeSafeAI/modex/issues/379#issuecomment-6811858808) | | [#380](https://github.com/TypeSafeAI/modex/issues/380) LM-5 | 6/13 | S5 shared capacity/cancellable waiting/explicit status and AC3/AC5 are reconciled on protected main. Redirect privacy, action-safe compaction, genuine OOM and older-client presentation remain open | Complete remaining engineering/native gates; preserve #475's owner disposition. [Delivered S5 evidence](https://github.com/TypeSafeAI/modex/issues/380#issuecomment-6815410455) | | [#381](https://github.com/TypeSafeAI/modex/issues/381) LM-4 | 5/13 | S2/S4/S5/AC3/AC6 are supported on main. Native compatibility-test/qualification UI, full state/keyboard/minimum-window coverage and phone inference provenance remain incomplete | Complete the remaining implementation and producer/human/device gates; product admission stays separate. [UI/race and S5 safety receipts](reviews/2026-11-12-local-model-refresh-preservation.md#literal-s2-and-ac3-reconciliation) |