fix(cluj): re-add seq to static trip_id to disambiguate duplicate CTP CSV depTimes

[ciotlosm]

fix(cluj): re-add seq to static trip_id to disambiguate duplicate CTP CSV depTimes

What broke

[release] daily multi-feed pipeline to R2 has been failing every day since 2026-08-22 with:

[validate] cluj-napoca.gtfs.zip: 1 error(s), 1 warning(s)
  ERROR   stop_times: non-monotonic stop_sequence for trip 63_0_LV_2255 (11 -> 0)
  WARN    13589 trips have no stop_times entries

Root cause: CTP's orar_M21_lv.csv (route 63 LV) has 22:55 on TWO rows in dir0:

75: 22:35,21:20
76: 22:55,21:40   <- dir0 has 22:55 here
77: 22:55,22:00   <- dir0 has 22:55 again
78: ,22:15

The CSV parser doesn't dedupe. The trip-emission loop calls makeTripId(routeId, dir, serviceId, depTime) for each row and produces the same id 63_0_LV_2255 twice. The second emission appends a fresh 12-row stop_times block to a trip that already has 12, so stop_sequence goes 0,1,2,...,11,0 - validate catches the jump and the whole pipeline dies.

Why the prior 4-part format was wrong

A previous "shorten trip_ids" cleanup reduced the static format from

<route>_<dir>_<serviceId>_<seq>_<HHMM>     (old, 5 parts)

to

<route>_<dir>_<serviceId>_<HHMM>          (current, 4 parts)

on the assumption the seq "was never consumed by anyone." Wrong: it absorbs the (rare) CTP CSV case where the same departure time appears on two rows in the same direction. The seq is load-bearing.

The fix

Reinstate the per-departures index in the trip_id:

<route>_<dir>_<serviceId>_<seq>_<HHMM>     (new, 5 parts, seq re-added)

i is already in scope at the call site. Two 22:55 rows on M21 LV dir0 become <route>_0_LV_1_2255 and <route>_0_LV_2_2255, each with its own monotonic stop_times block.

Verified locally with a mini fixture (3-stop Tranzy pattern + CSV with the same minute twice in dir0): output trips.txt has two distinct trip_ids, each with stop_sequence 0,1,2.

What this does NOT change

• HHMM is still last. neary's parseLiveStartMin regex fallback (_(\d{3,4})$) and the verify-trip-id-format.ts self-check (_\d{4}$) both keep working unchanged.
• Frequency anchors keep their <route>_<dir>_<serviceId>_FREQ_<HHMM> shape (built directly in derive/frequencies.ts, not via makeTripId). They don't need a seq because the HHMM is a window-start and windows are unique per direction.
• The cluj RT quirk in src/rt/cluj.ts parses the LIVE RT feed's trip_ids, not static - different format, different concern, no touch.
• The neary reconciler matches by (route, direction, time) and never compares trip_ids by string equality, so the additional seq slot is invisible to the JOIN.

Files

• adapters/cluj-napoca/src/assemble/emit/trips.ts - makeTripId gains a seq parameter; the trip-emission loop passes i; full JSDoc rewritten to document the CTP-CSV-duplicate-time case and what the seq must NOT break.
• adapters/cluj-napoca/tests/reconcile.test.ts - existing trip_id regex bumped from 4-part to 5-part (and the _FREQ slot is now optional in the seq position, not the HHMM position); the existing 35_0_LV_0600 reference updated to 35_0_LV_0_0600; one new regression test that feeds a CSV with a duplicate depTime and asserts two distinct trip_ids, each with a monotonic 0,1,2 stop_times block.

Test

• pnpm test -> 209/209 pass (was 208/208 + 1 new regression test)
• pnpm check (tsc strict on src + test) -> clean
• pnpm build -> clean

What happens after merge

Next 00:30 UTC cron run will pick up the fix, validate passes, the publish step lands fresh cluj-napoca SQLite + zip on R2, and the live n3ary.com app pulls the new data on its next SW refresh (the existing Aug 21 R2 data is still serving fine, so this is a clean cutover, not a recovery).