title: "Extract — "What do we NOT see?"" source: "tasks/TFW-38__quality_enforcement/research/extract.md"
Extract — "What do we NOT see?"¶
Parent: HL-TFW-38 Goal: Structural analysis — why workflows cause skipping, what patterns enforce completion.
Findings¶
E1: Template-Workflow Disconnect — The Root Cause¶
The RF template (.tfw/templates/RF.md) has 8 sections: §1-5 (core) + §6 Fact Candidates + §7 Strategic Insights + §8 Diagrams.
The handoff workflow (handoff.md Phase 3, line 73) lists 6 mandatory items:
- What was done
- Test results
- Known limitations or tech debt
- Deviations from TS
- Screenshots / logs
- Observations
§6, §7, §8 are NOT in this list. There's a "Fact Candidates" mention in the informational blockquote below (lines 81-87), but it's framed as a suggestion ("As you work, capture strategic knowledge..."), not as a mandatory section enumeration.
The disconnect mechanism: 1. Agent reads handoff.md Phase 3 → sees the 6-item mandatory list 2. Agent opens RF template → sees §1-8 3. Agent fills §1-5 from the mandatory list → "Observations" maps to §5 4. Agent stops. §6-8 have no corresponding mandatory item in the workflow. 5. The template's §6-8 sections with their blockquote instructions are invisible to the agent's task execution context — the agent has already decided "what to do" from the workflow and is using the template only for "how to format."
This is the exact pattern from HL S1: "Agents treat workflows as WHAT to do, templates as HOW to format. When they disagree, workflow wins."
E2: Word Count Budget Analysis¶
| Workflow | Current words | Budget (1200 max) | Headroom |
|---|---|---|---|
| handoff.md | 825 | 1200 | 375 words |
| review.md | 816 | 1200 | 384 words |
| research/base.md | 748 | 1200 | 452 words |
| docs.md | 443 | 1200 | 757 words |
All four target workflows are well within budget. The HL risk R1 (word count exceeds 1200) is LOW probability — there's ample headroom for enforcement additions.
Estimated word cost of proposed changes: - handoff.md Phase 3 §6-8 enumeration: ~30 words - review.md Step 1.5 audit section: ~80-100 words - review.md checklist item #10: ~10 words - research/base.md Findings Map mention: ~10 words - docs.md diagram collection section: ~60-80 words - conventions.md §14 new anti-patterns: ~30-40 words
Total: ~220-270 words across 4 files. Well within budget.
E3: Positive Example Analysis — What Made HD PhaseG/H Fill §6-8?¶
Two recent HD RFs (PhaseG = 2026-04-14, PhaseH = 2026-04-14) are the only RFs with all three sections filled. Analysis:
HD-4 PhaseH RF (the best example): - §6: 5 fact candidates with Category/Candidate/Source/Confidence columns - §7: 3 strategic insights with Insight/Implication columns - §8: 1 mermaid sequence diagram (auth flow) - This RF was produced by an agent that had a conversation where the user explicitly cared about §6-8 quality (this is the same session where TFW-38 was conceived)
HD-3 PhaseG RF: - §8 Fact Candidates: 5 entries (renamed from §6 → §8 — likely template confusion) - No §7 Strategic Insights section - No diagrams section
Conclusion: The positive examples correlate with user attention to quality, not with workflow instructions. The agent filled §6-8 because the user was actively discussing quality enforcement, not because handoff.md told it to.
E4: REVIEW Template vs Workflow Gap¶
REVIEW template has 9-item checklist (#1-9). Item #9 is "Observations collected."
Missing from both template and workflow: - No item for "RF §6-8 completeness check" - No instruction to verify RF claims independently - review.md Step 1 says "Read RF thoroughly" and "Examine changed files" — but "examine" is passive, not adversarial
Existing reviewer behavior (from Gather G3): - TFW-36 REVIEW self-assessed: "I trusted executor's self-reported fix instead of verifying independently" - TFW-19 REVIEW: only one with explicit "§3. Independent Verification" section - HD-3 Full REVIEW: flags "not verified independently" for coverage claim — honest but not structural
The fix: Add item #10 to checklist ("RF completeness: §6-8 present or explicitly N/A") + add audit step to workflow.
E5: research/base.md Step 6 Gap¶
Step 6 Synthesis (lines 83-93) lists what to include:
3. HL Update Recommendations (table)
4. Fact Candidates — review conversation history first
5. Iteration Status block (mandatory) — see RES template
6. Conclusion (1 paragraph)
Missing: Findings Map. It's in the RES template (lines 73-82) but not mentioned in Step 6.
Template section is there, workflow instruction is not. Same disconnect pattern as RF §6-8.
E6: docs.md Diagram Collection Opportunity¶
docs.md has 6-item checklist:
1. Architecture changed?
2. New decision?
3. Something deprecated?
4. New tech debt?
5. New principle/convention?
6. Fact Candidates present?
Missing: "Diagrams/process docs to collect?" — no mechanism to extract mermaid diagrams or architecture visuals from RF §8 or RES Findings Map into persistent project documentation.
docs.md is at 443 words (757 headroom) — adding a diagram collection step fits easily.
Checkpoint¶
| Found | Remaining |
|---|---|
| Template-workflow disconnect is the exact mechanism | None |
| Word budgets have 375-757 word headroom each | None |
| Positive examples are user-attention-driven, not workflow-driven | None |
| REVIEW template missing §6-8 completeness check | None |
| research/base.md Step 6 misses Findings Map | None |
| docs.md has no diagram collection | None |
Sufficiency: - [x] External source used? (Confirmed by RLCF/TICK research in Gather) - [x] Briefing gap closed? (All structural gaps identified and quantified)
Stage complete: YES → User decision: ___