Skip to content

docs(runbook): the two views-models blockers are cleared - #141

Open
Polichinel wants to merge 1 commit into
developmentfrom
docs/blockers-cleared
Open

docs(runbook): the two views-models blockers are cleared#141
Polichinel wants to merge 1 commit into
developmentfrom
docs/blockers-cleared

Conversation

@Polichinel

Copy link
Copy Markdown
Contributor

Docs only. MONTHLY_REFRESH.md listed views-models#399 and #403 as blocking a delivery; both resolved overnight.

Verified from this side rather than taken on trust — a merged PR is not the same as a correct declaration:

#399deliveries/un_crafd.py now declares intent = live(since=date(2026, 8, 14)), backdated to when the first delivery actually ran rather than to the merge. Interlock chain traced end to end: livestatus.upload_armed()config_meta.py's wire_upload_enabled → the manager's upload gate. A run now delivers.

#403 — both launchers pin views-postprocessing 1.1.1. Both is load-bearing: they share one conda prefix, so moving only un_crafd would have left the armed FAO leg on the defect. views-models verified the sequencing before merging #399 — armed on 1.1.1 the pin-safety guards pass 6/6; on 1.1.0 they fail 4/6.

Also recorded, with its caveat: the launcher compares direct_url.json against the pin to confirm what pip actually installed (views-models#385) — but warns and continues if that file can't be read, so a silent no-op reinstall stays possible in that narrow case.

uv run pytest -q -m "not integration" → 1096 passed.

…s verified

views-models#399 merged and #403 closed overnight. MONTHLY_REFRESH.md listed both
as blocking a delivery; it no longer should. Checked from this side rather than
taken on trust, because a merged PR is not the same as a correct declaration:

  #399  deliveries/un_crafd.py now declares intent = live(since=date(2026, 8, 14)) —
        backdated to when the first delivery actually ran, not to the merge. The
        interlock chain was traced end to end: live -> status.upload_armed() ->
        config_meta.py's "wire_upload_enabled" -> the manager's upload gate. All
        three hops present, so a run now delivers rather than staging locally.

  #403  both launchers pin views-postprocessing 1.1.1. BOTH is the load-bearing
        part — they share one conda prefix, so moving only un_crafd would have left
        the armed FAO leg on the defect. views-models verified the sequencing before
        merging #399: with both armed on 1.1.1 the pin-safety guards pass 6/6, on
        1.1.0 they fail 4/6.

Also recorded: the launcher verifies what pip actually installed by reading
direct_url.json and comparing it to the pin (views-models#385, after a
version-string pin silently no-opped a reinstall). With the caveat that if that
file cannot be read it warns and CONTINUES rather than stopping — so a silent
no-op reinstall remains possible in that narrow case. Their trade-off, stated here
so a reader of this runbook knows the guard is not absolute.

Nothing else changed. The refresh is still manual, and data-freshness.yml still
alarms if a month is missed.

Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
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