parse_speed__autoslice_poc.runbook.md

AutoSlice POC Runbook — Performance: Parse Speed

This runbook describes the fully/mostly automated flow for a single slice in one conversation.

Step 0 — Prep

• Confirm toggles in 90_prompts/agent_autoslice_poc.md.
• Decide libraries in scope (default: all four).

Step 1 — Kickoff (choose mode)

• Fully automated: Paste the Kickoff Prompt from 90_prompts/agent_autoslice_poc.md.
• Hybrid deep-research: Paste the Kickoff Prompt from 90_prompts/agent_autoslice_hybrid.md; the model will prompt for up to 4 sources, then proceed automatically.

Wait for the model to return four artifacts:

• evidence_csv (rows in EvidenceLog schema)
• matrix_row_csv (one row)
• draft_md (Markdown section)
• redteam_md (Markdown checklist)

Step 2 — Apply outputs to repo

• Append evidence_csv to 02_sources/EvidenceLog.csv (append-only; normalize IDs and dates if needed).
• Update/add the Performance: Parse Speed row in 03_feature_matrix/FeatureMatrix.csv with the returned matrix_row_csv.
• Save draft_md to 05_drafts/parse_speed.md.
• Save redteam_md to 05_drafts/parse_speed_redteam.md.

Step 3 — Commit (if auto_commit disabled)

• Evidence: Evidence: Parse Speed — add EVID rows
• Matrix: Matrix: add Parse Speed row
• Draft: Draft: Parse Speed section
• Red-Team: Red-Team: Parse Speed

Gates

• Optional: run quick hygiene check before updating the matrix.
• Freeze the row if satisfied; otherwise, iterate once.

Notes

• Keep published benches’ dataset/hardware/method in notes; avoid mixing apples-to-oranges.