Specialist fleet: pilot sweep results + automatability decomposition

Pilot sweep (all 4 gates, small_plus + medium):
  verification_type  84.3%  (medium, no_gain_stop_scaling)
  worker_type        80.8%  (small_plus, scale_more)
  automatability     78.3%  (medium, decompose_or_relabel)
  prereq_op          72.2%  (small_plus, decompose_or_relabel)

Automatability: 2-stage decomposition resolves Directional Collapse.
  Stage 1 binary (low_entropy vs needs_routing): 100%
  Stage 2 4-way (specialist/slm/llm/human): 100%
  Combined 5-way pipeline: 100%
  Script: specialists/scripts/split_automatability_decompose.py

Worker_type: combined dataset (sprint plans + projects) improves
  implementer class from 42.9% → 75.0%, overall 80.8% → 92.2%.
  Combined train/eval TSVs added.

Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
This commit is contained in:
Bill
2026-04-19 13:44:38 -06:00
parent 3fc9deac5b
commit 15125dce25
14 changed files with 2513 additions and 0 deletions

View File

@@ -0,0 +1,46 @@
1 0 name=LibrarySymbolAdvisor
1 0 name=SelfHostTranspileAudit
0 0 name=TrainingDataGenerator
1 0 name=IntentTranslator
1 0 name=TypeSystemTranslator
0 0 name=ContextBundleMinimizer
1 0 name=SecurityPreservingTranslation
1 0 name=IncrementalOptimizer
0 0 name=BufferOpsLifecycle
0 0 name=MigrationAcceptanceContract
1 0 name=ApprovalEscalationPlanner
0 0 name=Sprint163IntegrationSummary
1 0 name=ArchitectReviewInterface
1 0 name=MemoryModelTranslator
1 0 name=AgentCodeGen
0 0 name=Sprint238IntegrationSummary
0 0 name=CppJSBindingEmitter
1 0 name=OperationClassifier
0 0 name=Sprint36IntegrationSummary
0 0 name=SelfHostHarness
1 0 name=SymbolSelectorAPI
1 0 name=ArchitectReviewSurface
0 0 name=Sprint111IntegrationSummary
1 0 name=LibrarySymbolAdvisor
1 0 name=TaskitemLibraryAnnotator
0 0 name=SyntaxHighlighterElisp
0 0 name=SelfHostCodebaseAudit
1 0 name=PerTaskLibrarySelector
0 0 name=SyntaxHighlighterGo
1 0 name=AgentRegistry
1 0 name=OperationClassifier
0 0 name=WorkflowTemplates
1 0 name=AgentMutationApproval
1 0 name=AgentMutationPreview
0 0 name=RustPythonBindingEmitter
1 0 name=TransformEngine
1 0 name=OperationClassifier
0 0 name=CompilerPhaseSpec
1 0 name=ArchitectReviewSurface
1 0 name=OperationSelectorAPI
1 0 name=AgentMutationPreview
1 0 name=AgentMutationApproval
1 0 name=AgentTaskStatusOverlay
1 0 name=ScopeMilestoneDecomposer
1 0 name=SecurityExceptionReviewBoard
0 0 name=CompositionBuilder
1 1 0 name=LibrarySymbolAdvisor
2 1 0 name=SelfHostTranspileAudit
3 0 0 name=TrainingDataGenerator
4 1 0 name=IntentTranslator
5 1 0 name=TypeSystemTranslator
6 0 0 name=ContextBundleMinimizer
7 1 0 name=SecurityPreservingTranslation
8 1 0 name=IncrementalOptimizer
9 0 0 name=BufferOpsLifecycle
10 0 0 name=MigrationAcceptanceContract
11 1 0 name=ApprovalEscalationPlanner
12 0 0 name=Sprint163IntegrationSummary
13 1 0 name=ArchitectReviewInterface
14 1 0 name=MemoryModelTranslator
15 1 0 name=AgentCodeGen
16 0 0 name=Sprint238IntegrationSummary
17 0 0 name=CppJSBindingEmitter
18 1 0 name=OperationClassifier
19 0 0 name=Sprint36IntegrationSummary
20 0 0 name=SelfHostHarness
21 1 0 name=SymbolSelectorAPI
22 1 0 name=ArchitectReviewSurface
23 0 0 name=Sprint111IntegrationSummary
24 1 0 name=LibrarySymbolAdvisor
25 1 0 name=TaskitemLibraryAnnotator
26 0 0 name=SyntaxHighlighterElisp
27 0 0 name=SelfHostCodebaseAudit
28 1 0 name=PerTaskLibrarySelector
29 0 0 name=SyntaxHighlighterGo
30 1 0 name=AgentRegistry
31 1 0 name=OperationClassifier
32 0 0 name=WorkflowTemplates
33 1 0 name=AgentMutationApproval
34 1 0 name=AgentMutationPreview
35 0 0 name=RustPythonBindingEmitter
36 1 0 name=TransformEngine
37 1 0 name=OperationClassifier
38 0 0 name=CompilerPhaseSpec
39 1 0 name=ArchitectReviewSurface
40 1 0 name=OperationSelectorAPI
41 1 0 name=AgentMutationPreview
42 1 0 name=AgentMutationApproval
43 1 0 name=AgentTaskStatusOverlay
44 1 0 name=ScopeMilestoneDecomposer
45 1 0 name=SecurityExceptionReviewBoard
46 0 0 name=CompositionBuilder