feat: inspect real configured MAC workflows in the native desktop

docs/plans/fluid-accuracy-composition/execution-ledger.md

@@ -17,3 +17,50 @@ Next implementation slice remains G1. Method-selection and experimental-validati
 - Opened draft PR120 from content commit 788d5e5163d7153b726b71a7168da9311bc60850. Final gate inspection found root source-inventory recursively rejects extra Lean files, so opt-in proofs moved beside the plan; root checker/inventory remain unchanged. Requalified final paths and recorded the correction rather than weakening the gate.
 
 - Final plan-local kernel run returned exit 0: both modules, 25-declaration audit and 3 negative probes passed; unchanged root exact-source/dependency gate passed. P0 proposal/proof artifact accepted for delivery in draft PR120; production stays Deferred, next G1 unstarted.
+
+
+## 2026-10-20 — G1 admitted
+
+Explicit user implementation direction resolves prior G1 deferral. Frozen main remains bc2f472, PR115 included; protected audit119/plan120/pressurewarmstart59 untouched. New branch implement/g1-native-workflow-20261020. Source discovery required the bounded write-set correction recorded in plan.md before edits. Additive workflow controls/terminal profiles preserve current archives. Native interaction and performance remain prospective; no physical/GPU claim.
+
+Independent architecture review caught incomplete cancellation/fallback observation and incomplete request-control identity before qualification. Bounded re-plan in plan.md admits existing provider dispatcher/workspace observation hooks and an internal source accessor; no numerical-method change. Rejection, fallback initiation and cancellation remain distinct observed facts.
+
+### G1 core observation checkpoint
+
+Seven focused core contracts pass: actual candidate/fallback/appliedzero, short actualdt and exact baseline trajectory, before/after/in-stage cancellation and immutable state, canonical actual control/native state identity, illegal prefix/crossfield/limit rejection. Independent architecture review passes after reproducing and correcting missing-node, nativeRHS, cancellation and partial-diagnostic defects. Eight additive persistence contracts and sixty existing persistence integration tests pass independently; separate persistence review has no remaining findings. Native UI and fixed capture+cachedprojection budget remain pending; no G1 acceptance asserted.
+
+### G1 source implementation checkpoint
+
+Additive RCP4/RCT2 custody and native canvas/text/details/evidence integration are implemented. Independent architecture and persistence reviews pass; native UI source review has no remaining findings after six focused real-egui interaction/cache tests pass. Existing worker/runtime/producer-exit and old persistence contracts pass. Source moved Active → Implemented, then `verify` → Verifying. Actual native display workflow and fixed release capture+projection budget remain pending and must pass before acceptance. The core observation checkpoint b3db38d is pushed; the branch remains retained for the repository lead, with no merge/cleanup action.
+
+### G1 measured budget and retained native evaluation
+
+Source checkpoint bc494c1 is pushed in draft PR122 against plan120. Broad CPU run passed 597 tests in 74 suites with zero failures; current workflow contracts passed 15/15, existing persistence integration 60/60, strict all-target library/desktop Clippy and formatting passed. The first frozen release capture-plus-validated-projection evaluation passed: 30 paired intervals after five warmups, 0.4037% median added time against the unchanged 5% threshold, bitwise identical accepted native state after every pair. Independent reviewer recomputed raw hashes/math and approved the limited CPU workload claim; no optimization or criterion change was needed.
+
+The first actual normal-App run is retained as failed at `/tmp/rheon-native-display/failure-bc494c1`: live configured fallback was reached, but End/Left selection observation failed. Product versus isolated Xorg keyboard-map cause remains under investigation; stronger key-release/idle-frame event coverage passes. No native acceptance is asserted. G1 stays Verifying and PR122 stays draft. The 00:41 executor disconnect was transient: subsequent read/write/check calls succeeded; no worker or test was duplicated. Details and exact evidence hashes are in reports/g1-qualification.md.
+
+### G1 native verification finding — bounded repair
+
+In an explicit evdev XKB display environment, the unchanged bc494c1 binary passed actual pointer/keyboard/focus, exact live fallback interval and retained evidence selection, actual candidate/accepted residuals/limits/units, real cancellation (run 6, attempt 22, accepted generation 21, no publication), provider-absent cold reopening with identical run 4/ordinal 2 controls and archive hash, and owned App/provider cleanup. Independent native reviewer confirmed these observations. The first failed keyboard evaluation remains retained without a confirmed causal claim.
+
+The same review found the evidence window can exceed the 1000-pixel display because a multiline provider-binding Debug dump has no scroll area, hiding lower retained fields and Close. G1 returns Verifying → Active to repair only this bounded UI within its frozen write set. Add bounded scrolling/compact binding display and an overflow/close regression, then independently review and return Implemented → Verifying with a fresh source-bound native procedure. No numerical, archive, or acceptance-criterion change is admitted; PR122 stays draft. C1 readiness remains dependent on final G1 qualification.
+
+The repair now uses a viewport-relative bounded evidence body, compact hex provider identities and a footer Close outside the scroll area. A real configured receipt with bounded long binding values at 1000×600 verifies RHS initially offscreen, visible after real wheel scrolling and pointer-close with the workflow still open; key-release/idle-frame End/Left coverage also passes. All nine active workflow UI cases pass, two opt-in helpers remain ignored. Independent native reviewer approved the exact repair source. G1 returns Active → Implemented → Verifying; a fresh source-bound normal-App run and release budget evaluation remain required before acceptance.
+
+### G1 actual cold pointer/pan finding
+
+Checkpoint 3a86991 was pushed and built as the ordinary source-declared native binary. Fresh native interaction confirmed the bounded evidence repair, numerical details, live pointer/keyboard, real early cancellation (attempt 2 after accepted generation 1, no publication) and clean App exit. Partial screenshots and condition-wait diagnostics are retained separately; none constitutes full acceptance. The 00:56:43 disconnect callback did not lose source or processes: access checks found clean 3a86991 and the original App/driver active; work resumed without duplicate workers/runs.
+
+Actual cold interaction again showed canvas pan reaching Publication while details stayed Geometry. Independent pixel analysis identified the pan vector as exactly the pointer relocation from the previous control to the clicked node. A focused regression reproduced it: prior pointer at archive trigger (87,990), then pointer-move plus node press in one frame changed pan by (190.1,−635.5) before child hit testing. This is a product background/node gesture conflict, not resolved by keyboard-map changes or merely waiting for screenshots. G1 returns Verifying → Active for a bounded repair: retain actual press origin and initial pan, admit only background-origin gestures, and derive movement from the held pointer. Preserve node click/held movement without panning and actual background drag. Same source write set and unchanged acceptance criteria; no numerical or persisted-profile edits. Fresh independent review and source-bound native/performance evaluation follow.
+
+The gesture repair is implemented and independently approved: background-origin ownership, held displacement from actual press, exclusion of node origins, and disposal on release/reset/open/close or source invalidation. All eleven active workflow UI cases pass, zero fail; the two opt-in helpers remain ignored. Regressions include the exact 1600×1000 archive-trigger relocation, coalesced click/release/End with idle frames, node-held movement without pan, actual background displacement and invalidation during a held gesture. G1 returns Active → Implemented → Verifying. Final normal-App and fixed budget evaluations use the next source checkpoint, with native procedure starting only after both builds finish.
+
+### G1 accepted; bounded C1 preparation complete
+
+Implementation checkpoint `f6b7c8cd4628f6939c41567cecdf4a5231f60386` is pushed. Final fixed release evaluation passed 30 pairs after five warmups: 1.880860% median capture plus validated cache adoption against unchanged 5%, identical accepted native state after every pair. Independent architecture reviewer verified raw/binary/source hashes, workload, variability and order; PASS.
+
+After all builds completed, the same-source ordinary native binary passed the full normal-App live/evidence/numerical-details/cancel/close/provider-absent-cold/close-reopen workflow. Independent native reviewer verified 73 original artifact hashes, 22 accepted captures, exact unchanged run 4/ordinal 2 archive and controls, actual cancelled run 6/attempt 12 after accepted 11 with publication None, source/binary binding and complete owned App/provider/display absence. Drivers and live/cold App/display supervisors exited zero. No individual provider PID was sampled; only observed cleanup facts are claimed. Transient callbacks caused no observed work/process loss or duplicate runs. Failed prior evaluations remain retained alongside final evidence under `/workspace/rheon-g1-evidence`; actual archives and recorded executables are preserved there too.
+
+G1 transitions Verifying → Accepted, limited to G1-C/G1-U and the recorded performance scope. Claim register and visibility-gap disposition are synchronized; numerical F1–F8/physical/GPU/learned claims are unchanged. Source evidence and failure dispositions are in reports/g1-qualification.md, reports/g1-review.md and reports/g1-manifest.json.
+
+Only after G1 qualification, prepared reports/c1-readiness.md: C1-A Planned/dependency-ready with frozen two-module/declaration-only write set, existing pressure-v1/native-owner adapters, distinct family states, separately gated projection/whole-MAC candidates and explicit unavailable transfers. Independent architecture/contracts preparation review PASS; requested canonical G1-Accepted/C1-A-Planned status synchronization completed. C1 source implementation/acceptance is not claimed. P1/Q1 native method/reference work remains independent; repository lead coordinates shared declarations/status and all stacked-PR integration. No merge, protected-branch modification or branch cleanup performed.

