erik/acdream
1
0
Fork 0
Code Issues Pull requests Projects Releases 6 Packages Wiki Activity Actions

docs(overhaul): close #474 mutation proof and record retail review

Browse source
This commit is contained in:
Erik 2026-09-05 09:51:13 +02:00
parent b913d542e0
commit 7c1dde3f3a
2 changed files with 41 additions and 11 deletions
Download patch file Download diff file Expand all files Collapse all files

42
docs/research/2026-09-01-overhaul/474-round0-verification.md
View file

@ -1,6 +1,8 @@
# #474 round0 — independent lead verification
Status: candidate verified offline; retail/behavior review pass1 RUNNING.
Status: candidate verified offline; retail/behavior pass1 found no code defect
and one docs-only finding, now corrected by the lead. Controller mutation proof
is complete. Production/lifetime review pass2 RUNNING.
No production commit, graphical PASS or G4 acceptance is claimed.
## Frozen candidate and contract
@ -53,7 +55,7 @@ projection-subscription10. The unchanged direct TryCancel case is included in
the physics-publisher26; the implementer also recorded a separate1/1 run.
These are focused deterministic tests, NOT the canonical full solution lanes.
## Mutation evidence and its explicit limit
## Mutation evidence and restoration
The lead read each failed TRX and verified its SHA256, test name and first
assertion. All use the genuine composed T1, not a fake ACK/preset completion.
@ -83,12 +85,27 @@ Service tested-cycle pre/post hash matched
`B2FDD7EB6B5FD8AB5A645F2079680557A06192DEA3A5D252417C65611D4ACA5C`;
the later remarks-only correction yields the final hash below.
**OPEN PROOF GAP:** the controller's tested mutation had different pre/post
The original controller tested mutation had different pre/post
raw hashes. A later byte-identical remove/restore cycle was NOT test-run.
That does not meet the contract's single-cycle restoration proof. The lead
will repeat the tested mutation against the frozen final source after the
current read-only review, restore it byte-identically and rerun the green test.
No claim that this proof already exists.
That did not meet the contract's single-cycle restoration proof. After retail
review finished, the lead repeated the tested mutation against the final source:
raw controller hash BEFORE removal and AFTER restoration both
`8A6AA9202390E8DF9A2916196EA539D18E52D33972FE71D2F9D65808A18B01D3`.
The mutation exited1,0/1, at the intended Assert.False early-readiness assertion
(test line60), not at a setup/compile failure. Byte-identical restoration then
passed1/1, exit0. This closes the single-cycle proof gap.
Literal lead commands (same scratch; fresh zero-client guard before each):
```powershell
dotnet test tests/AcDream.App.Tests/AcDream.App.Tests.csproj -c Release --no-restore --filter 'FullyQualifiedName=AcDream.App.Tests.Streaming.LandblockPlacementPublicationCompositionTests.CollisionRefresh_WithMissingGraphicalBackend_PublishesBackendBeforeExactRestoreAck' --logger 'trx;LogFileName=474-lead-controller-mutation.trx' --results-directory 'artifacts/474-lead/controller-mutation' -v minimal
dotnet test tests/AcDream.App.Tests/AcDream.App.Tests.csproj -c Release --no-restore --filter 'FullyQualifiedName=AcDream.App.Tests.Streaming.LandblockPlacementPublicationCompositionTests.CollisionRefresh_WithMissingGraphicalBackend_PublishesBackendBeforeExactRestoreAck' --logger 'trx;LogFileName=474-lead-controller-restored.trx' --results-directory 'artifacts/474-lead/controller-restored' -v quiet
dotnet build AcDream.slnx -c Release --no-restore -v quiet
dotnet test tests/AcDream.App.Tests/AcDream.App.Tests.csproj -c Release --no-build --no-restore --filter 'FullyQualifiedName~GraphicalRemotePlacementServiceWindowTests|FullyQualifiedName~LandblockConcretePresentationPipelineTests|FullyQualifiedName~LandblockPhysicsPublisherTests|FullyQualifiedName~LandblockPlacementPublicationCompositionTests|FullyQualifiedName~LandblockPresentationPipelineTests|FullyQualifiedName~StreamingControllerReadinessTests|FullyQualifiedName~RuntimePlacementPresentationSinkTests|FullyQualifiedName~RuntimePlacementShadowCompositionTests' --logger 'trx;LogFileName=474-lead-app-final.trx' --results-directory 'artifacts/474-lead/app-final' -v minimal
```
The final solution build exited0,0 warnings/errors; final App lane exited0,
135/135 with no skips. The App.dll hash stayed unchanged. The only permanent
