Per-ticket audit trail: record model + effort + outcome ON the issue/PR #310

New issue

Closed

opened 2026-07-01 18:19:32 +00:00 by igor · 0 comments

igor commented

2026-07-01 18:19:32 +00:00

Collaborator

Goal

An audit trail on the Forgejo ticket itself — read a task and see how it went. Not (only) the cost ledger; the story should live where you read the work.

What it looks like

When the agent works an issue -> PR, the harness posts/maintains a structured run log on the PR (and/or issue) capturing each step:

who did the work and how hard: worked: sonnet@high
each review pass + verdict + effort: review: opus@xhigh -> REQUEST_CHANGES (round 1)
rework escalation: reworked: sonnet@xhigh
final disposition: review: opus@xhigh -> APPROVE, merged / escalated to Josh / blocked

So one glance at a ticket tells you the whole journey: what model + effort touched it, how many rounds, and how it ended.

Where the pieces are

Model is already known per call.
Effort lands with --effort (#308).
Outcomes (verdicts, rework rounds, merge/escalate) the review tick already tracks in .review state.

So this is mostly surfacing what we already have, as a comment on the ticket, plus effort once #308 wires it. The cost ledger stays the aggregate/spend view; this is the per-ticket provenance. Companion to #308.

## Goal An **audit trail on the Forgejo ticket itself** — read a task and see how it went. Not (only) the cost ledger; the story should live where you read the work. ## What it looks like When the agent works an issue -> PR, the harness posts/maintains a structured **run log** on the PR (and/or issue) capturing each step: - who did the work and how hard: `worked: sonnet@high` - each review pass + verdict + effort: `review: opus@xhigh -> REQUEST_CHANGES (round 1)` - rework escalation: `reworked: sonnet@xhigh` - final disposition: `review: opus@xhigh -> APPROVE`, `merged` / `escalated to Josh` / `blocked` So one glance at a ticket tells you the whole journey: what model + effort touched it, how many rounds, and how it ended. ## Where the pieces are - **Model** is already known per call. - **Effort** lands with `--effort` (#308). - **Outcomes** (verdicts, rework rounds, merge/escalate) the review tick already tracks in `.review` state. So this is mostly *surfacing* what we already have, as a comment on the ticket, plus effort once #308 wires it. The cost ledger stays the aggregate/spend view; this is the per-ticket provenance. Companion to #308.

joshtronic was assigned by igor

2026-07-01 18:19:32 +00:00

igor referenced this issue

2026-07-01 18:20:00 +00:00

RFC: escalating review ladder (Sonnet -> Opus -> Fable), escalate-on-non-approval #308

igor changed title from ~~Telemetry: track model + effort + outcome per call (tuning data for the effort ladder)~~ to Per-ticket audit trail: record model + effort + outcome ON the issue/PR

2026-07-01 18:32:41 +00:00

igor referenced this issue

2026-07-01 20:13:59 +00:00

feat(review): escalating effort ladder -- reviewer flat->max, worker climbs (Closes #308) #312

igor referenced this issue from a commit

2026-07-01 20:13:59 +00:00

feat(review): escalating effort ladder -- reviewer flat->max, worker climbs (igor#308)

igor referenced this issue from a commit

2026-07-01 20:51:29 +00:00

feat(review): audit-trail rework comment -- the ticket tells its own story (igor#310)

igor referenced this issue from a pull request that will close it,

2026-07-01 20:51:29 +00:00