Reading a run — what was ingested, skipped, quarantined

A run cannot look clean while it lost rows. Every row read is written, skipped, or set aside with its reason: IN = OUT + SKIPPED.

The counters

$ lm run get 8ffcd5ae
ID        CONNECTOR  ENDPOINT  STATUS     IN  OUT  SKIPPED  DEAD-LETTERS  STARTED           ENDED
8ffcd5ae  orders     default   Completed  10  8    2        2             2026-10-19 01:47  2026-10-19 01:47
⚠ 2 rows quarantined — see: lm dlq list --run 8ffcd5ae-c387-45b9-8a57-091043af6c59

10 = 8 + 2, and both skipped rows are dead letters.

Counter	Counts
`IN`	the rows read from the source
`OUT`	the source rows delivered to the table. A row already there, unchanged, counts too. A row `split-row` cut in three still counts one
`SKIPPED`	the rows not written: filtered out on purpose, or set aside
`DEAD-LETTERS`	the part of `SKIPPED` set aside with its reason, for you to look at

Not written	Reasons	A dead letter?
on purpose	a `filter-where` or `skip-empty-rows` dropped it; `parser.short_rows: skip-leading` skipped a line before the table	no
set aside	`MalformedSource` (the file is broken), `MappingFailed` (a hook could not read a value), `ValidationFailed` (`validate-row`), `WriteFailed`	yes

The lines under the table

lm run get adds a line for each of these, on stderr:

Line	Means	Do this
`⚠ N rows quarantined — see: lm dlq list --run …`	rows set aside	Diagnosing a failed run
`⚠ N cell(s) not delivered as declared (a hook, or a promoted value of the wrong type) — the rows were still written`	a value a hook could not read was emptied, or left as it was; or a promoted value did not read as its type and was emptied	cell_errors
`• N row(s) removed — the source no longer holds them`	a one-file `csv-file` connector followed its file	Full refresh
`• N file(s) unchanged since they were last read — not read again`	the file did not change: `IN 0` is normal	to read it again: `lm endpoint reset-cursor <connector>/<endpoint>`
`• N file(s) read but not archived — still in the source folder …`	the archive step failed; the rows are in	check the archive folder's rights
`• script hooks:… sha256 …`	the Kotlin script version the run used	Kotlin hooks
`✖ run failed: …`, `• run partial: …`, `• run cancelled: …`	why the run ended so	read the reason; it names the gesture

A full refresh that would remove too much fails, and removes nothing:

$ lm run get a569be52
ID        CONNECTOR  ENDPOINT  STATUS  IN  OUT  SKIPPED  DEAD-LETTERS  STARTED           ENDED
a569be52  orders     default   Failed  5   5    0        0             2026-10-19 01:47  2026-10-19 01:47
✖ run failed: 3 of 8 rows would be removed: the file holds 5 where the table held 8, more than refresh.max_removed (30%) allows — nothing was removed. If the file really lost them, raise refresh.max_removed (or set it to off) and run again.

From a count to the row

lm dlq list --run <id> lists the dead letters; lm dlq get <id> shows one, with the row exactly as it was read. Fix the source, run again, then close the entry. The walk, step by step: Diagnosing a failed run.

In your monitoring

/q/metrics counts runs by status and dead letters by reason: Observability.