@@ -15,8 +15,8 @@ A task's checks are active from the moment it lands, as listed in the design's t
 | T1.5 `sdm-core` Run state machine | 1a | done | 1aae431; CI run 37788929024 green (22 `sdm-core` tests: 14 run-state, 8 dispatch, composition root, profiles and capability checks) | `validate_query` dispatches and resolves its profile, then refuses as unsupported until T1.6a puts the interim parser behind `sdm_core::pql`; the model capability checks run over `QueryFeatures` the adapter fills; the other three operations refuse as unsupported until Phase 1b |
 | T1.6a Interim parser port and adapter | 1a | done | c01d79e, dbb10d8, 7a425a3, 23acca0; CI run 37802846877 green (`rust-lint` checks both schema locks; `rust-test`: the 417-query corpus through the adapter, 416 agreeing, the 34 journey queries recorded on the reference, the adapter's own tests) | The archive's parser and seven validators behind `sdm_core::pql`, capability checks in `sdm-core` (the pre-resolution `MODEL_NO_TEMPORAL_TARGET` check; the two filter capabilities consulted in resolution, E11); contract version 2 with PQL's row and column spans (E7); the validated-query format, version 1, locked apart (E8); the allowlist holds `pyint`, `float_repr` and `escapes` (E9); an interim date reader without D4's warning and D8 (E10); the corpus and journey queries live in `sdm-core/tests/pql` (E12). The one corpus difference is D4's warning; message text is not promised and differs |
 | T1.7 `sdm-host` | 1a | done | a7a5a02, 4ebb1ae, 1b7b68e; CI run 37804761504 green, every job (`rust-test`: `sdm-host` codec 7, admission 8, runs 12 tests; `sdm-core` `Run::end`) | Codecs and the 64 KB cap; the admission table, with `VALUE_OUT_OF_RANGE` for a value the admitted type cannot hold; `HostRun` with bytes in and out, contained calls, `RUN_BUSY`, E2's attachment rules, admission refusals ending the run (`Run::end`), and zones per run from the host facts (E13). The batch 2 code review's blockers (an uncatchable stack overflow on long AND chains; refusals of types the reference accepts) and should-fix items are T1.7c's exit (H27); the exit refuses only families that cannot convert losslessly |
-| T1.7b Contract v3 | 1a | in progress | Topic branch `rust-core-phase-1-t1.7b` at 305c73c (after the review fixes), one package, CI run 37830953836 green on it, every job; `BASE_SHA` c17d51891b2720e931309f55b9738bcd05d625ef (the phase tip at package start), `CHECKPOINT_SHA` the same (the candidate's merge base with it); before T1.14 the candidate's own harness runs and no acceptance is claimed. Exit tests: `sdm-host` `contract_v3` (11: another contract version refused first; the 5,000-column schema with a query past 64 KB, the start envelope and attachments either side of the cap, the wide `Fetch` step, the `validate_query` result, a real validation past the cap, the error and the warnings, each arriving whole; the run-numbered attachment; retryability; POSIX local zones), `codec` (the codec-level `Done` with a `SourceDescription` either side of the cap); `sdm-contract` `messages` (seed `2**70`, `Span` rows from 1, `ErrorCode`, provider errors without text) and `schema_lock` (the annotation-free hash); `sdm-core` `dispatch` (start attachments, the negative seed); `tests/test_rust_contract.py` (working entry, merged entries, the projection) | Contract version 3 (E15): input-sized fields as attachments (H5); exhaustive `ErrorCode` with `retryable` (H6); the hex seed, the annotation-free hash, over-cap core messages as `attached` (H7); errors from `HostRun` as control messages; the lock collapsed to one working entry (H4); test-only operations behind `test-ops`. Awaiting independent review and landing; owner check-in after it (H12) |
