bytecode tango 3

[faktorial-ai]

User report of this issue

---

Rebaseline the Finite Bytecode Retirement Inventory

Context: #3377 asks to complete the entire new bytecode execution layer and retire the IR execution layer. The maintained source of truth is docs/plans/bytecode-burndown-checklist.md, with route/status context in docs/bytecode-progress.md and drift inventories in docs/unified-bytecode-expansion-contract.md. Keep each child bounded to one source-owned lane, and do not claim final retirement until manifest/source-absence gates prove it.

• Rebaseline the current open bytecode checklist, proof manifest, and expansion-contract drift inventories, then update docs only where they are stale against current source.
• Convert the remaining open checklist rows into owned child slices with exact source files, proof anchors, and dynamic-residue boundaries before runtime changes continue.

Close Remaining Admission Gaps

Context: These children should admit one semantic family at a time by updating the production eligibility gate, compiler lowering, VM/runtime semantics, route-hit tests, no-route neighbor tests, and checklist/proof manifest rows together.

• Own the A1/A2 dynamic activation residue boundary: captured dynamic chains, bounded direct-eval/arguments gaps, retained with-chain boundaries, and async/generator parity where safe.
• Own the A51 compiler-leaf leftovers by retiring or narrowing named UnsupportedPlanShape buckets only when selector, compiler, VM, and proof coverage are all present.
• Own the B24h class-expression computed/static-block environment bridge, especially runtime-source direct eval and non-production static-block plans.
• Own the B36 resumable declaration leftovers for dynamic/eval helpers and complex class declarations outside the currently admitted public subsets.

Retire Fallback Tiers in Bounded Steps

Context: Accepted production unified bytecode must remain all-or-nothing. Do not move semantics into runner-time fallback. Delete or hard-quarantine fallback entry points only after the corresponding non-dynamic admission surface is proven.

• Remove remaining tier-1 ExpressionProgram execution surfaces that are not terminal dynamic residue, replacing them with standalone unified bytecode or deleting dead bridges.
• Remove remaining ordinary/script ExecutionPlanRunner fallback entry points after all non-dynamic declines are gone or explicitly quarantined as dynamic residue.
• Retire async/generator/class fallback runner bridges after resumable admission parity is proven for their non-dynamic surfaces.
• Keep direct eval, awaited/live with, eval-injected bindings, and Function(...) produced bodies as the only documented terminal interpreter residue.

Final Gates and Documentation

Context: Final completion is not a route-width claim; it requires machine-checkable absence of the old execution layers for non-dynamic code.

• Add or tighten standing CI/source-absence gates so non-dynamic production declines, legacy expression evaluation bridges, and ExecutionPlanRunner entry points cannot silently return.
• Run the canonical focused proof packs for bytecode admission, manifest drift, AST/IR fallback absence, and allocation-stability checks required by the repo playbooks.
• Update docs/bytecode-progress.md, docs/plans/bytecode-burndown-checklist.md, docs/plans/bytecode-proof-manifest.json, and docs/unified-bytecode-expansion-contract.md to state the final retired-state evidence.

---

Investigation

Delivery strictness: full - This task edits machine-checked bytecode governance surfaces and may drive child planning for runtime route/fallback retirement; the work depends on source-presence/source-absence gates, cross-module VM/compiler/runner behavior, and proof-manifest consistency.

