Harden deregistration attribution evidence

[Eli Pinkerton (wallstop)]

Summary

• add a handler/bus/bus/handler deregistration palindrome with symmetric 3% interpretability gates
• preserve round-trip raw values and explicit gate decisions in one structured diagnostic record
• capture the exact IL2CPP deregistration call chain, including inline/shared/specialized bodies and hypothesis-negative evidence
• document why independent minimized windows are diagnostic prerequisites, not acceptance evidence

Why

Session 217 rejected the typed-cache deregistration specialization after byte-identical repeats reduced its apparent handler-layer improvement to 1.99%. Its broad generated-C++ capture also did not prove the exact typed call shape.

This PR deliberately leaves production deregistration unchanged. It hardens the evidence needed to decide whether a future exact-MessageBus specialization has a real interface-dispatch mechanism before another licensed A/B bracket.

Evidence contract

The new palindrome runs four independent existing benchmark windows in handler A, bus A, bus B, handler B order. It rejects non-finite or non-positive raw values/excesses, is arm-swap invariant, treats exact 3% boundaries as inclusive, and records diagnosticOnly=true acceptanceEvidence=false candidateCompared=false.

The Performance Numbers workflow structurally extracts complete generated method bodies, ignores forward declarations, follows concrete wrappers to shared bodies, accepts specialized bodies, rejects prefix collisions, and records interface/direct/unrecognized dispatch evidence before any hypothesis verdict. An embedded StrictMode self-test covers the extractor's zero/one/many and body-boundary behavior.

Validation

• Unity focused contract fixture: 40 passed, 0 failed, 0 skipped
• two live palindrome runs were interpretable
• npm test: 421 passed
• npm run validate:all
• Prettier, markdownlint, cspell, CSharpier, actionlint, and git diff --check
• final Unity editor state: main stage, clean scenes, no prefab stage, idle

Advances #414.

Exact-head verdict

• Performance Numbers run 31900934703 passed at 8678a2ca.
• Artifact perf-6000.5.2f1-standalone-internal (9251194997) records one typed interface-dispatch evidence line, zero direct-bus lines, and typedBusDispatchRecognized=true.
• The exact typed wrapper appears byte-identically at three generated locations and delegates to one shared body containing GenericInterfaceActionInvoker1 on the stored message bus.
• The final palindrome failed handler drift (10.38%), bus drift (5.37%), and handler-excess spread (14.66%) against the inclusive 3% gates, so interpretable=false.
• Static, devcontainer, performance, and all four Unity-version all-mode legs passed.
• The final Cursor review finding was fixed with independent case-distinct and Unicode-distinct body fixtures; its thread is resolved, and the adversarial re-audit found zero material or minor issues.

The interface-dispatch mechanism is proven, but repeated stable attribution is not. This PR therefore retains the diagnostic/evidence prerequisites and leaves the production exact-MessageBus specialization deferred.

---

Note

Low Risk
Changes are confined to perf benchmarks, contract tests, CI evidence scripts, and documentation; production deregistration/runtime code is untouched.

Overview
Hardens deregistration attribution evidence without changing production messaging behavior. The PR adds diagnostic-only benchmarking and CI codegen capture so future specialization decisions rest on exact IL2CPP call chains and noise-gated timing, not broad string matches.

A new deregistration palindrome runs direct handler → direct bus → direct bus → direct handler after the published attribution rows. AnalyzePalindrome and DeregistrationAttributionPalindromeDiagnostic apply symmetric 3% gates on same-path drift and handler-minus-bus excess spread, emit DXM_DEREGISTRATION_ATTRIBUTION_PALINDROME, and always record diagnosticOnly=true, acceptanceEvidence=false, and candidateCompared=false so interpretable=true is only a prerequisite, not proof of improvement. Contract tests cover boundary classifications, arm-swap invariance, and log shape; benchmark order shifts the new diagnostic to 3 and the dispatch twin to 4.

The Performance Numbers standalone step replaces bounded Select-String markers with Get-GeneratedMethodDefinition (brace-balanced full bodies, ordinal dedupe, embedded self-tests), then walks DeregistrationAttributionState.Execute → typed TypedHandlerDeregistrationState.Deregister (including shared/_gshared bodies) → MessageBus_Deregister, recording interface vs direct dispatch counts in typed-deregistration-codegen.txt. The perf methodology doc and unity-perf.test.js workflow assertion are updated to match.

^{Reviewed by Cursor Bugbot for commit 8678a2c. Bugbot is set up for automated code reviews on this repo. Configure here.}

[Copilot]

Copilot was unable to review this pull request because the user who requested the review has reached their quota limit.

[github-actions]

Performance Numbers

Measured commit: 8678a2c
Trusted reporting code (PR base): 3a5fa60
Workflow run: 31900934703 attempt 1

Benchmark or harness paths changed, so historical deltas are non-comparable:

• Tests/Runtime/Benchmarks/DispatchThroughputBenchmarks.cs
• Tests/Runtime/Benchmarks/RegistrationLifecycleBenchmarkContractTests.cs
• Tests/Runtime/Benchmarks/RegistrationLifecycleBenchmarks.cs

Historical Standalone delta

Delta direction: + is better; - is worse.

Historical delta omitted because benchmark or harness code changed in this pull request.

Current TargetMap evidence

These raw rows are current-run diagnostics. Use fresh bracketed controls before accepting a micro-optimization.

[Standalone] DXM_TARGET_MAP_BENCHMARK scenario=TargetMap_16_Churn keyCount=16 operation=Churn totalOperations=1940000 operationsPerSecond=387978.405 wallClockMs=5000.278 gcAllocations=-1 gcAllocatedBytes=-1 targetMapEntries=16 targetMapCapacity=32 observedInvocations=1960000
[Standalone] DXM_TARGET_MAP_BENCHMARK scenario=TargetMap_16_Hit keyCount=16 operation=Hit totalOperations=54000000 operationsPerSecond=10799537.348 wallClockMs=5000.214 gcAllocations=-1 gcAllocatedBytes=-1 targetMapEntries=16 targetMapCapacity=32 observedInvocations=54020000
[Standalone] DXM_TARGET_MAP_BENCHMARK scenario=TargetMap_16_Miss keyCount=16 operation=Miss totalOperations=61640000 operationsPerSecond=12327326.928 wallClockMs=5000.273 gcAllocations=-1 gcAllocatedBytes=-1 targetMapEntries=16 targetMapCapacity=32 observedInvocations=0
[Standalone] DXM_TARGET_MAP_BENCHMARK scenario=TargetMap_1_Churn keyCount=1 operation=Churn totalOperations=2140000 operationsPerSecond=427881.083 wallClockMs=5001.390 gcAllocations=-1 gcAllocatedBytes=-1 targetMapEntries=1 targetMapCapacity=4 observedInvocations=2160000
[Standalone] DXM_TARGET_MAP_BENCHMARK scenario=TargetMap_1_Hit keyCount=1 operation=Hit totalOperations=54170000 operationsPerSecond=10833621.040 wallClockMs=5000.175 gcAllocations=-1 gcAllocatedBytes=-1 targetMapEntries=1 targetMapCapacity=4 observedInvocations=54190000
[Standalone] DXM_TARGET_MAP_BENCHMARK scenario=TargetMap_1_Miss keyCount=1 operation=Miss totalOperations=59970000 operationsPerSecond=11992947.499 wallClockMs=5000.439 gcAllocations=-1 gcAllocatedBytes=-1 targetMapEntries=1 targetMapCapacity=4 observedInvocations=0
[Standalone] DXM_TARGET_MAP_BENCHMARK scenario=TargetMap_256_Churn keyCount=256 operation=Churn totalOperations=1830000 operationsPerSecond=364401.952 wallClockMs=5021.927 gcAllocations=-1 gcAllocatedBytes=-1 targetMapEntries=256 targetMapCapacity=512 observedInvocations=1850000
[Standalone] DXM_TARGET_MAP_BENCHMARK scenario=TargetMap_256_Hit keyCount=256 operation=Hit totalOperations=42840000 operationsPerSecond=8563891.559 wallClockMs=5002.399 gcAllocations=-1 gcAllocatedBytes=-1 targetMapEntries=256 targetMapCapacity=512 observedInvocations=42860000
[Standalone] DXM_TARGET_MAP_BENCHMARK scenario=TargetMap_256_Miss keyCount=256 operation=Miss totalOperations=56620000 operationsPerSecond=11323204.205 wallClockMs=5000.351 gcAllocations=-1 gcAllocatedBytes=-1 targetMapEntries=256 targetMapCapacity=512 observedInvocations=0
