grouped-source-intake-backfill

name	grouped-source-intake-backfill
description	Backfill grouped source-intake batches into durable notes with per-channel compaction, sourced provenance, and checkpoint-safe writeback.

Grouped Source Intake Backfill

Use this skill when draining a batch of source records from vadimgest/intake-style exports into Obsidian or another durable note store.

When to use

• Multiple records arrive grouped by source/channel/thread.
• The batch contains a mix of signal and noise.
• A note must be written back with source provenance.
• A checkpoint or cursor must be advanced only after writeback is complete.

Workflow

1. Compact by group first.

   • Treat each channel/thread/group as the unit of analysis.
   • Deduplicate repeated facts and repeated messages inside the group before writing.
   • Do not create one note line per raw row when several rows belong to one conversation.

2. Answer the four intake questions for each group.

   • Q1: What needs to be done?
   • Q2: How can I help now?
   • Q3: What facts changed?
   • Q4: What patterns/signals changed?

3. Choose the durable home.

   • If the target entity is clear, write/update the entity note with sourced provenance.
   • If routing is uncertain or cross-entity, write a sourced Inbox note.
   • If the group is noise or empty, skip it and record the skip reason in the batch summary.

4. Prefer append-style log entries for people and entities.

   • Add a dated log section with src, mentions, summary, and facts-touched.
   • Keep state facts compact and sourced.
   • Avoid overwriting unrelated sections when a small append is enough.

5. Verify before cursor advancement.

   • Re-read the destination note after editing.
   • Run the exact checkpoint / commit snippet only after triage and writeback are complete.
   • Commit only the listed source lines for the batch.

Pitfalls

• Do not split one conversation across many note entries when a single grouped summary is sufficient.
• Do not commit a source line that was not fully processed.
• Do not treat headers / empty threads as substantive content.
• When editing an existing note, read enough of the file to avoid truncation surprises before patching or replacing.
• If the batch is mostly noise, say so plainly rather than inflating it into false signal.

Verification

• Batch note exists and includes source line ranges or representative URIs.
• Entity notes, if updated, contain sourced log entries rather than bare claims.
• Checkpoint advancement matches the processed source lines exactly.

References

• See references/whatsapp-edge-backfill-example.md for a concrete batch example and writeback pattern.