backlog: file #1331 -- one test name over six assertions makes a merge-blocking flake read as six bugs - #533
Merged
Conversation
wshallwshall
enabled auto-merge (squash)
August 22, 2026 23:01
wshallwshall
disabled auto-merge
August 22, 2026 23:01
…e-blocking flake read as six bugs Content routed here by the dispatcher seat, which allocated no number and cited none. Filing routes to the lander because allocation and the commit that files it cannot be split across worktrees. WHAT IT COST, mine: test_connscale_smoke_end_to_end blocked THREE PRs in one evening -- 516, 530 and 529 -- on required contexts. WHY IT DOES NOT READ AS ONE PROBLEM, which is the row: it wears two different assertions. A throughput SLO (fixed_aggregate at N=24, 25.1 against a prior 34.5 * 0.75) and an intake loss (engine_read 35 below confirmed sent 36). Two seats hitting it twice see two unrelated bugs. THE DISPATCHER MEASURED THE CAUSE: the test is 273 lines and the end-to-end case asserts at least SIX separate properties under one name. That is why nobody has owned it, and it is why the fix direction is to make the failure NAME itself rather than to re-run it. ATTRIBUTION IS IN THE ROW because it is the reusable half: attribute by FILE SCOPE before re-running. 529 and 530 each touch zero files under pipeline, store or transports and neither touches the test, so neither can reach the intake path. That is what made both re-runs earned rather than manufactured. AND WHAT IS NOT MEASURED IS STATED: whether the two assertions share a root cause is unknown, neither failure was reproduced locally, and the three-PR count is one evening rather than a rate. Verified: parse_items 339 items / 245 open against main's 338 / 244, zero lost from either side, #1331 present and open. backlog_status_check green over 575.
wshallwshall
force-pushed
the
lander/1331-connscale-flake
branch
from
August 22, 2026 23:03
43b905e to
55f0acb
Compare
wshallwshall
enabled auto-merge (squash)
August 22, 2026 23:03
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
1 participant
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.
Ledger-only, one file, +28/-0. Content written by the dispatcher seat, which allocated no number and cited none; filing routes here because allocation and the commit that files it cannot be split across worktrees.
The row
tests/test_connscale_smoke.py::test_connscale_smoke_end_to_endblocked three PRs in one evening -- #516, #530 and #529 -- ontest (ubuntu-latest, py3.14)and the Windows legs, which are three of the thirteen required contexts.Why it does not read as one problem: it wears two different assertions.
The dispatcher measured the cause on
origin/main: the test is 273 lines and the end-to-end case asserts at least six separate properties under one name. Two seats hitting it twice see two unrelated bugs.Fix direction is in the row so it does not become "re-run it": make the failure name itself -- split the assertions, or carry the failing property into the message.
The attribution rule, which is the reusable half
Attribute by file scope before re-running. #529 and #530 each touch zero files under
pipeline/,store/ortransports/, and neither touches the test, so neither can reach the intake path. That is what made both re-runs earned rather than manufactured -- and both went green, which is the check on the method.What is NOT measured, stated in the row
Whether the two assertions share a root cause is unknown. Neither failure was reproduced locally; both figures are read off CI. The three-PR count is one evening's observation, not a rate.
Verification
parse_items: 339 items / 245 open against main's 338 / 244, zero lost from either side, #1331 present and open.backlog_status_check.pygreen over 575 items across both files. Ledger gate passed on commit, which is what confirms the number was allocated rather than chosen.Opened unarmed, behind #458 -- arming a second PR now would only re-BEHIND the one carrying #1005 and #1314.