grouped-source-batch-writeback

name	grouped-source-batch-writeback
description	Turn grouped vadimgest/intake source batches into durable Obsidian notes with provenance, deduplication, and checkpoint-safe writeback.

Grouped Source Batch Writeback

Use this skill when a pre-batched intake dump arrives grouped by source/channel/thread and you need to compact it into durable notes without over-writing one raw row at a time.

Core idea

Treat the batch as a grouped evidence window, not a flat list. Compact each group first, then decide whether the signal belongs in:

• a clear entity note,
• an Inbox/backfill note,
• or a skip record if the row is pure noise.

Procedure

1. Read the batch structure first.

   • Identify source, thread/group keys, line ranges, and any exact checkpoint commit lines.
   • Honor the batch's scope/exclusions.

2. Compact each group before writing.

   • Deduplicate repeated facts/messages inside the group.
   • Collapse multiple raw rows from the same conversation into one summary.
   • Keep representative vadimgest://... URIs or source line ranges.

3. Answer the four intake questions for each meaningful group.

   • What needs to be done?
   • How can I help now?
   • What facts changed?
   • What patterns/signals changed?

4. Route by certainty.

   • If the target entity is clear, update that note with sourced facts.
   • If routing is uncertain or cross-entity, write one sourced Inbox note for the batch.
   • If the record is pure noise, record it as skipped with reason/hint.

5. Write durable notes with provenance.

   • Inbox summaries should include line ranges or representative URIs per group.
   • Entity notes should use compact state/log facts with source citations.
   • Avoid one note/log entry per raw row when the rows belong to the same conversation.

6. Commit the checkpoint only after triage/writeback.

   • Run the exact checkpoint snippet supplied by the batch.
   • Commit only the listed source lines.
   • Do not advance the checkpoint for any source you could not fully process.

Pitfalls

• Do not treat newsletters, alerts, or other low-signal records as meaningful facts unless they change a real state.
• Do not skip compaction just because the batch is small.
• Do not write raw-message-per-row notes; that creates noise and loses the group-level story.
• Do not invent extra checkpoint updates beyond the exact batch instructions.

Verification

• Every written note should point back to source URIs or line ranges.
• The final report should state which note paths were written and whether the exact checkpoint snippet ran.

Reference material

• See references/batch-intake-patterns.md for a reusable compacting and writeback template.