MovilBA decoupling: the fork was asked for and never landed - gates the retirement stub
Elazar asked bin-lezama-cc to fork gitpush.sh exclusively for MovilBA so gitpush.sh could be changed freely without risking MovilBA CI-CD. MEASURED ON VENUS 2026-08-31: no gitpushmovilba.sh, and no MovilBA-named fork of any kind, exists anywhere in sh.git. movilba-web still has no .gitpush-strict marker - gitdeploy.sh:4 states this in its own version comment as the reason its v1.2 guard was inert - and gitpush.sh still carries movilba-web-specific branches at :1046 and :1140. What is NOT known from venus and must not be guessed: whether MovilBA agents still INVOKE gitpush.sh/gitdeploy.sh, or have moved to their own path outside sh.git. That is host-local. One answer settles it. This item is the decoupling itself, and it GATES the retirement stub (OPS-177): retiring gitpush.sh while MovilBA still calls it breaks a CI-CD with multiple branches and remotes. Either the fork lands, or MovilBA is confirmed off gitpush.sh, before OPS-177 ships.
Activity
-
parent=#3117
-
ANSWERED BY MEASUREMENT, AND THE ANSWER UNBLOCKS THE RETIREMENT. All four mba-l lanes replied in one round. MovilBA does NOT invoke gitpush.sh. Dev is 'gitdeploydev.sh --bump "msg"', prod is 'gitdeployprod.sh' on a fresh explicit per-deploy GO from Elazar, never carried over. coder02-mba-l-cc measured it on lezama instead of asserting it: gitdeploydev.sh does not call gitpush.sh anywhere - the only textual hit is a comment at :205 ('same intent as gitpush.sh website flow') - and it pushes with raw git push at :402 and a loop over extra remotes at :439. So the named fork Elazar asked for was never written, and the coupling it was meant to remove is nonetheless already gone. The risk he was protecting against - changing gitpush.sh and breaking MovilBA's multi-branch multi-remote CI-CD - does not exist today. OPS-177 (the retirement stub) is UNBLOCKED on MovilBA's account. Residual coupling is in the OTHER direction and stays: gitpush.sh:1046 and :1140 still carry movilba-web-specific text, and gitdeploy.sh:4 still cites movilba-web's missing .gitpush-strict marker. Dead references to a repo that no longer calls it - delete with the monolith, not before. NEW HAZARD, out of scope here, filed separately: the two coder lanes run two DIFFERENT COPIES of gitdeploydev.sh.
-
CORRECTING MY OWN CLOSE. bin-lezama-cc measured on lezama, the host movilba-web actually lives on, and falsified one of my claims. Verified on venus before accepting it - both lines read, both are prose: gitpush.sh:1046 is a COMMENT inside the --review design rationale ('the project token is not the repo name either (audit-mba-l-ca's repo is movilba-web)'), and :1140 is the TEXT OF AN ERROR MESSAGE explaining why .gitpush-review must be declared rather than derived. Neither is a branch, neither tests for movilba, and gitpush.sh contains ZERO movilba-specific code. I wrote 'still carry movilba-web-specific branches at :1046 and :1140' into OPS-173 and into my report to Elazar. That was wrong: I ran a case-insensitive grep and read filename-and-line-number hits as control flow without opening either line. Same wrong-instrument shape as reading absence from git status - a grep answers 'does this string appear', never 'does this code branch'. CONSEQUENCE FOR THE PLAN, and it is the useful half: there is nothing to fork OUT of gitpush.sh for MovilBA. bin-lezama-cc's verdict - 'do not build the fork, it would be a 4th uncoordinated copy, the exact drift you reject in OPS-174' - is correct and I withdraw the fork as a deliverable. Elazar's original instruction to fork was sound when given; the coupling it targeted has since disappeared by another route, so executing it now would ADD the problem it was meant to prevent. Their dating, which I did not have: last real gitpush.sh run inside movilba-web was 2026-08-22 18:03 ART, 24 invocations total, all on 08-21/08-22. Coverage stated honestly - 16 CC transcripts back to 2026-08-03 plus ~/.codex/sessions with 0 hits, so the four -ca lanes are covered; gitdeploydev.sh was NOT executed, and the finding says nothing about repos other than movilba-web. Also theirs, and it stays open under someone else's name: movilba-web TRACKS .gitpush-pre.sh with GITPUSH_INFO_VERSION pinned at 4.41. No gitdeploy* script reads it, so it is inert on the live path - but a stray gitpush.sh run in that repo would rewrite a TRACKED file. Owner is pm-mba-l-cc. Flagged by bin-lezama-cc rather than edited, correctly.