Automattic / Automattic/markdown-database-integration
Generated wiki posts can disappear after primary-mode reload
- Dominant language
- PHP
- Stars
- 5
- Forks
- 1
- Avg merge
- 2h 38m
- Merged PRs (30d)
- 155
Description
## Summary
On `intelligence-chubes4`, Data Machine/Intelligence generated hundreds of `wiki` posts under `wordpress-com` during live ingestion, but after an Intelligence deploy/reload the WordPress.com corpus collapsed back to the markdown-backed seed tree.
## Observed
Before deploy/reload:
- `wp post list --post_type=wiki --post_status=publish --format=count` reported `486`.
- `wp intelligence wiki eval readiness --root=wordpress-com --sample=100` was green.
- Newly generated WordPress.com pages included high-ID posts like:
- `Dotcom Updates — 2022-12-06` (`6971`)
- `Avalancha Team Updates - June 6th` (`6975`)
- `The Ankler launch` (`6974`)
- Gutenberg rollout dated pages (`6755`, `6758`, `6763`)
After deploying Intelligence and re-running WP-CLI:
- `wp post list --post_type=wiki --post_status=publish --format=count` returned `111`.
- `wp intelligence wiki status wordpress-com --format=json` reported only `12` pages under `wordpress-com` and readiness false.
- `wp-content/markdown/wiki/wordpress-com/` contains only the seed tree, not the generated pages.
- Searching `wp-content/markdown/wiki` found no generated files for `gutenberg-rollout`, `dotcom-updates`, or `ankler`.
## Why this matters
`intelligence-chubes4` runs Markdown Database Integration in primary mode. If generated wiki posts are not mirrored durably to markdown (or are otherwise lost on markdown-index reload), Data Machine/Intelligence ingestion can appear successful in SQLite but disappear after a deploy/reload.
## Expected
Generated `wiki` posts written through normal WordPress APIs should persist durably across primary-mode reloads, either by writing markdown files or by having an explicit supported path for generated/transient content.
## Notes
The new Intelligence operator status report in Automattic/intelligence#482 surfaced the issue by reporting only the durable 12-page `wordpress-com` tree after deploy. Debug log entries show HTML-to-block conversion warnings for generated pages around the time they were created, but the posts were visible/readable before reload.
Contributor guide
No contributing guide indexed for this repository
Assessment
This issue has not been assessed yet.