Repository navigation
A-4: sign my own row - #466
Conversation
Ticks only my row in A-4's ratification table (cve_mapping joining the check vocabulary). Nothing is ratified by this, and nothing changes in code: cve_mapping stays out of pipeline.CHECKS until the table is full. Touches only this row, so it merges in either order with #381, which signs Samika's.
patelankeet2
left a comment
There was a problem hiding this comment.
Verified against the real current code, not just the description.
Diff is exactly what's described: one line in docs/finding-format.md, Shubham's A-4 row flipped from a blank box to a tick. No code file touched.
Checked the reasoning behind the sign-off, not just the diff:
analysis/checks/risk.pydoes declare_SEVERITY_CEILING = {"cve_mapping": "medium"}, and_apply_ceiling()only ever holds a severity at the ceiling, never lowers below what a check or an R-rule already set -- matches the "not the same as downgrading" claim in the comment above it.cve_mappingis confirmed absent frompipeline.CHECKS(onlyaccess_control,policy_compliance,routingare registered), so A-4's gate still holds: nothing runs it yet.
CI green on both 3.12 and 3.13. mergeable_state: clean.
The merge-order note (adjacent-line conflict with #381) is accurate -- #381 is still open, so no conflict exists yet, but whichever of the two lands second will need the two-line resolution described.
Approving.
SamikaPerera
left a comment
There was a problem hiding this comment.
I checked the diff and each claim in the description against main (ee769b4).
The diff. One line in docs/finding-format.md: Shubham's row in the A-4 table (docs/finding-format.md:498 on main) goes from ⬜ to ✅. Nothing else changes, and no code file is touched.
Claims checked:
- "
cve_mappingstays out ofpipeline.CHECKS": true.CHECKSon main lists onlyaccess_control,policy_complianceandrouting(analysis/pipeline.py). - "
cve_mappingcaps itself atmediumandriskholds that ceiling against business-context escalation": true._SEVERITY_CEILING = {"cve_mapping": "medium"}is atanalysis/checks/risk.py:221, andapply_business_context()applies it through_apply_ceiling(). - "Its
foundis worded as 'worth checking'": true, atanalysis/checks/cve_mapping.py:327. - "#434, section 6": true.
docs/research/cvss-comparison.md:135is "### 6.cve_mapping's cap: ours is right, and CVSS explains why". - "A-4's note still says 'zero of four, as of 8 September'": true, at
docs/finding-format.md:501. - The merge-order note is accurate.
git merge-treeof this head with #381 (d62c209) gives one conflict indocs/finding-format.md: your ✅ on Shubham's row and my ✅ on Samika's row are adjacent lines. Keeping both resolves it.
Run: this branch merged with current main (ee769b4), Batfish running: 2324 passed, 13 skipped, 0 failed. CI on be0847e is green on 3.12 and 3.13 (run 42757236470).
There is no code or test to mutation-check here.
Not checked: whether the A-4 note should drop its count after both rows land. That is a follow-up you name in the description, not part of this diff.
|
This now conflicts with Verified the combined result rather than guessing: merged both locally (not pushed, this is someone else's branch) and ran the full suite clean (2392 passed, 13 skipped, 0 failed). The real conflict is only the two adjacent table rows plus the count note right below them, which both PRs update independently without seeing each other: Not pushing this to your branch -- rebase/update is yours to do. Happy to re-verify (full suite) the moment it's updated. |
# Conflicts: # docs/finding-format.md
|
@patelankeet2, thank you for testing the combination. Resolved in Rather than changing it to "two of four", which goes stale again with the next tick, the note now says "NOT RATIFIED until the table above is full" and keeps its history (zero of four from 8 September, one of four from 5 October). That's the same fix we made to A-3's key set in #449, and what this PR's description proposed. Everything else in the note, including the rule that Merged result: 2418 passed, 13 skipped, ruff clean. This is a merge with a conflict resolution, so per M-2 it needs a fresh look. |
SamikaPerera
left a comment
There was a problem hiding this comment.
Approving 40310c4. Since be0847e there is one new commit, Shubham's merge of main (Thu 19 Nov 14:31). It kept both A-4 ticks, mine from #381 and his, and it rewrote the note below the table, which is new text in neither parent: 'NOT RATIFIED until the table above is full', keeping the history (zero of four from 8 September, one of four from 5 October). A-4 now reads Arsh not ticked, Ankeet not ticked, Shubham ticked, Samika ticked. Merged with current main: clean. The diff against main is Shubham's tick plus that note, nothing else; the tick count goes from 16 to 17, so no row loses a tick and the A-3 tables from #449 are intact. Full suite on the merge: 2453 passed, 63 skipped, with Batfish not running, so the Batfish tests did not run. One thing for the merger: CI on 40310c4 ran against b6b3865, and main moved after that (#449 at 15:02), so rule 3a needs a re-run on the merged result before merging. Not checked: CI on the current main.
ARSH871-bot
left a comment
There was a problem hiding this comment.
Reviewed 40310c4 (main merged into your branch). Approving.
The diff against main is the one row you describe plus the note: your A-4 row goes ⬜ to ✅, and the "one of four, as of 5 October" count is replaced by "NOT RATIFIED until the table above is full. The table is the record", the same change #449 made for A-3's key set. After #381 the table on main reads Samika ✅ and the other three ⬜, so this makes it two, and the note no longer needs touching when a third lands.
Your reasons for the tick check out against the code: cve_mapping pins MAX_SEVERITY = "medium" and ignores the advisories when choosing it (analysis/checks/cve_mapping.py:97-108), and risk holds the ceiling against escalation (_SEVERITY_CEILING = {"cve_mapping": "medium"}, analysis/checks/risk.py:221). It merges cleanly with main (GitHub reports MERGEABLE; #381, the conflicting one, has already landed). A documentation-only change, so I did not re-run the suites.
Ticks only my row in A-4's ratification table (
cve_mappingjoining the check vocabulary, from #239/#304). Nothing is ratified by this, and nothing changes in code:cve_mappingstays out ofpipeline.CHECKSuntil the table is full, as A-4's note says.Why yes. My row's concern was that a new check changes what a scan's totals mean, because
policy_complianceshares the results list.cve_mappingcaps itself atmediumandriskholds that ceiling against business-context escalation, so a train-level match can't outrank a confirmed access-control or policy finding. Itsfoundis worded as "worth checking", not "vulnerable". The CVSS comparison (#434, section 6) reached the same view of the cap from the standard's side.Merge order: this changes one table row only, but my row sits directly above Samika's, and git counts changes to adjacent lines as a conflict (checked with
git merge-tree). So whichever of this and #381 lands second hits a two-line conflict, resolved by keeping both ✅. I'll resolve it if this one is second. Once both land, A-4's note still says "zero of four, as of 8 September", and #381 updates that line. After that, the note should stop repeating the count, as we did for the A-3 key set in #449, so the table is the only record.