implementation-report records 0 tokens for some implementations (impl 2 and 3 of the #568 batch) #569

New issue

Open

opened 2026-09-14 12:08:10 +00:00 by jared · 0 comments

jared commented

2026-09-14 12:08:10 +00:00

Owner

Context

Per-ticket stats for the responsibility-refactor batch #568, read from .sdlc/pipeline.db on 2026-09-14:

ticket	implementation_id	dispatches	tokens recorded
#564	1	14	13.55M
#565	2	11	0
#566	3	20	0
#567	4	32	29.74M

Implementations 2 and 3 have zero in every token column while 1 and 4 recorded fine. The agents.model_id + token columns landed in schema 23 (#553, ADR-0175).

Expected

Every dispatch row carries in/out/cache_read/cache_creation tokens once its agent stops.

Reproduce

sqlite3 .sdlc/pipeline.db "select implementation_id, sum(tokens_in), sum(tokens_out) from dispatches group by implementation_id" (adjust column names to the schema).

Suspects

Usage stamping happens at implementation-report time (#553: "stamp usage on the session's latest implementation"). If the stamp ran only for the session's latest implementation, sessions that ran two tickets back-to-back would leave the earlier one unstamped. Check whether 565 and 566 shared a session with a later ticket.

Origin

Verdict: keep. Root, not symptom: the stats are the whole point of #553 and two of four batch tickets have none; a user asked for per-ticket stats today and could not get them.
Session: fedf1982-c8e2-4f03-9f41-bc1d84fdd823

## Context Per-ticket stats for the responsibility-refactor batch #568, read from .sdlc/pipeline.db on 2026-09-14: | ticket | implementation_id | dispatches | tokens recorded | | --- | --- | --- | --- | | #564 | 1 | 14 | 13.55M | | #565 | 2 | 11 | 0 | | #566 | 3 | 20 | 0 | | #567 | 4 | 32 | 29.74M | Implementations 2 and 3 have zero in every token column while 1 and 4 recorded fine. The `agents.model_id` + token columns landed in schema 23 (#553, ADR-0175). ## Expected Every dispatch row carries in/out/cache_read/cache_creation tokens once its agent stops. ## Reproduce `sqlite3 .sdlc/pipeline.db "select implementation_id, sum(tokens_in), sum(tokens_out) from dispatches group by implementation_id"` (adjust column names to the schema). ## Suspects Usage stamping happens at implementation-report time (#553: "stamp usage on the session's latest implementation"). If the stamp ran only for the session's latest implementation, sessions that ran two tickets back-to-back would leave the earlier one unstamped. Check whether 565 and 566 shared a session with a later ticket. ## Origin Verdict: keep. Root, not symptom: the stats are the whole point of #553 and two of four batch tickets have none; a user asked for per-ticket stats today and could not get them. Session: fedf1982-c8e2-4f03-9f41-bc1d84fdd823

jared added the

plugin/os-sdlc

filed-by/agent

labels

2026-09-14 12:08:10 +00:00

jared referenced this issue

2026-09-14 12:20:39 +00:00

Project::Yaml owns .sdlc/project.yaml: one path, one reader, one guarded writer (supersedes ADR-0088 in part) #567

jared referenced this issue

2026-09-14 12:25:19 +00:00

Map: responsibility-refactor batch 1 on project.rb (#564 #565 #566 #567) #568