vrr/codeburn
2
0
Fork 0
mirror of https://github.com/AgentSeal/codeburn.git synced 2026-10-21 05:23:19 +00:00
Code Issues Projects Releases Packages Wiki Activity Actions 4
codeburn/docs/drill-through.md

4.3 KiB
Raw Blame History

Drill-through: from aggregates to the sessions behind them

CodeBurn Desktop can open any aggregate — a day in the Overview chart, an expensive session, a project, a model, a task category, a pull request — as a session list that explains exactly what composes it, without losing your place.

Using it

  • Click a day bar in Overview's Daily spend chart. Sessions opens with a Day chip and shows the sessions that were active that day — including sessions started earlier that were still working on it, within the source's day granularity.
  • Click a session in Overview's "Most expensive sessions" to land on the sessions list with the drawer already open on that exact session (matched by provider + id, so a duplicate id or title under another provider never opens the wrong one).
  • Click a model name (Overview's models table or the Models report), a task category (Overview's Top activities or the Models by-task rows), a project (Spend → By project → expand → "View sessions"), or a pull request (Pull requests → expand → "View sessions").
  • While an investigation is active, clicking another aggregate adds to it: same-dimension values union (two days, two models), different dimensions intersect (day AND category).

Reading the numbers

The destination's summary always separates two figures:

  • in selection — the spend the listed sessions contribute to the pressed value. A session that spent $100 overall but only $20 in the selected category contributes $20 to the selection sum.
  • full cost of these sessions — the whole cost of the listed sessions, labeled separately so the two can never be confused. Rows show their contribution first ($20 of $100), and the side drawer shows both next to each other.

Pull requests keep the by-PR report's attribution semantics: turns split across a multi-PR set contribute their share to each, so PR rows are not an exclusive partition. A session's spend that is tied to no selected PR is labeled "Not tied to a specific PR". When older transcripts cannot be attributed per turn, the rows carry the by-PR report's ~ even-split marker.

Sessions that cannot be attributed to the selection (no per-turn detail survived for them) are excluded and counted in the summary ("N sessions could not be attributed") instead of being silently dropped or zero-valued. The same disclosure applies to ambiguous session identities and old cached reports without per-model usage when a model filter is active.

Days follow each call's timestamp, so a turn crossing midnight contributes to both days. Model filters use the model's actual calls and tokens, independently of its cost; free models retain their usage. Supplementary accounting retains cost and tokens without adding requests.

Drawer, history, and focus

  • Clicking a session row opens the side drawer: metadata, full totals next to the in-selection figures, and the per-model / per-category / per-branch / per-day / per-PR breakdowns. PR URLs open outside the app. Escape closes the drawer and focus returns to the row that opened it.
  • The top bar's ‹ / › (or ⌘[ / ⌘] on macOS, Alt+← / Alt+→ elsewhere) walk the in-app history: filters, sort, list depth, and the open drawer are all restored, and Back returns you to the exact chart you left.
  • A data refresh keeps the drawer open while its session still exists in the refreshed population; if the session is gone (or the provider/project scope changed), the drawer closes rather than showing a session that no longer reconciles.
  • Reloading the app restores the last position (section, selection, drawer); the restored drawer revalidates against the fresh report.

Data path

The renderer stays a pure view over CLI JSON. The drill-through report is codeburn sessions --format json --contributions — the same session rows and the same --project/--exclude/--provider filtering semantics as the plain report, plus per-session contributions.segments (call-day slices with category, git branch, model costs and usage, and active PR set) and canonical identities (projectId, provider, sessionId). Transcripts never cross the IPC boundary; the report is fetched only while a selection is active, and every chip combination is computed client-side over the full fetched population.