fix: ritual horizon reaches the end of its last day, not now's time-of-day

[davidorban]

The bug

_upcoming_confirmed_occurrence_ids in core/ritual_intelligence/ritual_match.py builds its horizon as current + timedelta(days=days_until_end_of_week). That arithmetic preserves now's time-of-day on the target date rather than reaching the end of it.

On a Sunday days_until_end_of_week is 0, so the horizon collapses onto the current timestamp and every occurrence later that same day falls outside the window. Confirming a ritual on Sunday morning silently drops the Sunday evening session from prep generation. The same off-by-a-few-hours truncation clips the final day of every other weekday's window too; Sunday is just where it degenerates completely.

The lower bound is already normalised with datetime.combine(today, datetime.min.time(), ...). The upper bound was not.

The fix

Anchor the horizon to the end of the last day in the window:

horizon_end = datetime.combine(
    today + timedelta(days=days_until_end_of_week), datetime.max.time(), tzinfo=current.tzinfo
)

The Friday special case now extends days_until_end_of_week before that combine rather than building a second timestamp, so both branches go through one code path.

Verification

test_confirmed_ritual_on_sunday_includes_that_same_evening fails on current main with the exact symptom:

AssertionError: Sunday-evening occurrence was excluded by a horizon that stopped at Sunday morning

and passes with the fix. It uses relative dates so the events always fall inside the matcher's 28-day window. core/tests/test_ritual_match.py is 6 passed; the ritual-selected suite across core/tests/ is 19 passed.

Found and fixed downstream, where it shipped as davidorban/Dex#84.

🤖 Generated with Claude Code