REQ-ACCEPTANCE-004: prove MTFX-T5 insertion-order invariance

[zendev-author]

Closes #268

Achieved outcome

The M3 golden-gate acceptance suite (REQ-ACCEPTANCE-004) now proves Handoff/04 §39 MTFX-T5 ("Proportional seller/buyer rationing conserves quantity and is insertion-order invariant"), one of the enumerated remaining acceptance artifacts named on the current ledger row. This is one bounded slice of the composite requirement, not full closure — MTFX-T4, MTFX-T6, MTFX-I5, MTFX-I6 and the full local-shortage orchestration golden-gate scenario remain outstanding, exactly as before this PR.

Tested revision

a3b0c5ab803be7c469f0be8cd0385628cd921862

Changed artifacts

• src/simulation/acceptance-004-m3-golden-gate.test.ts — added a new describe("MTFX-T5: ...") block. It reuses the exact multi-seller (3) / multi-buyer (2) shortage fixture already used by the file's MTFX-I3 block (sellable 50/30/20, effective demand 40/35, cleared quantity 75 < either side's raw total), which forces proportional rationing whose per-seller share for the 20-unit seller is not exactly representable in binary floating point — so computeLocalClearing()'s residual-correction pass actually executes rather than being a no-op. The test calls computeLocalClearing() three times with the identical intents in three different array orders (forward, fully reversed, and a third non-reversed shuffle) and asserts bit-identical per-participant summed fills and conserved total cleared quantity across all three, proving the stable actor/intent-ID sort — not input array position — drives both residual correction and two-pointer matching.
• docs/spec/implementation_status.csv / docs/spec/IMPLEMENTATION_STATUS.md — updated the existing REQ-ACCEPTANCE-004 row's PR field to this PR and appended this PR's contribution to EVIDENCE; STATUS remains PARTIAL (unchanged, correctly — this PR does not close the composite requirement).

Acceptance criteria

Issue #268's own acceptance criteria are for the full composite REQ-ACCEPTANCE-004 (all MTFX-T1..T6/I1..I6 plus the orchestration golden-gate). This PR is one bounded slice within that Issue's remaining work, matching the pattern established by the prior merged PRs #417/#419 against the same Issue:

• MTFX-T5 (Handoff/04 §39): proportional rationing conserves quantity and is insertion-order invariant — proven by the new test.
• MTFX-T4, MTFX-T6, MTFX-I5, MTFX-I6, full orchestration golden-gate scenario — still outstanding, unchanged by this PR.

Checks

Check	Outcome	Evidence
npm run typecheck	passed	clean, no errors
npm test	passed	564/564 tests passed (40 files; 563 pre-existing + 1 new)
npm run build	passed	vite build succeeded
dotnet build --configuration Release --no-restore	passed	0 warnings, 0 errors
dotnet test --configuration Release --no-build	passed	45/45 tests passed (REQ-MIGRATION-003 maintained)
python scripts/implementation_status.py --check	passed	ledger/rendered table in sync after regeneration

Not checked

• No UI/browser verification — this PR touches only TypeScript unit-test/domain logic, no docs/ Pages artifacts.
• Did not re-verify MTFX-T4/T6/I5/I6 or the orchestration scenario; they are explicitly out of scope for this bounded slice and remain tracked on Issue REQ-ACCEPTANCE-004: Golden-gate M3 local-market acceptance test #268.

Assumptions and unknowns

• Assumption: the existing MTFX-I3 fixture's numeric shape (sellable 50/30/20, demand 40/35) is a legitimate reuse for MTFX-T5 since both invariants examine the same proportional-rationing pass from different angles (fill conservation vs. order invariance). No new economic fixture semantics were invented.
• Assumption: toBe (exact equality) is an appropriate assertion strength here because the production sort key is deterministic and the same arithmetic executes in the same order regardless of input array position; if a future legitimate change makes the arithmetic order genuinely non-deterministic (e.g. real concurrency), this test would need loosening — not expected pre-M11.

Highest-risk area for review

The strength of the toBe (bit-exact) equality assertions across the three orderings. If this is judged too strict for legitimate floating-point non-determinism, the fallback is a tight epsilon (e.g. 1e-12) rather than the file's usual quantityEpsilon (1e-8), to keep the test meaningfully sensitive to an order-dependent regression rather than passing vacuously.

Remaining gate

None for this bounded slice. Issue #268 stays open and status:ready after this PR (once merged) for the next AUTHOR run to pick up the next enumerated slice (MTFX-T4, T6, I5, I6, or the orchestration golden-gate scenario). REQ-ACCEPTANCE-004 correctly remains PARTIAL in the ledger.

[zendev-acceptor]

ACCEPT — head revision 6fdb2772363501ba9444239ae1c1a9313d44165a.

1. Checks green at head: build-and-test (CI), typescript (CI), policy-guard (CI), both write-mergeability-status runs, and the mergeability status context are all SUCCESS at this exact head SHA (confirmed via gh pr checks/gh pr view). Branch is not behind master (git log origin/master..pr-424 shows only this PR's own two commits; git log pr-424..origin/master is empty) and mergeable: MERGEABLE.
2. Independent re-execution at this head (checked out into a worktree, not read from the checks tab):
   • npm ci && npm run typecheck — clean
   • npm test — 564/564 passed (40 files), matching the PR body's claim exactly
   • npm run build — succeeded
   • dotnet build --configuration Release — 0 warnings/errors
   • dotnet test --configuration Release --no-build — 45/45 passed (REQ-MIGRATION-003 maintained)
3. Scope: diff touches exactly the three files declared in "Changed artifacts" — src/simulation/acceptance-004-m3-golden-gate.test.ts (new MTFX-T5 describe block, purely additive), docs/spec/implementation_status.csv and docs/spec/IMPLEMENTATION_STATUS.md (ledger row update). No workflow/AGENTS.md/docs/zendev policy paths touched.
4. No invariant or test weakened: purely additive test file change — no deletions, no .skip, no loosened assertions. The new test uses toBe (bit-exact) equality on real per-participant fills computed by the actual exported computeLocalClearing(), deliberately choosing a fixture (20-unit seller's 15-unit share is not binary-exact) that forces the residual-correction pass to actually execute, then checks three distinct intent-array orderings (forward, reversed, shuffled) for bit-identical fills and conserved total — this is a real, non-vacuous proof of MTFX-T5 (insertion-order invariance), reusing the same fixture and helper pattern already established and accepted in the merged MTFX-I3 PR (#419).
5. No secret/credential/personal data/local path anywhere in the diff (read in full).
6. Handoff record complete: achieved outcome, tested revision, changed artifacts, acceptance-criteria table, checks table, not-checked section, assumptions/unknowns, highest-risk area, and remaining-gate statement are all present and accurate. Ledger STATUS correctly stays PARTIAL — this is one bounded slice of the composite REQ-ACCEPTANCE-004, consistent with the pattern of prior merged slices (#417, #419) against the same still-open Issue #268.

Linked Issue #268 carries priority:high, type:feature, area:simulation-core, Goal/Evidence/Scope/Non-goals/Acceptance-criteria/Verification are all present. No standing CHANGES_REQUESTED from any account; no prior reviews on this PR. No status:needs-decision label.

Merging.