Blast Radius

• docs/plans/bytecode-burndown-checklist.md - maintained finite checklist, current open-row inventory, row counts, and source-owned lane boundaries.
• docs/plans/bytecode-proof-manifest.json - executable proof index for admitted, open, hard-quarantined, and retired bytecode rows.
• docs/bytecode-progress.md - human-readable current route/status map and final-proof evidence summary.
• docs/unified-bytecode-expansion-contract.md - drift inventory for decline codes, proof commands, and expansion-contract rows.
• tests/Asynkron.JsEngine.Tests/BytecodeProofManifestTests.cs - manifest/checklist consistency, source-presence/source-absence, and classified fallback gates.
• tests/Asynkron.JsEngine.Tests/ExpressionProgramCoverageMapTests.cs - expansion-contract and expression coverage drift gate.
• tests/Asynkron.JsEngine.Tests/UnifiedBytecodeProductionEligibilityTests.cs - route/no-route proof for production eligibility boundaries.
• tests/Asynkron.JsEngine.Tests/UnifiedBytecodeProductionInvocationTests.cs - production invocation, classified fallback, and source gate coverage.
• src/Asynkron.JsEngine/Execution/UnifiedBytecode/UnifiedBytecodeCompiler.cs - A51 compiler diagnostic owners and emitted opcode coverage.
• src/Asynkron.JsEngine/Execution/UnifiedBytecode/UnifiedBytecodeProductionEligibility.cs - A1/A2/A51/B24h/B36/E5 eligibility and decline classification.
• src/Asynkron.JsEngine/Execution/UnifiedBytecode/UnifiedBytecodeVirtualMachine.cs - VM ownership for admitted rows and fallback-free execution constraints.
• src/Asynkron.JsEngine/Ast/TypedAstEvaluator.UnifiedBytecodeResumableActivation.cs - A2 resumable direct-eval and dynamic activation residue boundary.
• src/Asynkron.JsEngine/Ast/TypedAstEvaluator.IrSyncGeneratorInvoker.cs - E5d sync-generator creation-time IR route-selection anchor.
• src/Asynkron.JsEngine/Ast/TypedAstEvaluator.AsyncFunctionInvoker.cs - E5d async-function declined-body runner residue anchor.
• src/Asynkron.JsEngine/Ast/Legacy/StatementNodeExtensions.cs - E5c classified script and terminal dynamic script fallback anchors.
• src/Asynkron.JsEngine/Ast/ClassDefinitionExtensions.cs - B24h/B36 static-block production route and classified static-block residue.

Dependencies

• None identified.

User Stories

Story 1: As a bytecode runtime maintainer, I want the finite retirement inventory rebaselined against current source and proof gates, so that future runtime work starts from real open blockers instead of stale checklist text.

Story 2: As a Faktorial build agent, I want the remaining open rows converted into owned child slices with exact source files, proof anchors, and dynamic-residue boundaries, so that implementation work can be scheduled without parking ordinary fallback work in A1/A2 or terminal residue.

Story 3: As a reviewer, I want every retirement claim backed by manifest and source-absence evidence, so that final IR/expression fallback retirement is machine-checkable rather than narrative-only.

Approach

• Start the build stage by capturing a pre-edit baseline: current HEAD 85de9bb6cd6ba1cece67906f71c93bf374a24942, checklist totals 137 / 161, A+B+C+D 127 / 146, Phase E 5 / 10, manifest item count 32, non-done manifest items A1 A2 A51b A51a2 A51c A51j A51h A51h1 A51h2 A51k A51f3 A51f5 B24h B36 E5b E5c E5d E5e, and the focused runner/E4 source scans before edits.
• Treat docs/plans/bytecode-burndown-checklist.md as the human source of truth, docs/plans/bytecode-proof-manifest.json plus BytecodeProofManifestTests as the executable source of truth, and docs/unified-bytecode-expansion-contract.md as the decline-template drift inventory.
• Reconcile the four named docs against current source. Update docs only where stale; if a manifest row moves, update BytecodeProofManifestTests in the same change so the row is executable.
• Convert every remaining open checklist lane into independently shippable child slices. Preserve these current groups: A1/A2 dynamic activation residue, A51 compiler leaves including A51g/A51i lanes that currently lack manifest closure rows, B24h class-expression/static-block environment bridge, B36 resumable declaration/class-definition residue, and E5b-E5e runner-retirement/exclusion rows.
• Do not reopen E4 unless current production source contradicts the manifest. Current evidence shows the E4 dynamic executor and old standalone/lowered expression bridges are absence-ratcheted and production-source scans return no matches; runner-internal expression execution belongs to E5.
• Keep runtime changes blocked behind child conversion unless the build finds a tiny proof-manifest/doc drift that is required to make the child plan accurate.

Risks

• A documentation-only rebaseline can still break the build because BytecodeProofManifestTests parses and enforces checklist/manifest/source-gate consistency.
• Flattening A1/A2, D1-D4, B24h/B36, E4, or E5 rows would hide the owner that future runtime work must actually retire.
• Broad source searches can overcount test/docs references; production-source absence must be scoped to the source paths used by the manifest tests.
• Route-hit/profile numbers in the docs are dated evidence. Build must capture current numbers before claiming a new final-proof baseline.
• If build discovers the durable proof declaration is wrong, repair the durable task/proof contract first; do not weaken a proof to pass an unrelated edit.