docs/plans/fluid-accuracy-composition/issues.md

@@ -12,7 +12,7 @@ The frozen audit remains authoritative for baseline observations; this table is
 | F6 coarse channel/interface diffusion | Medium | Q1 measured accuracy envelope retains old failure; Q2 transport selection. Open. |
 | F7 area-lumped obstacle pressure/load | Medium | Q2 common B/M/S and independent flux/traction/refinement; new weight alone insufficient. Open. |
 | F8 uncalibrated cohesion/contact work | Medium | Q3 capillary stress/drop/wave + impulse/work ledger; moving/coupled capability separate. Open. |
-| Workflow visibility gap | New architecture requirement | G1 descriptor + actual receipts and native inspection; HTML only a provisional design aid. Open production requirement. |
+| Workflow visibility gap | New architecture requirement | Closed for existing configured sealed MAC by [qualified G1](reports/g1-qualification.md): authoritative descriptor, real receipts, native live/cold interaction and independent review. Other families/editing remain outside that acceptance; HTML is still only a design aid. |
 | Hidden provider state/side effects | Contract risk | C1 explicit prepare/commit/abort or stateless/disposable candidate; old pressure seam cannot roll back external side effects. Open implementation gate. |
 | MARIN numeric observations | Specific missing input | V1 physical validation only blocked; no block to reference/graph/contract work. |