fix(paperless-ngx): widen email cron to */10 to reduce double-ingest

[ClaydeCode]

Problem

Fixes #30.

On Paperless NGX, an email attachment gets ingested twice. The first ingest succeeds; the second always fails and raises a failure alert in the Paperless UI — the document actually imported fine, so users are confused by an error that means nothing.

Root cause

apps/paperless-ngx/docker-compose.yml.template overrode the mail-check cron to every minute:

- PAPERLESS_EMAIL_TASK_CRON=* * * * *

Paperless's mail-check task has no overlap guard, and it writes the message to its ProcessedMail dedup table only after the attachment is consumed (post-OCR). So a mail-check fired by the next cron tick can pass the dedup .exists() check and consume the same attachment again before the first pass recorded it. First import succeeds; the second is rejected by document-hash duplicate detection → the failure alert.

Change

One line: widen the interval to */10 * * * * — Paperless upstream's default.

Because the dedup record lands only after OCR completes, the race window can span minutes; 10 minutes gives a comfortable margin over a single fetch+OCR pass while email ingestion stays acceptably responsive. The prior every-minute value was chosen for responsiveness (per git history) but is the aggressive knob that triggers the race.

Scope note: this mitigates, not eliminates. A genuine fix needs an upstream overlap lock; and the user-configured multi-rule case (two mail rules matching one message, which upstream treats as not-a-bug) is out of the template's control.

Verification

• python3 build_store_data.py exits 0 (generated zip/metadata are untracked; only the template changed).
• No other file or doc references the cron value.
• Behavioural repro (send one attachment, watch for two consume tasks) requires a live shard and is not runnable in CI.

Review panel

Adversarial reviewer (always-on) — no blocking findings. Advisories, all addressed:

• "*/5 reduces but doesn't eliminate the race; dedup is written post-OCR so the window is minutes-long — prefer upstream default */10 for 2x margin." → Fixed: adopted */10 instead of the originally-committed */5.
• "Commit says 'stop double-ingest' — overclaim, cron tuning only mitigates." → Fixed: reworded commit + PR to "reduce/mitigate" and documented the residual race and the multi-rule case.
• "Issue says second ingest 'always' fails; a pure overlap race is intermittent — could be a second mechanism (multi-rule match / post-OCR dedup write)." → Noted: the post-OCR dedup write makes the window minutes-long, which explains near-deterministic failure under an every-minute cron; the multi-rule case is user-config and out of template scope. Called out in the scope note above.

No other specialist (test / DevEx / security / DB / API / UX) was dispatched — a one-line config value change touches none of those lenses.

---

🤖 Generated with Claude Code