Department & care

Delivering, checking and publishing the schedule

Local preview, two test suites, double check of half-days and names, then publication through a branch and a pull request, and a check that it actually works online.

Claude CodeCoworkGitHub15 min per version

Deliver and publish: four-step circuit, verification report, before / after changes.
Deliver and publish: four-step circuit, verification report, before / after changes.

Screenshot of a real tool · fictitious data, names blurred

The problem

A new version of the schedule can erase data, break a tab or display old names on the team’s workstations. And the publication procedure, done once a month, is forgotten from one time to the next.

Step-by-step method

  1. Start again from the online source, not from a local copy: archive it as is, with the current state of the data.
  2. Always deliver four things: a local preview to open right away, the changes on a dedicated branch (ready for a pull request), the archived source, and any emails, drafted in the conversation.
  3. Two test suites before any publication, both green: one suite on the data logic (merging by identity, odd/even-week attendance, migrations, the “also present” row), then the triple test in a driven browser:
    • no script errors, all modules loaded, with and without real data;
    • the key views: the month view shows all staff members, absences are displayed;
    • the real gestures: importing an absence email, drag-and-drop in a tab, assigning a stand-in.
  4. Double check of half-days: compare the template schedule with each activity tab, and list the discrepancies (someone scheduled for echocardiography in the template but missing from the echocardiography tab). Nothing is corrected without approval.
  5. Double check of names against the directory: spelling, identifiers, duplicates, people not found flagged as such.
  6. Bundle and ask for approval: nothing is published without your explicit approval. The session’s fixes are bundled into a single version, rather than one publication per fix.
  7. Publish through a branch, then a pull request: Claude pushes the version to a branch and opens a pull request summarising the changes, the results of both test suites and the discrepancies. The person in charge reviews and merges it themselves; going live follows automatically. There is no longer any direct manual upload to the live version.
  8. Check online, because deployed does not mean tested: hard reload (Ctrl + F5), version number in the title, new buttons present, and the fixed gesture repeated for real. No link is passed on without having been opened; otherwise it is marked “not verified”. A “not connected” banner goes away with the team password.
  9. Announce: a short email to the secretariat and to the mailbox of the activity concerned, with the others in copy, drafted in the conversation and sent by you.

Deliverable

A tested version, listed discrepancies, a verified publication, and a ready announcement.

Lessons learned

  • A frozen copy of the data inside the published file makes old names reappear on new workstations: remove it, and flag any workstation that is not connected.
  • Dropping a person onto an occupied cell can create a duplicate: test this gesture with every version.
  • A publication procedure must be written for someone who has never done it, including what they should see at the end.
  • Beyond the capacity of a cell, names “disappeared”: an “also present” row now shows everyone who overflows.
  • Two people editing the same cell at the same time: merge by identity (who is assigned), never by position in the list, otherwise one wipes out the other.
  • The schedule never feeds anyone’s personal calendar: they are two separate sources.

Starter prompt

Here is the schedule’s repository. Apply the requested fix on a branch, run both test suites, then deliver the local preview, the list of half-day and name discrepancies, and a pull request ready for review. Publish nothing without my approval.

← Back to the triptych