change after retail review was the service-window test's class-summary comment:
Near alone is no longer the complete predicate; it also requires presentation.
T4 covers reset/recenter × pre/post-backend starting edge, failed static
reconciliation retry, exact ACK completion, retirement and zero remaining
@ -108,21 +125,24 @@ to this scratch, not the campaign's later graphical build.
| src/AcDream.App/Streaming/StreamingController.cs | 8A6AA9202390E8DF9A2916196EA539D18E52D33972FE71D2F9D65808A18B01D3 |
| src/AcDream.App/Streaming/GraphicalRemotePlacementServiceWindow.cs | A1524051D58F5D69875273C689367FC956D5592D456DD4A1DE0E355E2BDCB656 |
| src/AcDream.App/Composition/SessionPlayerComposition.cs | 841C7E049A283D631CF0DA868E89E66DB7F4595A71591960FE466F8E074E6420 |
| tests/AcDream.App.Tests/Streaming/GraphicalRemotePlacementServiceWindowTests.cs | DF47F00AE3FD6322A120477F71F0E6F2F9B8DE1B9B436C3F05096B6ABF9B770E |
| tests/AcDream.App.Tests/Streaming/GraphicalRemotePlacementServiceWindowTests.cs | B54F7102F08F5F07000B777A4C025756AECC20DE54D9CE9FD714AD47B572212A |
| tests/AcDream.App.Tests/Streaming/LandblockPlacementPublicationCompositionTests.cs | AD40411A3F7E1564FB451BDD3FA639D689218887D1C25625E2D86EA0BB717793 |
| docs/architecture/acdream-architecture.md | 1B4F23E6CB06ACA00D63EAA56752CBC0FE330A12DF42B6A69A736C7B00C85A13 |
| docs/architecture/retail-divergence-register.md | CE1098C223E27676ED13039BBD9F6200656747514C95A1E4392A0A23A4A454DA |
| src/AcDream.App/bin/Release/net10.0/AcDream.App.dll | 6698A3A018E8C68F86835042F5F211B20326F33CCE99FB9121028C17EA551B90 |
| tests/AcDream.App.Tests/bin/Release/net10.0/AcDream.App.Tests.dll | 5E69DBE0699DCF23E77F1CA111DAB41FA8E479B072A4CFFE162C555C60473B1D |
| tests/AcDream.App.Tests/bin/Release/net10.0/AcDream.App.Tests.dll | 0024A86D56B4BEE2010A8A709D2757509C499135968167482B7A8239441FC53A |
| tests/AcDream.Runtime.Tests/bin/Release/net10.0/AcDream.Runtime.Tests.dll | E5B5840B2C55AE636E2190818170DC65AEE901E8C2C4051271788E9BE9BF890A |
| artifacts/474-lead/app/474-lead-app.trx | F2589C7A3EE3D200579BE3C5B45F286C519669E18259FD3317BCAD06FA958CF2 |
| artifacts/474-lead/runtime/474-lead-runtime.trx | 3BF20A03034A2320DD4B1482FA754D3B382934FE368B9BE0A5BD4C54B9D79CF8 |
| artifacts/474-lead/controller-mutation/474-lead-controller-mutation.trx | 530B0266C7D26A735C2B79214AB4FFD4E6FC036807E7BB7EDE8A5268D8F95D70 |
| artifacts/474-lead/controller-restored/474-lead-controller-restored.trx | C0B341734449CF36476448E3EC5BD66FBCFCBE924EF40ECF1EEF1BCE032D7886 |
| artifacts/474-lead/app-final/474-lead-app-final.trx | B7D7BCC67191100D1733EC0552164602AC98A503AD6BFE0D33D4CB8F083D201A |
## Remaining boundary
Named-retail/paired-PE proof is in the amended contract; the lead rehashed the
paired retail executable again during verification and got the same SHA256.
The change is AD-2's asynchronous adaptation, not literal retail machinery.
Independent reviews, the controller mutation restoration proof, integration,
canonical full lanes, the exact nine-hop graphical route and owner G4 remain
Production/lifetime review, integration, canonical full lanes,
the exact nine-hop graphical route and owner G4 remain
outstanding. Preserve C1c and geometry-witness work; never merge main before G4.

10
docs/research/2026-09-01-overhaul/s5-consumers-material-closeout-packet.md
View file

@ -4456,3 +4456,13 @@ open: the lead will repeat the tested cycle after that read-only review so the
candidate source remains frozen while reviewed. This is not a code failure or
a graphical result. Production/lifetime review and the actual graphical gate
remain pending; no production commit or G4 acceptance yet.
Retail/behavior review pass1 returned no production behavior defect and one
MINOR docs-only stale service-window test summary. The lead corrected that
comment directly. The lead also closed the controller proof gap with a fresh
single mutation/test/byte-identical restoration cycle: guard removal0/1 at
the intended early-readiness assertion, restored1/1, same raw controller hash
before/after. Final Release solution0W0E and affected App135/135 were rerun;
App.dll stayed unchanged. Exact commands/hashes are in the verification record.
Production/lifetime/gate-honesty review pass2 is RUNNING against the refrozen
eight-file return. No production code fix round or graphical PASS is claimed.