Additional quality gates

• Baseline evidence gate - Before editing files, record current HEAD, checklist totals, manifest non-done ids, runner AST-seam scan result, E4 tombstone scan result, and source owner scan notes in ## Build Update or the build handoff.
• Source-of-truth gate - Source Context and Faktorial API output remain authoritative for issue state; use source/docs only to verify current repo drift, not external GitHub reads.
• No parking gate - A1/A2 and terminal D/E residue must not absorb ordinary A51, B24h, B36, E4, or E5 work.
• Manifest closure gate - No row may move to done unless the manifest proof row proves admission, hard quarantine, fallback retirement, or source absence as applicable; non-done rows must retain at least one open or hard-quarantined proof.
• E4/E5 separation gate - E4 deleted bridge names remain source-absence ratchets; runner-internal expression execution and remaining ExecutionPlanRunner construction stay in E5.
• Change-risk routing - requested behavior: finite bytecode retirement rebaseline plus child conversion; risk lane: full; lane evidence: proof manifest, source-presence/source-absence tests, current source scans, route-hit probes, and allocation profile; required proof or human judgment: build must run the proof obligations and preserve owner classification; buyer consequence if missing: unsupported JavaScript can be falsely marked bytecode-owned or required fallback can be hidden; stop/downgrade trigger: if current source contradicts the manifest or proof declaration, stop runtime work and repair the durable proof contract before implementation.

Proof obligations

• proof-manifest | required | runner: repo-command | command: rtk dotnet test tests/Asynkron.JsEngine.Tests --filter "FullyQualifiedName~BytecodeProofManifestTests" | terminal: passes
• expansion-contract | required | runner: repo-command | command: rtk dotnet test tests/Asynkron.JsEngine.Tests --filter "FullyQualifiedName~ExpressionProgramCoverageMapTests" | terminal: passes
• production-eligibility | required | runner: repo-command | command: rtk dotnet test tests/Asynkron.JsEngine.Tests --filter "FullyQualifiedName~UnifiedBytecodeProductionEligibilityTests" | terminal: passes
• production-invocation | required | runner: repo-command | command: rtk dotnet test tests/Asynkron.JsEngine.Tests --filter "FullyQualifiedName~UnifiedBytecodeProductionInvocationTests" | terminal: passes
• prototype-vm | required | runner: repo-command | command: rtk dotnet test tests/Asynkron.JsEngine.Tests --filter "FullyQualifiedName~UnifiedBytecodePrototypeTests" | terminal: passes
• runner-ast-seam-scan | required | runner: repo-command | command: rtk bash -lc '! rg "EvaluateLegacyAstExpression\(|EvaluateLegacyAstExpressionSlow\(|ProfileEvaluateExpression\(" src/Asynkron.JsEngine/Ast/TypedAstEvaluator.ExecutionPlanRunner*' | terminal: passes
• e4-tombstone-scan | required | runner: repo-command | command: rtk bash -lc '! rg -n "EvaluateStandaloneExpressionProgram|EvaluateLoweredExpressionProgram|EvaluateDynamicExpressionProgram|UnifiedBytecodeExpressionProgramExecutor\.ExecuteDynamic|ExecutionPlanRunner\.ApplyStandaloneBindingTargetProgram|ProfileEvaluateExpressionProgramLoop|ProfileEvaluateLoweredExpressionProgramLoop" src/Asynkron.JsEngine tools/ProfileRunner -g "*.cs"' | terminal: passes
• route-hit-probes | required | runner: repo-command | command: rtk bash -lc './tools/profile forloop --route-hits && ./tools/profile propertyaccess --route-hits && ./tools/profile functioncalls-lite --route-hits && ./tools/profile activation-noargs-lite --route-hits && ./tools/profile forofiteration --route-hits' | terminal: passes
• allocation-stability | required | runner: repo-command | command: rtk ./tools/profile forloop --memory | terminal: passes

Recommended Labels

• bytecode
• unified-bytecode
• ir
• runtime
• compiler
• documentation
• planning
• size:XL
• priority: high

Acceptance Criteria (Refined)

AC-1: The build update captures the pre-edit baseline with current HEAD, checklist totals, manifest non-done ids, runner AST-seam scan result, E4 tombstone scan result, and source owner scan notes.

AC-2: docs/plans/bytecode-burndown-checklist.md, docs/plans/bytecode-proof-manifest.json, docs/bytecode-progress.md, and docs/unified-bytecode-expansion-contract.md agree on current open rows, counts, owner files, proof anchors, and dynamic-residue boundaries.