[Standalone] DXM_TARGET_MAP_BENCHMARK scenario=TargetMap_4096_Churn keyCount=4096 operation=Churn totalOperations=1390000 operationsPerSecond=276033.356 wallClockMs=5035.623 gcAllocations=-1 gcAllocatedBytes=-1 targetMapEntries=4096 targetMapCapacity=8192 observedInvocations=1410000
[Standalone] DXM_TARGET_MAP_BENCHMARK scenario=TargetMap_4096_Hit keyCount=4096 operation=Hit totalOperations=39370000 operationsPerSecond=7871769.298 wallClockMs=5001.417 gcAllocations=-1 gcAllocatedBytes=-1 targetMapEntries=4096 targetMapCapacity=8192 observedInvocations=39390000
[Standalone] DXM_TARGET_MAP_BENCHMARK scenario=TargetMap_4096_Miss keyCount=4096 operation=Miss totalOperations=59020000 operationsPerSecond=11802535.777 wallClockMs=5000.620 gcAllocations=-1 gcAllocatedBytes=-1 targetMapEntries=4096 targetMapCapacity=8192 observedInvocations=0
[Standalone] DXM_TARGET_MAP_BENCHMARK scenario=TargetMap_4_Churn keyCount=4 operation=Churn totalOperations=2150000 operationsPerSecond=428300.231 wallClockMs=5019.843 gcAllocations=-1 gcAllocatedBytes=-1 targetMapEntries=4 targetMapCapacity=8 observedInvocations=2170000
[Standalone] DXM_TARGET_MAP_BENCHMARK scenario=TargetMap_4_Hit keyCount=4 operation=Hit totalOperations=50440000 operationsPerSecond=10087909.814 wallClockMs=5000.045 gcAllocations=-1 gcAllocatedBytes=-1 targetMapEntries=4 targetMapCapacity=8 observedInvocations=50460000
[Standalone] DXM_TARGET_MAP_BENCHMARK scenario=TargetMap_4_Miss keyCount=4 operation=Miss totalOperations=57510000 operationsPerSecond=11500129.389 wallClockMs=5000.813 gcAllocations=-1 gcAllocatedBytes=-1 targetMapEntries=4 targetMapCapacity=8 observedInvocations=0
[Standalone] DXM_TARGET_MAP_CONSTRUCTION keyCount=1 wallClockMs=0.000100 operationsPerSecond=10000000.000 gcAllocations=-1 gcAllocatedBytes=-1 targetMapCapacity=4
[Standalone] DXM_TARGET_MAP_CONSTRUCTION keyCount=16 wallClockMs=0.000603 operationsPerSecond=1658031.088 gcAllocations=-1 gcAllocatedBytes=-1 targetMapCapacity=32
[Standalone] DXM_TARGET_MAP_CONSTRUCTION keyCount=256 wallClockMs=0.007750 operationsPerSecond=129032.258 gcAllocations=-1 gcAllocatedBytes=-1 targetMapCapacity=512
[Standalone] DXM_TARGET_MAP_CONSTRUCTION keyCount=4 wallClockMs=0.000214 operationsPerSecond=4664179.104 gcAllocations=-1 gcAllocatedBytes=-1 targetMapCapacity=8
[Standalone] DXM_TARGET_MAP_CONSTRUCTION keyCount=4096 wallClockMs=0.125400 operationsPerSecond=7974.482 gcAllocations=-1 gcAllocatedBytes=-1 targetMapCapacity=8192

[Copilot]

Copilot was unable to review this pull request because the user who requested the review has reached their quota limit.

[Copilot]

Copilot was unable to review this pull request because the user who requested the review has reached their quota limit.

[Copilot]

Copilot was unable to review this pull request because the user who requested the review has reached their quota limit.

[cursor]

Cursor Bugbot has reviewed your changes and found 1 potential issue.

^{❌ Bugbot Autofix is OFF. To automatically fix reported issues with cloud agents, enable autofix in the Cursor dashboard.}

^{Reviewed by Cursor Bugbot for commit f40358b. Configure here.}

[Copilot]

Copilot was unable to review this pull request because the user who requested the review has reached their quota limit.