Reading a session's detail panel
Click any row in Sessions and a detail panel opens showing everything ATTRIBUT knows about that session: its cost and token economics, the value-model breakdown (when the session has been judged), what work it's attributed to, and, for sessions where the coding tool reports that level of detail, which tools and subagents it used.
The panel opens docked on the right on wide screens, or as a slide-over on narrower ones. At the top it shows the session's title (or an "Untitled session" placeholder), its ID, the local start time and duration, and the coding tool plus model(s) used — a session that ran more than one model (for example an Opus lead with a Haiku subagent) lists the primary model first, followed by the others.
Economics
The Economics section is a grid of eight stats covering tokens, spend, and activity for the session:
Tokens | Total token count for the session. |
Cache | Combined cache-read and cache-creation tokens, shown in compact form (e.g. "820k" or "1.2M"). |
AI spend | What the session actually cost under your declared billing plan, or the API list-price equivalent if you haven't declared a plan. |
Value | The human-equivalent dollar value of the work done in the session. |
Leverage | Value divided by AI spend — how many times over the session paid for itself. Shows "—" when there's no value or no spend to compare. |
Output | Lines of code changed, shown as "+added −removed" (lines added vs. lines removed — not a net delta). Shows "—" when the session has no line data. |
Turns | Number of conversation turns in the session. |
Tool calls | Total number of tool invocations in the session. |
Turns and Tool calls only appear for sessions tracked through the Claude Code hook integration; sessions logged another way show "—" for both.
Value model
When a session has been run through ATTRIBUT's value model, a Value model section appears beneath Economics. It shows:
A work-type badge — Creation, Reduction, Repair, Understanding, or Orchestration — naming the dominant kind of work in the session.
A confidence percentage for the estimate. When confidence is low, the panel flags that you should lean on the range rather than the single point estimate.
The point estimate itself in EEH (effective expert-hours) — the value model's native unit, e.g. "6.4 EEH" — labeled with the method version that produced it (e.g. "structural_v1").
A dimension-mix bar: a horizontal bar split proportionally across whichever work types contributed, with a legend below listing each type and its weight as a percentage.
Sessions that haven't been judged yet skip this section entirely — their Value figure in Economics still shows, but from a simpler lines-of-code estimate instead.
Attribution
The Attribution section heading names the confidence level: Linked, Inferred, or Unlinked.
Linked — the session resolved to one or more pull requests. Each is listed with its repo, PR number, and title, linking out to the PR on GitHub. Underneath, the underlying commit (short SHA and subject line) is shown as its own link when known.
Inferred — a commit resolved to a repo, but no pull request has landed yet. The panel shows the repo, the commit SHA (linking to GitHub), and the commit's subject line.
Unlinked — no repo or commit match was found. The panel simply reads "No linked work."
Tool use
For sessions that logged individual tool calls, a Tool use section lists every tool the session invoked — for example Edit, Bash, or Read — each shown as a chip with the number of times it was called.
Subagents
If the session spawned any subagents, a Subagents section appears, titled with the total count. Each subagent is listed with its agent type (or model name, if no type was recorded), followed by its model, total tokens, number of tool calls, and duration. When the subagent's cost is known, it's shown alongside in dollars.
Tool use and Subagents both depend on the coding tool reporting this level of detail — sessions logged through a tool that doesn't report it simply omit these sections.