-| T1.7c Batch 1 and 2 code fixes | 1a | in progress | Topic branch `rust-core-phase-1-t1.7c`, one package, stacked on T1.7b's candidate 305c73c because its exit needs v3 (over-cap output, test-only operations through `HostRun::start`); `BASE_SHA` c17d51891b2720e931309f55b9738bcd05d625ef and `CHECKPOINT_SHA` the same until T1.7b lands, then rebased onto that landing with both reset. Exit tests: `sdm-host` `faults` (check (a): an extra column, swapped same-typed columns, a decimal scale, a timestamp unit, metadata, a later batch; nulls in a non-nullable column; check (b): a column lacking, added, renamed, reordered, of another type, without required metadata or with nulls the request does not take; panics at start, in the body and on the adapter's stack; `fields.panic` not poisoning; canaries in malformed messages and attachments, Arrow values and a credential, and panic payloads, in errors and, in a child process, stderr; the adapter-stack panic is inside the interim parser), `widening` (each widened type's values, validity and received type; split and single-value refusal at a test-only capacity; ), `admission` (refusals name the type received; the still-refused families, generated from the family table, against HOST-2), `runs`; `sdm-core` `pql_adapter` (10,000-link chains, trees tall by NOTs under a chain or nested groups refused in bounded time, depth boundaries, DT-9 pinned without quoting the literal, E11 precedence), `pql_stack` (one stack per call, the tallest accepted query of each shape and mix round-trips), `pql_corpus` (`KNOWN_DIFFERING`, readings, multiset errors), `pql_journey` (34 cases, six skipped by name) | H16, H17 and H27 as built (design, "Core, host and binding" and "Interim parser"). Before T1.14 the candidate's own harness runs; no acceptance is claimed. Awaiting independent review and landing |
+| T1.7b Contract v3 | 1a | done | Landed at 6be9820 (fast-forward of the reviewed candidate 305c73c plus its landing commit); reviewed at 5cbcde8 and 305c73c, no blockers; CI run 37833188732 green on 6be9820; `BASE_SHA` and `CHECKPOINT_SHA` c17d518. | Contract version 3 (E15): input-sized fields as attachments (H5); exhaustive `ErrorCode` with `retryable` (H6); the hex seed, the annotation-free hash, over-cap core messages as `attached` (H7); errors from `HostRun` as control messages; the lock collapsed to one working entry (H4); test-only operations behind `test-ops`. Awaiting independent review and landing; owner check-in after it (H12) |
+| T1.7c Batch 1 and 2 code fixes | 1a | done | Rebased onto the T1.7b landing 6be9820, so `BASE_SHA` and `CHECKPOINT_SHA` are 6be9820; reviewed at 7dbd907 (one blocker: tree height) and 7564501 (blocker and should-fix closed, no new blocker); landed by its landing commit. | H16, H17 and H27 as built (design, "Core, host and binding" and "Interim parser"). Before T1.14 the candidate's own harness runs; no acceptance is claimed. Awaiting independent review and landing |
 | T1.8 `sdm-py`, the client wheel and CI rewiring | 1a | not started | | From T1.7: bind `HostRun` (bytes in and out, attachments by id) and raise `_native.CoreError` from the error it returns; containment is already in `sdm-host`. From re-review: the `pyo3-log` bridge with `tracing`'s `log` feature, GIL, clean-import and abort tests, and a delivered log record (H8, F5); the local macOS build for the reference comparisons, its SHA-256 in identity (owner answer 5) |
 | T1.9 Python host and routing seam | 1a | not started | | The exhaustive error mapping (H6); the level-cache reset; no chained driver exception (H17); a wide-schema `materialize()` through the installed wheel (H5) |
 | T1.10 First S0 case on every runner | 1a | not started | | Guard name list updated on `rust-core-reference` before any reference observation is recorded; records through `run_python.py` in `python-ref` mode until T1.14 lands `gate.sh`; `python-old` is a smoke runner (H10); a recorded `RuntimeError` is an outcome, not a runner failure |