Skip to content

Report

Report is the considered version of your AI-coding review. Where Realtime is in-the-moment and Daily Recap is a quick end-of-day read, a Report is a full write-up over a range you choose — the place to step back and see patterns: what’s improving, what keeps costing you, and what to change next. It lays that analysis out with charts, and it shows your own work.

Everything you need to make a report sits in the toolbar at the top:

  • Range — a dropdown of presets (Today, Yesterday, This week, Last week, This month, Last month) plus a Custom… date picker for any window.
  • Language — the language the report is written in. A report keeps the language it was generated in, so an English report stays English even if you switch the dashboard afterward.
  • Generate — builds a fresh report for the current range. It reads Queuing… while it starts, and you can Stop a run you no longer want.

The page opens to your history, not a blank form — a table of everything you’ve already run:

ColumnWhat it shows
GeneratedWhen the report was created
RangeThe window it covers
LangThe language it’s written in
Prompt ScoreThe headline score for that range
StatusWhether it’s ready, still generating, or failed
ActionsExport or open

Select rows to delete several at once, or use a row’s export menu to pull it out as Markdown, DOCX, or PDF without opening it first.

Click a row to open the report in a side panel. It builds as a series of chapters that stream in as they finish, so you can start reading the summary while the later chapters are still computing. From top to bottom:

  • A status strip flags anything incomplete — a chapter still generating, or a token snapshot that’s catching up — so you know when a number will firm up.
  • A header with the range, who it was generated for, when, its scope, and its language.
  • Cost optimization — an estimate of what you could have saved over the range, or a note when there isn’t enough priced data to say.
  • If you’re new, a Learning mode notice explains that early reports are light until you’ve built up a history of scored work.
  • The chapters themselves, in order: a short executive summary; your score and how it moved; how cleanly your sub-sessions closed; what your prompts cost to get where they got; a snapshot of the underlying figures; the work you actually did; where cost concentrated and what was recoverable; and specific recommendations for what to change next.
  • In a weekly report, the metrics-snapshot chapter also carries an activity heatmap — hour of day against day of week, shaded by token volume — so you can see when you tend to code.

Alongside the ones you generate, Prism produces reports automatically on a weekly and monthly cadence. Each opens to its own summary view with its headline tiles up top, so a recurring report reads at a glance before you dive into the detail.

From an open report you can regenerate it — after a new batch of work lands, say — or export it with the export menu, which offers Markdown, DOCX, and PDF (handy for pasting into a doc or a self-review). Reports are visible only to you.