@@ -38,7 +38,7 @@ A task's checks are active from the moment it lands, as listed in the design's t
 | T1.15 Coverage generator for S0 | 1a | not started | | Narrowed by the owner 2026-10-11 (E66(f)) to `--check`, the cut proposed at checkpoint 4 |
 | T1.16 L0 guard self-tests | 1a | merged into T1.10c | | Re-review round 2 |
 | T1.17 Clean-container wheel smoke | 1a | not started | | The E16 floor and newest lanes, both gating. Follow-up from T1.8's review: `docs/get-started/installation.md` tells Databricks users to pin the runtime's NumPy, which conflicts with `numpy>=1.26` on DBR 14.3 LTS (NumPy 1.23); the install guidance must be revalidated on DBR 14.3 under E16 before the epic merges |
-| T1.18 L2 in CI | 1a | in progress | Started on lane `rust-core-phase-1-lane-1a` at e7950b3, rebased onto T1.26a's landing; `BASE_SHA` and `CHECKPOINT_SHA` 30d27a2d; candidate complete, for review; lands by its landing commit, whose SHA and CI run the next package records (E25) | Runs before T1.15 and T1.17 (orchestrator, 2026-10-11: approved outcome changes from lane 1b need this integration first, and T1.15's cut is with the owner). `contract-python` per runtime cell (three jobs), `rust-test` once, both through the checkpoint's `gate.sh`; `contract/tests` in `ref-pandas3` on the effective expectations (`ratchet.py effective`); the gate's running failing closed; H1's cell rule in `compare.py`, the floor cell report-only for this run alone (E65(c)), HOST-3 proposed; git's paths NUL-separated; E65; as built in [as-built.md](as-built.md#t118). Measured locally: `gate.sh --base 30d27a2d --run python-core,python-old,rust-host` passes in `ref-pandas3`, as do `python-core,python-old` in `dbr17-pandas2`; `floor-pandas2.2` has 38 `warnings` differences under the checkpoint's comparator, none under the cell rule. E49 probes: no route changed; the cell check refuses 100,000 levels by name; each new rule seeded on missing, empty and retyped evidence. No `status.py`, no new comparator: `compare.py`, `compare-policy.yaml` and the recorded observations consumed. Size: about 300 lines of non-test code and CI (`ci.yml` +160, `compare.py` +66, `ratchet.py` +67, `gate.sh` +7), about 470 of tests |
+| T1.18 L2 in CI | 1a | in progress | Started on lane `rust-core-phase-1-lane-1a` at e7950b3, rebased onto T1.26a's landing and then T1.26b's; `BASE_SHA` and `CHECKPOINT_SHA` c52f2800; candidate complete, for review; lands by its landing commit, whose SHA and CI run the next package records (E25) | Runs before T1.15 and T1.17 (orchestrator, 2026-10-11: approved outcome changes from lane 1b need this integration first, and T1.15's cut is with the owner). `contract-python` per runtime cell (three jobs), `rust-test` once, both through the checkpoint's `gate.sh`; `contract/tests` in `ref-pandas3` on the effective expectations (`ratchet.py effective`); the gate's running failing closed; H1's cell rule in `compare.py`, the floor cell report-only for this run alone (E65(c)), HOST-3 proposed; git's paths NUL-separated; E65; as built in [as-built.md](as-built.md#t118). Measured locally: `gate.sh --base c52f2800 --run python-core,python-old,rust-host` passes in `ref-pandas3`, as do `python-core,python-old` in `dbr17-pandas2`; `floor-pandas2.2` has 38 `warnings` differences under the checkpoint's comparator, none under the cell rule. E49 probes: no route changed; the cell check refuses 100,000 levels by name; each new rule seeded on missing, empty and retyped evidence. No `status.py`, no new comparator: `compare.py`, `compare-policy.yaml` and the recorded observations consumed. Size: about 300 lines of non-test code and CI (`ci.yml` +160, `compare.py` +66, `ratchet.py` +67, `gate.sh` +7), about 470 of tests |
 | T1.19 Local gate | 1a | not started | | Shrunk by the owner (E53(b)): `gate.sh --only` and `--locked`, each locked check shown passable and failing (H26), and a seeded fault only for a check that still lacks one, the existing comparator fault tests counting; the route-proof faults named and kept: a `python-core` observation without its `HostRun` operations, a `python-old` observation with `_native` loaded, and the request-UUID control; its exit no longer requires `--locked 1a`, which T1.20 runs |
 | T1.6b PQL parser completion | 1a | not started | | Before T1.20; DT-9 set to `rejected`; the D8 and DT-9 cases join acceptance; from T1.6a: D8 (fields of five digits or more, now `UNSUPPORTED_REQUEST`) and D4's warning, with the channel it travels on; the 29 quirks and the custom-profile tests from the archive; `KNOWN_DIFFERING` (from T1.7c) empty; DT-9's reference-accepted query flips to accepted; the current year from the local zone; a deep-query subprocess test covering traversal and destruction; product work on the permanent parser, and the crate's rename from `sdm-pql-interim` (E18) |
 | T1.20 Phase 1a review | 1a | not started | | Completed independent review, blocking findings resolved; every Phase 1a check active; held-out cases written with their SHA-256 committed (H11); the S0 differential run once (H11); `python-ref` regenerated; owner check-in after it (H12); carried from T1.18's review to this check-in: a run that is not a push skips the gate, and `land.py` takes the newest run of any event (both before T1.18; `land.py` takes bug fixes only); by H1, a regression the same on both routes is reported, not gated, in a cell other than the reference's |