Last updated:
Explore Pull RequestsRelease historyThis feature was introduced by Pull requests in AI SDLC Insights on October 19, 2026.
The Pull requests tab lists changes across the repositories visible in your dataset, with their AI cost and the sessions behind them. For GitLab, the tab uses Merge requests instead. This guide uses pull request for both providers.
Connect GitHub or GitLab and select a dataset that receives its request history. Coding-harness telemetry supplies the linked sessions and their cost.
Find a pull request
Open Darkplane → Insights → Pull requests and pick a time range of at most 90 days. The list shows requests with an event in that range, described as they stood at its end. Columns include state, title, repository, number, type, AI cost, and last activity, ordered by last activity.
Use the shared filter bar to narrow the list by repository and agent activity. Request-specific conditions include assistance, type, activity, reviewed by, status, and review depth. For example, select status = awaiting review to investigate requests still waiting for review, or activity = merged in range to inspect completed work.
The search condition matches text in the title or author's login without regard to case, or matches the request number exactly. It searches the whole range, not just the loaded page. Multiple search conditions must all match, with up to 20 search terms at once.
When you select assistance = not assisted, session-based filters do not apply because those requests have no linked coding session. Some chart drilldowns also add agent filters = not applied to preserve the chart's comparison population. Check the filter chips before comparing the list to another chart.
Read the heatmap
The default Heatmap view groups requests by time on the horizontal axis and the selected metric on the vertical axis. Cell color represents the number of requests. Hover a cell to read its count and bounds.
- AI cost counts requests opened in the range that have an AI cost, grouped by their opening time and cost.
- Time to merge counts only requests merged in the range, grouped by their merge time and time to merge.
An empty AI cost heatmap does not mean the dataset has no requests. Unassisted requests have no AI cost to plot. Likewise, an empty Time to merge heatmap can mean that nothing merged in the range.
Drag a region to narrow the list to that time and value range. The selection indicators above the list show the bounds; clear the selection to return to the broader list. Use Set time range, Set ai cost filter, or Set time range & ai cost filter, depending on the region you selected, to keep the selection as filter chips and a time-picker range. For Time to merge, the button names that metric instead. Switching metrics clears the temporary selection.
The heatmap always uses its metric's population: opened requests with an AI cost, or merged requests. It ignores existing AI cost, time-to-merge, and activity bounds so you can select a new region without hiding the surrounding distribution.
Switch to Stats
Select Stats to replace the heatmap with totals for the filtered list across all its pages: the request count, AI-assisted, Not assisted, and AI cost. These totals follow the time range, filters, and search conditions.
Inspect linked sessions
Select a request row to open its side panel. You can follow the provider link to GitHub or GitLab, inspect the request's facts, and see its linked sessions, AI cost, tokens, and model cost breakdown. Open a linked session to inspect its session details and conversation.
The panel's session cost, tokens, and duration count only work on this request's branch. A session that also worked on other branches can therefore cost more on the Sessions tab. A missing session can also reflect telemetry retention, rather than a request with no AI assistance.
If Dash0 found no assistance, the panel explains that no coding-agent session recorded work on the repository and branch in the 90 days before the request opened or merged. This is a statement about recorded telemetry, not proof that nobody used AI.
Start from a chart
Select a linked chart segment on another Insights tab to open the requests it counted. The destination carries the chart's time range and relevant conditions as removable filter chips. A selected time-series span can narrow that range further.
For example, drill into a review-depth bucket on Productivity, then open an individual request to inspect the sessions behind it. This lets you check which changes account for a summary before acting on the summary alone.