Find the blocker. Get back to work.
Choose the symptom. Inspect its evidence. Take the next useful step.
Check the project first
phasedrift doctor
phasedrift statusConfirm the project, tracker URL, running release, and latest import. Then open the symptom that matches what you see.
I don't see recent commits
Collection needs a running tracker. Check Settings for new-commit collection, its profile, and queued or failed jobs. Structural collection doesn't rerun every full analyzer on each commit.
If the tracker is stopped, run phasedrift init from the project to reuse its connection and data.
I ran a full scan, but the historical chart is still missing checks
A full working-copy scan measures the code that's on disk now. It doesn't fill every old commit with compiler, lint, or React results. Historical full analysis is separate and requires the correct revision environment.
Inspect the scan result in Code health. Missing prerequisites and partial analyzers have their own explanation.
I see zero unused items
Open the analyzer result. Confirm that the check actually ran, what population it scanned, and whether its result is complete. Zero measured findings and an unavailable check are different states.
I don't see a new agent report
Run phasedrift session status. Agents need to follow the installed reporting instructions or use an explicit adapter. Without native identity, task entry returns an attempt handle. Use its exact closeout command. Check that the current conversation loaded the refreshed project instructions.
A failed submission may be available for retry:
phasedrift session retry <submission-id>Retry replays the frozen report; it does not rebuild it from newer task text. Keep the returned submission ID.
Reports are present, but not classified
Reports and model classifications are separate. Classification is optional and depends on the configured analysis policy and provider. Source-linked suggestions can still create local proposals without it.
The trial has no before-and-after number
Check its metric, comparison window, model filter, scope, and contributing turns. A baseline needs comparable recorded evidence. Explicit host reports alone don't contain native runtime or usage.
Custom metrics and check-recovery plans can collect real check receipts without a computed effort comparison. Inspect those checks rather than expecting an unsupported score.
The tracker is on an old release
Install the new application build, then run:
phasedrift upgrade --repo /path/to/your/projectManaged upgrades verify ownership, back up the database, and preserve the data directory. The same managed upgrade updates an installed login service. Then run phasedrift session setup --upgrade --hooks and reload project instructions.
A scan is slow
Open Code health or Settings to see the current job. The first full baseline, dependency analyzers, and historical dependency preparation do more work than a structural scan. Inspect progress and the exact blocker before repeating the scan.
Inspect host coverage → · Read the evidence →
I have graphs but don't know what to do next
Open Agent work → Reports. Ask the agent to find one repeated obstacle, cite its original examples and propose a small fix. Choose that fix, then track it in Improvements with a matching check and quality condition. Use the first-change prompts.
If there are no relevant reports, finish one normal task first. More graphs are not a substitute for evidence about a specific problem.