Briefing — Iteration 2¶
Parent: HL-TFW-38 Goal: Design the staged review structure and resolve remaining open threads from iteration 1. Predecessor: RES iteration 1
Predecessor Decisions to Build On¶
- D1: Explicit §6-8 enumeration in handoff.md — CONFIRMED, ready for TS
- D2: Reviewer audit step with triage gate — CONFIRMED, but may be superseded by staged review
- D3: Diagram collection → index, not copy — needs concrete format
- D5: Review should have staged structure — USER CLARIFIED: domain-specific gates (requirements, code quality, security), not literal research flow
Open Threads from Iteration 1¶
- Staged review workflow — design domain-specific review stages. User clarified: like industry multi-gate patterns (requirements validation, code quality, security). NOT exact research flow stages.
- Two classes of TFW workflows — investigative (staged) vs procedural (linear). Map all workflows.
- Diagram indexing format — concrete KNOWLEDGE.md §2 format for diagram references.
User-Injected Direction (post-RES1)¶
"Better for review: requirements validation, code quality, security — something like this. When I told like research, I mean the flow or idea, but not exact research flow."
Research Plan¶
Gather¶
- Search: how do production AI code review systems structure multi-gate reviews? What stages/gates are standard?
- Analyze existing TFW REVIEW files — what "natural stages" do good reviewers already follow?
- Check review.md word budget — can staged structure fit in 1200 words?
Extract¶
- Design the concrete review stage set for TFW
- Map how the REVIEW template checklist maps to stages
- Define what the reviewer does in each stage (action, not just read)
- Propose diagram index format for KNOWLEDGE.md
Challenge¶
- Test: does staged review create stage-file overhead that breaks the review's purpose?
- Test: can the 9-point (now 10-point) checklist be decomposed into stages cleanly?
- Test: does this exceed the 1200-word workflow budget?
Hypotheses (from HL §10 + iteration 1)¶
| # | Hypothesis | HL Status |
|---|---|---|
| H1 | Explicit §6-8 enumeration stops skipping | 🟢 confirmed (iter 1) |
| H2 | Audit verification step changes reviewer behavior | 🟢 confirmed (iter 1) |
| H3 | Domain-specific review stages (not single-pass) will produce more reliable reviews | open (new, from user direction) |
Scope Intent¶
- In scope: Review stage design, REVIEW template adjustment, diagram indexing format, workflow classification
- Out of scope: Implementing the changes (that's TS/handoff), changing the research workflow, handoff.md changes (already confirmed in iter 1)
Guiding Questions¶
- Should review stages produce intermediate files (like research stage files) or stay within the single REVIEW artifact?
- Is the staged structure mandatory for ALL reviews, or only code/implementation reviews (with a triage bypass for docs)?
Stage complete: NO