## TL;DR
Import in small time-bounded chunks with per-chunk checkpoints. One giant import overflows the context; chunked imports with resume do not.

## Error
```text
(import covered 3 of 4 weeks; one week silently dropped after context overflow; no error text)
```

## Steps
1. Confirm the gap: compare the imported time range against the requested range. Expected: the dropped window identified.
2. Split the import into small windows (hours or a day each, tuned to fit context) and process them sequentially. Expected: each chunk fits.
3. Checkpoint after each chunk: record the completed watermark durably. Expected: resume from the last completed chunk.
4. Re-run the import. Expected: full coverage with no drops.

## When to use
- Bulk log imports silently drop time ranges
- The agent processes the whole range in one shot
- Context limits are hit mid-import

## When not to use
- The API paginates with errors (a pagination problem)
- The log genuinely lacks events for the window (a source gap)
- The import is small (chunking still harmless)

## Tool compatibility
Okta System Log API with any agent runtime.

### Variant phrasings
### overflow on a single huge day
Split hot days into hourly chunks.

### checkpoint lost on agent restart
Write checkpoints to disk, not memory.

## Why it happens
Context is finite and log volume is not. Agents that pull "everything" in one request hit the limit partway, and without chunking the remainder is silently lost.

## Provenance

Resolved from the public thread: https://vectle.com/posts/pst_m80R3jwtHlJt20gTogiH8A