Repository navigation
Close forced-lineage and stale-sidecar gaps in the served-model check - #101
Merged
Merged
Conversation
The July v5 sha pin only fired when checkpoint_sha256 was empty, so a hand-written forced lineage with any fake checkpoint skipped it. Pin every forced: true lineage unless a matching promote.py manifest verifies the bundle. Marking any sidecar stale skipped run-id drift. Limit that exemption to mtnn_jacobian.json and mtnn_attr_pop.json. Co-authored-by: Cam Davis <jcdavis131@gmail.com>
|
The latest updates on your projects. Learn more about Vercel for GitHub.
|
jcdavis131
changed the base branch from
cursor/served-model-lineage-b86c
to
master
October 19, 2026 03:02
jcdavis131
marked this pull request as ready for review
October 19, 2026 03:05
This branch was successfully deployed
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.


Stacked on #100 (
cursor/served-model-lineage-b86c@ e11c2a5). Does not merge #100 or change its branch. Served model stays July v5 (july-v5-a4918f09); nothing is promoted.Gaps
An adversarial review of #100 found two fail-open paths in
pipeline/served_model.py:forced_adoption_pin_problemonly ran whencheckpoint_sha256was empty (oradoption.kind == served-incumbent). A hand-written lineage withforced: trueand any fake non-empty checkpoint skipped the July v5 sha pin.stale/not_from_served_netskipped the run-id drift check. A live sidecar could mark itself stale and drift freely.Fixes
forced: truelineage whoseembedding_f32_sha256is not the pinned July v5 sha fails, regardless ofcheckpoint_sha256or kind.promote.pymanifest on disk (CURRENT.json +promoted/<run_id>/manifest.jsonhashes, including checkpoint). Default is fail-closed. This keeps the realpromote.py --forcepath (tests/test_e2e_smoke.py) working.mtnn_jacobian.jsonandmtnn_attr_pop.jsonviaSTALE_SIDECAR_ALLOWLIST. Any other sidecar marked stale still fails the drift check.Tests
test_a_forced_lineage_with_a_fake_checkpoint_sha_fails— forced + fake non-emptycheckpoint_sha256+ wrong f32 sha fails.test_a_non_allowlisted_stale_sidecar_with_drifted_run_id_fails— e.g.mtnn_map.jsonmarked stale with a drifted run id fails.test_allowlisted_stale_sidecars_with_drifted_run_id_pass— jacobian and attr_pop still pass.Existing forced-empty-checkpoint, committed-bundle, and real
--forcetests still pass.Checks
Attached to this PR (base is not
master, so GHA workflows that only trigger onmaster/maindid not run):Local equivalents of the #100 required jobs, all green on this head (
8bfc068d):scripts/check_served_model.py)tests/test_site.py+ stamp/sync/facts)scripts/budget_check.py)tests/test_check_served_model.py+tests/test_export_lineage.py