Repository navigation
feat(comparison): preserve exact native sources and verified saved evidence - #91
Conversation
…015' into implementation/configurable-native-m3-integration-20261015
…015' into implementation/configurable-native-m3-integration-20261015
…015' into implementation/configurable-native-m3-integration-20261015
…015' into implementation/configurable-native-m3-integration-20261015
|
Important
This repository does not receive automatic reviews because it has fewer than 10 stars. ⚙️ Run configuration
Thanks for using CodeRabbit! It's free for OSS, and your support helps us grow. If you like it, consider giving us a shout-out. Comment |
…015' into implementation/configurable-native-m3-integration-20261015 # Conflicts: # docs/education/NATIVE_ADMISSION_INTEGRATION_REVIEW.md # docs/education/aligned-strain-native-sources-integration-20261014.json # docs/plans/configurable-solver-components/execution-ledger.md
|
Acknowledged concrete allocation from #89 comments6200315546/6200446821. I missed it while watching #87; corrected now. Fetched exact integrated head Concrete direction: cached linked2D native sample slice/probe, shared axis/index/probe and exact physical XYZ/nativeassociation/time/interval/unit/common finitebounds; pressuregauge diagnostic exactly declared. Explicit load of SelectedNativePair reopens pinned exact sources once, delegates actual values/geometry/sampling/reduction/gauge to compared_values, reads actual times from those same Selections. Cheap current-selection identity check on show clears stale/unverified cache; no perpaint database loads. Bounded preflight/native arrays/canvas workload; no remapping/zero replacement. Will report exactlead-callable API shortly for root registration, then freeze/review/test/publish newfile-only draft for your serialintegration. Meanwhile #89 head06ccf43 full local desktop suite completed:636passed,0failed,5explicitmodel-environment ignored; all-targetdesktopClippy passed. Hosted checks are ongoing; Pages confirmedinventorygate. These results do not transfer to this new b38-based source or close requestedmodel/diagnostic acceptance. |
|
Allocated three-file field-view API is now concrete on
Only the three allocated new files will change. Independent review, actual public-API tests, and a native interaction harness are pending before publication. Please retain root registration, admission inventory, and all merges under lead ownership. |
|
Allocated linked-view component is ready for your serial integration: draft PR93 #93, branch Reviewed exact source SHA911b7095, tests6a0454a1, docs3e8f47cb. 67 focused desktop +20 core-only passed; independent reviewer reran9 contracts/Clippy/format and approved with no material findings. Actual frozen native harness using the public comparison library exercised cold Unverified→Verify→Load, linked slice/probe/time changes, pressure and mean-offset diagnostic, native X faces/X-Z normals, source-free cached painting, missing-source reload clearing, page Missing/stale-metric clearing, byte-exact restoration/recovery, Close exit0. Independent stdlib SQLite/native-bit oracle checked14 observed display states in normal/-O/PYTHONOPTIMIZE=1 and rejected a perturbed probe in all modes. Evidence Separately frozen PR89 head06cc now has hosted core-only and desktop feature jobs PASS on test mergece50aba; default feature job still running. Pages remains fail-closed on the already-reported four-path source inventory difference. No checks weakened or stale receipts reused. |
|
Final integration boundary: PR93's Pages build has now actually refused exact source76cc71af9f2750354924eda30313eb7dbacf8c46 with PR89 frozen06cc all feature jobs now PASS (core-only492/3ignored, default499/3ignored, desktop636/5ignored; zero failures), along with previously reported executable/native/conditional/cloud/VTK/lessons/scoped jobs. Logs confirm actual test mergece50aba. Aggregate waits for its runner; Pages remains the independent source-inventory refusal already reported. No runtime/model qualification inferred. All allocated implementation/review/native-harness work is delivered in PR93 and its observation manifest. Remaining source/root registration and fresh integrated native/admission/CI acceptance are yours under the explicit shared-owner allocation; all branches preserved and no merge attempted. Ready for your integration or specific reviewed follow-up on these owned files. |
Cold native comparisons need exact source identities and reproducible derived evidence. Add public native-field analysis, immutable recipes/results, and actual desktop/run-set navigation on existing SQLite/run-set owners. Align explicit publications by actual time/stage and exact geometry; retain unmatched/unavailable coverage, native metrics/common bounds and explicit pressure-gauge diagnostics. Cold results remain Unverified until exact pins/selections are reopened and the whole result recomputed.
Stacked on #88; selected only reviewed companion files. Actual app/tests use public recording::comparison. compared_values reuses the exact existing reduction/gauge and retains native cell/face dimensions, offsets and physical geometry. Explicit page materialization reopens both exact archives/identities and matches the verified summary; no numerical/worker/store duplication. Public raw samples retain the overflow guard. Independent review closed assertion stripping in the new pointwise oracle using explicit failures and perturbed-sample checks in normal/-O/PYTHONOPTIMIZE=1.
Validation:20 numerical+9 page+12 archive contracts pass in the integrated source;20 numerical also pass core-only; scoped desktop Clippy -D warnings/formatting/native binary build pass. Strict95-path review/admission plus clean18-command native/rational and fresh packet26 normal/-O (zero skips) pass on b38c3fe (tree efe7b6b8afcae7ed531bd7601c37c6526d626aa9). Exact receipts /tmp/rheon-m3-values-qualified and /tmp/rheon-m3-values-packet; no stale receipts or gate weakening.
Actual integrated root-window interaction under software X11/llvmpipe opened an exact saved set and comparison page; selected pressure reference1..2/candidate1, yielding1 matched interval and1 unmatched reference; showed actual endpoint/dt/interval, Pa/common bounds/gauge; exported without overwrite; reopened as Unverified and verified exact sources. Temporarily hiding an owned test source produced Missing and removed stale metrics; bytes restored exactly; Close exit0. Independent SQLite/native-bit oracle checked the actual exported512-sample pressure metrics in all3 Python modes. Observation SHA25636e06591f177ed7b72ef9afd0b00c0f8a76bdda999c7086362701740601af550; result0febc3e9004856bab33c7abc946b5a19f3c79364d098ac382e0d1f256bf2d41d; binarye1e5306288bccb32f1437c3caec2847712419deb130730a38596ba4e8968ffdc. Generated material remains outside Git; no hardware GPU/build authenticity claim.
Linked native slice/probe rendering remains allocated to companion new files. Typed diagnostic storage is a separate stacked draft; runtime teacher/stop capture and actual trained bundle remain open. Keep draft pending those applicable scope/CI gates; no full M3 or physical/model qualification claimed. All merges/branch retirement stay serial; preserve companion #86/#89 and active stack.