Current verification status
Parent tracker: #66. Logical 100% parity is incomplete. The historical adapter-based 4,708-file green result below is not proof that the original assertions exercised equivalent runtime behavior.
The native-only runner and corrected discovery are on main in 1d62a395b, with the config-argument correction in 82156e60d and scanner evidence preservation in 7fa9fc32d, all published as Chris Breuer without coauthors. The synthetic corpus prelude, rewrite engine and bootstrap fallback are removed. All 4,754 original pre-exclusion entries route natively: 1,863 test-runner entries and 2,891 scripts. The exact audit reports zero missing/extra paths or mode differences; 11/11 discovery tests pass. This is inventory evidence, not 4,754 passing files.
The first native-only production build passed 46/46 steps; the focused harness passed 5/6. The failed Node assertion control found split --config path arguments, corrected to upstream CI's --config=path. Direct Home and pinned Bun controls reach and fail the intended assertion with that spelling. Corrected optimized/focused builds are running; their final result is pending. Retained native routes passed HTML 145/one original skip, Blob 65, body/stream 9,467/four skips, Headers/Response 170, Request 24, FormData boundary 5 and microtasks 2. Independent HTML aggregate: 530 assertions, 145 passes and one skip. #701/#694 remain unresolved.
The triage scanner now preserves one attempt, its full individual log and exit code, refuses existing output, and returns nonzero for failed/incomplete/empty scans. It never retries or escalates outer bounds. Pinned Bun controls verified one original-deadline failure, one outer-bound interruption and one pass, each started once; overwrite refusal preserved all logs. Its strict-name discovery still is not the full CI inventory.
A separate preserved Home/Bun comparison of all 78 previously omitted entries is queued after the corrected builds. Docker guards without a daemon receive no Docker coverage credit. Remaining work includes broad native failures, the full CI launch profile, separate script/test/empty accounting, complete gate outcome retention, integration/vendor/service setup and platform/build ownership. This issue remains open and 100% parity remains incomplete.
#703 still owns the full CI launch profile, service/vendor setup, full outcome preservation and the broad runtime failures that native execution exposes. The native capture's 120-second outer bound and upstream CI launch options still need reconciliation; ordinary test deadlines have not been extended. #701 and #694 remain unresolved. Other platform execution and independent native build ownership remain incomplete under #66.
The user's requirement is no workarounds. Preserving an adapter's green status must never justify fabricated outputs, fake memory counters, reduced workloads, extra skips, retries, warm-ups or relaxed original deadlines. Temporary failures must be reported truthfully and tracked while the missing behavior is implemented. Final completion still requires the actual suite to pass logically.
Required completion evidence
- Preserve original executable sources, assertions and workloads; distinguish original upstream skips/TODOs from missing Home behavior and exclude them from implemented coverage.
- Execute the original tests and child processes in the real Home runtime, with faithful upstream discovery, fixture layout, startup context, flags and expected-failure handling. Remove fixture-specific semantic substitutes and fallback success responses.
- Publish exact native results and retain observed failures. A filename count or adapter summary is insufficient evidence of logical parity.
- Verify the complete intended upstream suite with zero unexpected failures, unsupported behavior, Home-injected skips/TODOs or substituted results, including the original memory/lifecycle assertions. Current per-file native matrices do not establish a complete single-VM aggregate.
The final objective is unchanged. The historical requirement to keep the adapter gate green throughout does not override truthful results or the user's no-workarounds instruction.
Historical issue statement
The text below is retained for context, not as a current verification claim:
Parent tracker: #66 Related execution tracker: #201
The full copied Bun test tree now executes green at the file gate (4,708/4,708 files, 0 failures, 0 unsupported registrations, 0 runner logs/leaks), and the copied tree exactly mirrors all 12,996 entries in ~/Code/bun/test aside from Home metadata files.
However, logical 100% still requires distinguishing upstream-intentional TODO/skips from Home-injected rewrites and eliminating the latter by porting the missing behavior. Current injection sites include:
- cron in-process firing and Bun.serve cookie suites
- WebCrypto generateKey WPT vectors
- CSS parser invalid-input fuzzing
- sparse-file behavior
- Bun.color exhaustive coverage
- subprocess IPC/stream blob/bytes behavior
- JSONL stress, OOM, UTF-8, streaming, and adversarial cases
- non-JS loaders and snapshot update/serialization workflows
- test.failing done-callback reporting
Acceptance criteria:
- Add accounting that separates upstream TODO/skips from Home-added TODO/skips.
- Remove each Home-added rewrite only after implementing and validating its underlying runtime behavior.
- Keep the full 4,708-file gate at 0 failures, 0 unsupported, and 0 leaks throughout.
- Reach zero Home-injected TODO/skip rewrites without weakening upstream assertions or stress semantics.
Track focused feature issues from this issue as needed.