Explore six days of Astra usage
Filter response counts and reviewed corrections by day and project. Repository figures cover the full window.
Recorded usage
Native counters, not unique text, billed spend, or productivity. Cached input is included in input; reasoning is included in output. Day 6 ends at the common cutoff.
Responses by study day
Responses by project
Responses and corrections by project
Select a cell for its exact counts. The optional 3D profile shows one project across six days, using a fixed scale so that project comparisons remain consistent.
Select a cell to inspect its exact counts and denominator.
Cells show retained observations. Zero does not prove no work occurred. A corrective count is not a failure rate. Use Tab and Enter to inspect matrix cells.
Contributions that needed steering
Protocol 1.1 distinguishes new requests from corrective steering using context. Counts reflect review judgments. No satisfaction or ease ratings were inferred. Read the method.
Repository activity
Always the full six-day window. The project filter applies here; the day filter does not. These are all-author repository events, with no proven Astra authorship or daily outcome allocation. One observed project identity could not be resolved.
Definitions, sources, and limits
This explorer uses only the versioned evidence file and a reproducible derived reader dataset. All projects are aliased, with relative days and no raw transcripts, task identifiers, machine paths or mappings. De-identification does not guarantee anonymity.
- Recorded usage: reconciled retained records, common cutoff, three measurement/coordination tasks excluded. This is not an account billing export.
- Coded observations: 538 substantive contributions, with typed and voice components reconciled. Native message items use a different denominator.
- Repository activity: complete net boundary diffs and timestamped events across 17 identifiable repositories. Activity is not quality or model authorship.
- Task counts and runtime unions are omitted from filtered totals because they are not additive.
Download the reviewed aggregate data · Full methods and limitations · Three.js release record · Canonical evidence JSON · Evidence methods