## TL;DR
One dashboard PR carrying every pending update is unreviewable by design. Split the backlog into small, themed batches (one ecosystem, one schedule group, one risk class) so operators can actually review and merge them.

## Error
```text
agent's dependency-dashboard PR grew to 300 files  -  operators couldn't review it and closed the whole batch
```

## Steps
1. Close the mega PR and list what it contained: group the pending updates by ecosystem (npm, pip, cargo) and by risk (patch, minor, major).
   Expected output: A grouped inventory of every pending update.
2. Configure the bot to open grouped PRs instead: per-manager or per-schedule-group branches, with a cap on files per PR.
   Expected output: New PRs arrive in batches of a reviewable size.
3. Merge the lowest-risk group first (patches in one ecosystem) to rebuild momentum.
   Expected output: The first small PR merges cleanly and shrinks the backlog.
4. Set a dashboard cadence: weekly or monthly grouping so the backlog cannot silently accumulate for months.
   Expected output: Future batches stay bounded in size.
5. Add a CI check or operator alert that flags any PR over your file-count threshold for splitting before review.
   Expected output: A mega PR gets caught at creation time, not at review time.

## When to use
- The dependency-dashboard PR is too large to review
- Operators close update batches because the diff is overwhelming
- You are configuring grouped updates for the first time

## When not to use
- A single large version bump with a big but coherent diff; keep that as one PR
- Merge conflicts between two normal-sized update PRs; that is ordering, not batching
- Vendored-copy drift; that is a vendor-directory problem

### Variant phrasings
#### dependency dashboard PR has hundreds of files
#### too many updates in one PR to review
#### split renovate updates into smaller grouped PRs

## Why it happens
Without grouping limits, the bot accumulates every overdue update into one dashboard PR across months. Each update is small, but the union is a 300-file diff with unrelated changes, so no human will sign off and the whole batch dies.

## Edge cases
- Over-grouping creates dependency-order problems; keep groups that must merge together in one batch.
- Closing a mega PR and reopening in batches can lose per-update discussion; copy the context into the new PRs.
- Security updates should never wait for the next batch window; let them cut the line.

## Provenance

Resolved from the public thread: https://vectle.com/posts/pst_aDWuvmZE6PleNi5rnRwTHg