chore(ci): refresh the slow lane's weights from run 37598785795

[NicholasEhsanRoy]

Regenerates tests/slow_lane_weights.json from the junit artefacts of the last full green slow-lane run, 37598785795 (commit fa800c98, all 12 shards green, 58 to 98 minutes per shard). The provenance names only this run, since it measured every file. The table now lists 211 files (185 before: 31 new, 5 dropped, the rest re-weighed). tests/usd was measured but is a directory, not a file, and is left out by the tool. The prediction sentence in docs/developer_guide/testing_standards.md moves from "57 to 67" to "71 to 81" minutes. No change to the shard count, workflow, tests/_sharding.py or any test; the previous weights commit did not touch CHANGELOG.md, so this one does not either.

Predicted minutes of test time per shard (six weighted shards)

lane	shards 1..6	worst
jax 0.10.2, old table, same measurements	58.3 69.1 79.3 90.6 56.4 95.7	95.7
jax 0.10.2, new table	71.1 76.2 79.1 74.0 75.1 73.8	79.1
jax 0.11.2, old table, same measurements	68.9 74.8 68.8 88.7 82.3 89.3	89.3
jax 0.11.2, new table	80.3 79.7 74.2 79.8 80.7 78.1	80.7

(Old-table rows: the old table's weights dealt with the tool's own deal, files it did not list by path hash, loaded with this run's per-file measurements. This reproduces the 58 to 98 minute spread of the run.) The worst shard drops from 95.7 to 79.1 minutes (jax 0.10.2) and from 89.3 to 80.7 (jax 0.11.2), well under the 100-minute target.

Heaviest files (minutes on jax 0.10.2, jax 0.11.2)

36.4  42.5  tests/property/test_sysid_truth_recovery.py
26.2  29.7  tests/compliance/test_guard_mutations.py
16.8  14.5  tests/property/test_differential_coupling_topologies.py
 9.7  14.4  tests/core/test_coupling_spectral_bound_on_the_transformed_reading.py
13.9  14.1  tests/property/test_rest_write_sequences_leave_a_graph_that_reloads.py
 9.0  14.0  tests/property/test_differential_coupling_interactions.py
13.5  13.7  tests/property/test_rest_requests_generated_from_the_schema.py
13.0  13.0  tests/core/test_coupling_claims_graphs.py
12.1  12.0  tests/property/test_coupling_nonlinear_search.py
12.1  10.7  tests/cloud/multigpu/test_run_pod_seeded_faults.py
11.8  11.4  tests/property/test_coupling_targeted_search.py
 6.1   9.8  tests/property/test_coupling_error_bound.py
 8.8   7.8  tests/property/test_round_trips.py
 8.5   7.0  tests/property/test_sysid_contract.py
 7.8   7.7  tests/core/test_coupling_diagnostics_leave_the_state_alone.py

No file is near 45 minutes alone except tests/property/test_sysid_truth_recovery.py at 42.5 minutes on jax 0.11.2 (36.4 on 0.10.2); that is the floor on any shard's load, and it still leaves the worst shard at 80.7 minutes.

Checked locally (collection and gates only; CI is the test run)

• tests/compliance/test_ci_sharding.py: 63 passed (includes the guard that no table entry names a missing file and the 100-minute cap).
• All nine scripts/check_*.py gates: exit 0.
• Collection only, -m slow, MADDENING_TEST_SHARD=k/6:weighted: 115 + 404 + 376 + 161 + 212 + 389 = 1657, equal to the unsharded 1657 slow tests.

🤖 Generated with Claude Code

https://claude.ai/code/session_013UkCde7g23gTziUjYAvnKD