AC-3: Remaining open rows are converted into owned child slices with exact source files, proof anchors, and explicit boundaries for A1/A2, A51 compiler leaves, B24h, B36, E5b, E5c, E5d, and E5e; A51g/A51i lanes must not be dropped just because they lack manifest closure rows today.

AC-4: E4 stays closed as source-absence/tombstone inventory unless current production source contains a live E4 bridge; any such contradiction repairs the manifest/checklist before runtime work proceeds.

AC-5: No child slice or doc wording treats terminal dynamic residue as a parking bucket for ordinary compiler, class-definition, expression-program, or runner-retirement work.

AC-6: BytecodeProofManifestTests continues to prove checklist/manifest status consistency, non-vacuous source discovery, exact source-presence/source-absence rows, and classified E5 fallback ownership.

AC-7: docs/unified-bytecode-expansion-contract.md keeps current decline-template inventories aligned with compiler and eligibility source; stale examples are removed or reclassified with proof.

AC-8: The proof obligations above pass before review or verification is requested; any failed proof is reported with the exact command, failure, and next repair owner.

AC-9: Final build notes record both pre-edit baseline and post-change final signal for the selected stable evidence metric, even if the build turns out to be evidence-only.

AC-10: The build handoff records created child task ids, their lineage back to gh3377, and any rows intentionally left as terminal residue or human-governed future work.

[faktorial-ai]

Build Update

Baseline timestamp: 2026-07-16T17:50:53Z
Baseline signal: rtk git rev-list --count HEAD..origin/main = 4 commits behind current origin/main
Final timestamp: 2026-07-16T17:52:40Z
Final signal: rtk git rev-list --count HEAD..origin/main = 0 commits behind current origin/main
Signal delta: -4 commits behind; branch is rebased onto origin/main f3e33c7

Repair focus: fixed the conflict-resolution bounce by integrating the closed fk12 main-red repair into gh3377 and recording explicit repair evidence. The prior rejection was procedural metadata, not a missing source implementation.

Committed state: branch agent-go/task-gh3377 is clean at 57d50bb. The rebased commits are 8130e1b (docs: rebaseline bytecode retirement inventory) and 57d50bb (docs: align bytecode runner retirement prose). No push was performed.

Changed paths: docs/bytecode-progress.md, docs/plans/bytecode-burndown-checklist.md, and docs/unified-bytecode-expansion-contract.md. Diff size against current origin/main is 3 files, 104 insertions, 41 deletions.

Conflict repair: resolved the docs/unified-bytecode-expansion-contract.md rebase conflict by preserving the gh3377 source-evidenced static-block/standalone-expression rows. Current source has no ExecutionPlanRunner.RunScript( call in ClassDefinitionExtensions.cs, and the static-block row now names classified-static-block-ir-fallback plus the source-absence ratchet.

Evidence: rtk git diff --check origin/main...HEAD passed; rtk rg -n "<<<<<<<|=======|>>>>>>>" docs/bytecode-progress.md docs/plans/bytecode-burndown-checklist.md docs/unified-bytecode-expansion-contract.md returned no matches; runner AST seam scan passed; E4 tombstone scan passed; rtk jq -r '.items | length' docs/plans/bytecode-proof-manifest.json returned 32; manifest non-done ids are A1 A2 A51b A51a2 A51c A51j A51h A51h1 A51h2 A51k A51f3 A51f5 B24h B36 E5b E5c E5d E5e.

Checklist/final signal: checklist grep count is 138 done / 23 open markdown rows; the maintained row-accounted doc text states 138 / 161 complete, A+B+C+D 127 / 146, Phase E 6 / 10, and 18 non-done manifest rows.

Child-lane handoff: the docs now state the repository ledger is the child-slice contract because gh3377 is not an automatic Faktorial plan-expansion task. Active related handoffs are gh3625 for the B36 deferred class-definition bridge proof and gh3495 for E5 runner-retirement push; A1/A2, A51 leaves, and B24h remain explicit not-yet-split lanes.

Faktorial runtime notes: none were present on the task. Proof declarations were not changed. No C# files were edited, so Rider file diagnostics were not applicable.

Deferred verification: xUnit, route-hit, and allocation proof commands were not run locally per build-lane rules; run-quality is requested for review to execute the canonical proof/quality gate.