Summary

By matchday 1 of a tournament, every group already shows "Grupo bloqueado — ya arrancó el primer partido" on the Predict page, which makes it impossible to refine 1°/2°/3° picks as the rest of the group plays out — even though most of those fixtures haven't kicked off yet.

This PR loosens the group-level cutoff to the latest per-match-safe point: a group's placement card stays editable while any of its group-stage matches is still unlocked, and only locks once every such match has kicked off or been manually closed by an admin. Placement still locks before any match has finished, so it stays a genuine prediction — but the user can keep refining picks throughout matchdays 1 and 2.

Per-match locking semantics are unchanged: each individual match still locks at its own kickoff or when admin closes it. Only the derived group-level lock changes.

What changed

• api/locking.cjs — isGroupPredictionLocked now requires every group-stage match in the group to be locked before reporting the group as locked; comment + module header updated to describe the new cutoff
• i18n: predict.groupLocked rewritten in en / es / pt / it / nl to read "last group match kicked off" instead of "first match kicked off" (Spanish: "ya arrancó el último partido del grupo")
• docs/ARCHITECTURE.md §7.5 — locking rules now describe the per-match-safe cutoff and the consequence for groups containing matchDate = null matches
• docs/QA_CHECKLIST.md §11.1 / §11.2 — manual lock test now explicitly exercises a partially-started group vs. a fully-started group, and the admin closure step calls out that closing one group-stage match no longer locks the whole group on its own
• tests/locking.test.mjs — three new unit cases:
  • partially-started group stays open
  • group locks once every group-stage match has kicked off
  • closing only one of multiple admin-closeable matches does NOT lock the group

Integrity note

Under the new rule, by the time a group's placement card locks, 5 of the 6 fixtures in a 4-team group have already been played. The final 6th match is still uncertain, so placement remains a prediction. This was the user's explicit preference over the more lenient "lock after all matches finish" option (which would have trivialized the placement scoring).

Test plan

• npm run lint — clean
• node --test tests/locking.test.mjs — 11/11 pass, including the 3 new cases
• npm test — full suite 50/50 pass (verified pre-cherry-pick on the parent branch)
• Manual: reload Predict on the seeded WC2026, confirm group cards no longer show locked while at least one fixture is in the future, and that the pill copy now reads "last group match kicked off" in each language

Made with Cursor