mixed-source-intake-backfill

name	mixed-source-intake-backfill
description	Process mixed vadimgest-style intake batches into sourced inbox notes or entity writebacks, then advance the checkpoint safely.

Mixed-source intake backfill

Use this skill when draining a batch of evidence from Gmail, Signal, WhatsApp, LinkedIn, Dayflow, GTasks, or similar mixed-source exports into MyBrain.

Core goal

Turn a raw batch into durable notes with provenance:

• compact each channel/thread group first
• dedupe repeated facts inside the group
• decide whether the output belongs in an entity note or a sourced Inbox note
• include source line ranges or representative vadimgest://... URIs
• only then advance the checkpoint

Workflow

1. Group first, not row-by-row.

   • Treat each channel/thread as the unit of analysis.
   • Merge duplicate or repeated records inside the same group before writing anything.

2. Answer the four intake questions per group.

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

3. Route durable facts to the smallest correct home.

   • Clear person/org/deal/project fact → entity note with provenance.
   • Cross-entity / uncertain routing → one sourced Inbox note for the batch.
   • Noise / stale notifications / header-only records → summarize as skipped with reason.

4. Write sourced evidence, not just summaries.

   • Include line ranges or representative URIs in every group summary.
   • If a group contains both signal and noise, keep the signal and explicitly classify the rest.

5. Checkpoint only after writeback is complete.

   • Use the exact commit snippet provided by the batch script.
   • Do not advance any source you could not fully process.
   • Do not substitute a broad vadimgest commit for the exact per-source checkpoint.

Good batch-note shape

• Timestamped batch header
• One subsection per channel/thread cluster
• For each subsection: 4 questions, then summary bullets with source references
• Final skip summary by source

Pitfalls

• Do not write one note per raw row when rows belong to the same conversation.
• Do not bury the only durable fact under a wall of noise.
• Do not advance the checkpoint before the inbox/entity writeback is finished.
• Do not invent routing when the target entity is unclear; use the Inbox.

Verification

• Re-open the written note and confirm the new batch section is present.
• Verify the checkpoint file was updated with the exact listed source lines.
• Make sure the final report states whether the exact checkpoint snippet ran.

Reference material

• See references/batch-example.md for a condensed example of how a mixed batch was compacted and checkpointed.