## TL;DR

The export's columns or structure changed and the agent's parser broke. Validate the export schema on ingest, map the new format in a versioned adapter, and pin or detect the export version so the next change fails with a clear message instead of silent corruption.

## Error

```text
agent failed when expensify changed the report export format
```

## Steps

1. Diff a new export against a known-good old export. Expected: the added, removed, or renamed columns are listed.
2. Add schema validation at ingest: check for the expected columns before parsing. Expected: format changes now fail fast with a named diff.
3. Update the parser's column mapping for the new format, keeping the old mapping behind a version flag. Expected: both formats parse.
4. Detect the export version (header row, metadata) and select the mapping automatically. Expected: the agent adapts without code changes next time.
5. Add a canary check: parse one export per day and alert on schema drift. Expected: advance warning of the next format change.

## When to use

- An agent breaks right after an Expensify export or UI update.
- The failure is in parsing the export, not in calling the API.

## When not to use

- API calls fail with auth errors (credential issue).
- The export works but the data looks wrong (check filters and mappings).

## Tool compatibility

- Expensify report exports (CSV); custom agent ingest pipelines.

## Variant phrasings

### Export format changed, parser broke

Same failure, plain wording.

### Unexpected columns in expense export

The schema-drift symptom.

## Why it happens

Expensify evolves its exports without versioning guarantees. Parsers written against column positions or exact headers break the moment the layout shifts.

## Provenance

Resolved from the public thread: https://vectle.com/posts/pst_-XLG-19qe9y2p8MlH7ZT2A