You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
{{ message }}
Repository navigation
Mapper: Auto Match starts a run with no matching algorithm, does nothing, and records it as failed #2872
Since multi-algorithm configuration (#2307), a new project starts with no matching algorithm. Auto Match still accepts Submit with Retrieve Candidates ticked. It sends no $match calls, closes the run in under a second, and records it as failed with 0 completed and 0 failed rows. The user gets no message: the dialog's "Your project is configured to use the following match algorithms:" is just followed by an empty list. Seen in production on 2026-06-23 and on 2026-10-02 (four runs in a row on a new LOINC project).
Auto Match can also run the AI Assistant alone, on rows that already have candidates. So the dialog has to work out what a run will actually do from the project's algorithms, the row scope and its two checkboxes. It should tell the user while they set the run up, not after Submit.
Acceptance criteria
With no algorithm configured, the dialog says so where it lists the algorithms and offers a way to open Configure. Retrieve Candidates is off and disabled.
With Run AI Analysis on and Retrieve Candidates off, the dialog shows how many rows in the chosen scope the AI will analyse and how many it will skip. It analyses rows that have candidates and no analysis against those same candidates yet.
These messages update as the scope and checkboxes change.
Submit is disabled, with the reason shown, only when the run would do nothing.
A run that would do nothing opens no AutomatchRun.
An AI-only run is recorded by what the AI did. Today summarizeRunCompletion counts only algorithm stages, so an AI-only run that analysed every row is recorded as failed (0/0).
Components:
oclmap AutoMatchDialog.jsx: isDisabled, the Retrieve Candidates checkbox and the algorithm chips.
MapProject.jsx: onGetCandidatesSubmit, getRowsResults, createAutomatchRun, and runBulkAIAnalysis → fetchRecommendation, which skips rows with no recommendable_concepts.
The issue has detailed acceptance criteria and names the affected files (AutoMatchDialog.jsx, MapProject.jsx, attribution.js), but the dialog logic has to account for algorithms, row scope and two checkboxes, and the AI-only run recording needs a design choice, so complexity and scope are moderate. The cause, #2307, is closed, so nothing blocks the work, and the risk is moderate because it changes how production runs are recorded and when they are created. The type stays bug because runs are wrongly recorded as failed.
Generated by ticket-triage pipeline v3. Model: anthropic/claude-sonnet-5.5 (openrouter).
Activity
ocl-automation commentedon Oct 2, 2026
Automated Triage — 2026-10-02
Decision:
decision/ai-readyType:
type/bugSignals:
signal/well-specifiedThe issue has detailed acceptance criteria and names the affected files (AutoMatchDialog.jsx, MapProject.jsx, attribution.js), but the dialog logic has to account for algorithms, row scope and two checkboxes, and the AI-only run recording needs a design choice, so complexity and scope are moderate. The cause, #2307, is closed, so nothing blocks the work, and the risk is moderate because it changes how production runs are recorded and when they are created. The type stays bug because runs are wrongly recorded as failed.
Generated by ticket-triage pipeline v3. Model: anthropic/claude-sonnet-5.5 (openrouter).
snyaggarwal commentedon Oct 5, 2026
@paynejd raised PR OpenConceptLab/oclmap#92
OpenConceptLab/ocl_issues#2872 | Create similar concept to not copy e…
OpenConceptLab/ocl_issues#2872 | review feedbacks
Merge pull request #92 from OpenConceptLab/issues#2872
OpenConceptLab/ocl_issues#2872 | auto match should not run when algos…