Skip to content

Level master with develop (CI green) - #3

Merged
jdogresorg merged 9 commits into
masterfrom
develop
Aug 18, 2026
Merged

Level master with develop (CI green)#3
jdogresorg merged 9 commits into
masterfrom
develop

Conversation

@jdogresorg

Copy link
Copy Markdown
Contributor

Levels master with develop so master carries the comment-hygiene cleanup (internal work-tracking references removed from comments and test labels).

Routine leveling, same shape as the previous one. The diff is develop's accumulated work; CI is green on develop at this SHA.

The contributor agreement says the template carries an agreement checkbox;
no template existed in any repo, so the sentence described something that
was not there.
With no ref, actions/checkout fetches the sibling repo's DEFAULT branch,
which D7 (2026-08-14) keeps at master (released code). Develop CI was
therefore comparing develop code against master siblings, which held only
while master stayed leveled with develop; the LIST-memo lane diverging
develop across xchain-indexer/sdk/explorer turned every cross-repo gate
red (drift-guards both directions, explorer schema canary). Same idiom as
the .ci-siblings SIBLINGS_REF blocks; the venue push gate already ships
siblings at the pushed branch (XC-1494). XC-1519.
The pre-push venue gate ran one of the several jobs ci.yml fans out on
GitHub, so a push could gate green locally and go red upstream on a job
the gate never ran (2026-08-15: that happened on three repos at once).
bin/ci-full.sh transcribes every push-triggered job's run-steps in job
order, fails loud on a missing sibling or a missing runtime rather than
skipping, and reports every red tier the way GitHub reports every red
job. When ci.yml gains or changes a job, ci-full.sh changes in the same
commit. XC-1520.
…1521]

The AML gap board flagged 16 files here for the sleep-flake rubric check. Seven
carried a real fixed settle and are converted onto the repo's existing
waitFor helper, waiting on the post-condition the following assertions read: a
socket reaching OPEN before a single-shot broadcast that a still-handshaking
subscriber would miss permanently, and replica row counts after a poll.

The other nine are left alone deliberately and are not defects: a sleep whose
following assertion says nothing happened is the test itself, and converting it
to a condition poll returns immediately and proves nothing.

server-websocket's isolation case now waits for both frames bitcoin is owed
rather than guessing 200ms, which is a strictly stronger window for the
litecoin-got-nothing assertion that follows it.

Unit suite green at 1748 passing; the e2e and integration suites need a live
stack this venue has not got, so their behaviour is unverified here.
Findings adjudicated in the 2026-08-16 xchain-platform review round.
Every claim was re-checked against current code and every remedy was
re-derived rather than applied from the finding's recommended option;
each verdict then passed an adversarial verify stage before booking.

Findings:
  #5073  tableLifecycle.js state_tree_nodes note is stale: the "awaited" mark-and-sweep pruner already shipped in retention.js
  #5101  sync_meta content-parity digest hashes node-local id/logged_at columns, guaranteeing a false TABLE_CONTENT_PARITY mismatch

Report: claude/reports/2026-08-16_review-round-xchain-platform.md
Replaces the inline sibling-checkout script with the composite action published
in the organization .github repo, so one definition serves every call site
instead of ten copies that had drifted apart.
@jdogresorg
jdogresorg merged commit 755facf into master Aug 18, 2026
12 checks passed
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant