## TL;DR
A sensor that evaluates cleanly but requests nothing is almost always a cursor or condition bug: the cursor already advanced past the new data, the condition never matches, or every tick yields SkipReason. Read the sensor tick logs and the cursor value first. Sensors do exactly what the code says; the code says skip.

```text
dagster sensor evaluated but no runs requested, sensor broken
```

## Use this when
- Sensor ticks show evaluations but zero run requests.
- New data arrives and the sensor ignores it.
- It worked before a sensor code change.

## Not for this skill when
- The sensor errors during evaluation. That is an exception in the sensor code.
- Runs are requested but fail to launch. That is a launcher problem.

## Steps
1. Open the sensor tick history and read the logs and the cursor value per tick. Note what the sensor saw. Verify: the sensor's view of the world is visible
2. Check the cursor: did it advance past the new data, or is it stuck behind? Note the cursor position versus the data. Verify: cursor behavior is understood
3. Review the run-request condition for logic errors: wrong comparison, wrong timezone, wrong field. Note the failing condition. Verify: the logic bug is found
4. Add explicit SkipReason messages so future ticks explain themselves. Note ticks now say why they skipped. Verify: silence becomes explanation
5. Reset or fix the cursor if it is wrong, then watch one tick request a run. Note the run is requested. Verify: the sensor works end to end

## Variant phrasings
### dagster sensor not triggering runs
The symptom phrasing.
### dagster sensor cursor stuck
The cursor-shaped search.

Compatibility: Dagster 1.x sensors. Tick history and cursor semantics are stable across 1.x.

## Why it happens
Sensors are pure functions of the cursor plus the external state they poll. If the function returns no RunRequests, Dagster did its job: it evaluated and found nothing to do. The bug is always in what the function observed or how it decided, most often a cursor that moved too far during testing and now sits ahead of real data. Because sensors fail silently by design (skip is a valid outcome), the tick logs are the only witness. Read them.

## Edge cases / pitfalls
- Resetting a cursor reprocesses everything since the new position. Backfill deliberately or you will launch a run storm.
- Minimum interval between ticks can delay the first run request after a fix. Wait out the interval before concluding it is still broken.
- Sensors that request runs for data that already has runs create duplicates. Make run keys idempotent.
- A sensor evaluating against a cached or stale external state skips correctly for the wrong world. Check the freshness of what it polls.

## Provenance

Resolved from the public thread: https://vectle.com/posts/pst_-Awde5Vpn7dFRMtSCLxXug
