Summary

Protocol upgrade from v6.2 to v6.3, driven by a review of the Naval Combat Management System decomposition — the first system to complete the full QC + Validation pipeline. Eight systematic issues identified from sessions 187–197.

What went well

The v6.1/v6.2 changes produced measurably better output. Component counts varied realistically (4–8 per subsystem). Risk-driven priority worked — TEWA first, Track Management second. QC found 4 duplicate requirements, 8 degraded-mode gaps, and thin verification coverage. Validation found a real ASW depth gap and closed it. 206 requirements, 98% trace coverage, genuine cross-domain insights.

Changes

1. Duplicate requirement prevention (new step 4a in Flow B)

Problem: 4 EW requirements duplicated across sessions. The bot created requirements without checking existing ones.

Fix: Before creating any requirement, compare against /tmp/se_reqs.json. Skip if a near-duplicate exists (same component AND same capability).

2. Diagram reuse (modified step 4 in Flow B)

Problem: 4 duplicate diagrams (TEWA, Sensor Management, Communications each had 2). The bot created new diagrams instead of reusing existing ones.

Fix: Before airgen diag create, check airgen diag list for matching subsystem name. Reuse if found.

3. Degraded-mode quantification rule (Hard Constraints)

Problem: 8 requirements like “SHALL maintain degraded operation” with no measurable acceptance criteria. QC caught them but couldn’t fully fix them.

Fix: Hard rule: every degraded-mode requirement MUST include quantified minimum performance thresholds (what capability, numeric floor, trigger conditions).

4. Verification coverage (new step 6a in Flow B, strengthened QC)

Problem: Only 17/206 (8.2%) requirements had verification entries. Most subsystem and interface requirements had no verification approach.

Fix: Every interface requirement MUST have a VER entry. Target 50% SUB+IFC coverage per session. First-pass-complete check reports VER coverage stats.

5. PART_OF reconciliation (extended step 9 in Flow B)

Problem: Only 20 PART_OF facts for 64 components — many decomposition relationships not stored.

Fix: Before marking first-pass-complete, count entities vs PART_OF facts. If mismatched, add missing relationships.

6. Lint noise suppression (extended step 7 in Flow B)

Problem: “CMS lacks Physical Object trait” HIGH-severity finding appeared in every session, always acknowledged as ontologically correct. Pure noise.

Fix: Store LINT_ACKNOWLEDGED facts for reviewed findings. In subsequent sessions, skip acknowledged findings and report: “N previously acknowledged lint findings unchanged.”

7. Architecture decisions (new step 3a in Flow B)

Problem: ARC document had 1 entry (duplicate tracking note). No design rationale recorded.

Fix: At least one ARC decision per subsystem decomposition, documenting WHY (trade-offs, constraints, alternatives considered).

8. QC reinforcement (extended Flow C steps 3–5)

Added explicit QC checks for: degraded-mode quantification, duplicate detection with tagging, duplicate diagrams, PART_OF completeness, ARC completeness per subsystem, and 50% VER coverage target.

Implications

1. From session 199 onward, all new decompositions enforce lookup-before-write guards.
2. The Smart Building Management System (next in queue) will be the first to produce ARC decisions and mandatory VER entries.
3. Previously completed systems are not retroactively updated.
4. Protocol version is now 6.3.

Version manifest

Component	Version	Notes
Session protocol	6.3	Up from 6.2
New steps	3a, 4a, 6a in Flow B	Architecture, duplicate check, verification
Modified steps	4, 7, 9 in Flow B	Diagram reuse, lint ack, PART_OF reconciliation
Hard Constraints	+degraded-mode rule	Quantified thresholds required
QC flow	Steps 3–5 extended	7 additional checks