feat: optimise SOBC command prompt via autoresearch (7.7 → 8.9)

[tractorjuice]

Summary

• Optimised the /arckit.sobc command prompt using the autoresearch loop (19 iterations, 5 keeps, 14 discards)
• Baseline score: 7.7 → Final score: 8.9 (improvement: +1.2)
• Aligned to HM Treasury Green Book 2026 methodology

Changes to prompt

Keep	Score	Description
Baseline	7.7	Unmodified command
iter 2	8.0	Added Appraisal Summary Table (AST) with BCR to Economic Case
iter 6	8.3	Added Cross-Case Traceability Matrix with quoted stakeholder goals
iter 12	8.6	Added inline stakeholder goal quoting throughout all five cases
iter 19	8.9	Added post-write verify-and-patch for Theory of Change, CSF Assessment Matrix, and switching values

Green Book 2026 elements now generated

• Theory of Change (Inputs → Activities → Outputs → Outcomes → Impacts)
• CSF Assessment Matrix (5+ CSFs × all options, RAG-rated)
• Appraisal Summary Table (NPV, BCR, stakeholder goals %, key risks)
• Switching values (% change in costs/benefits that breaks the recommendation)
• Cross-Case Traceability Matrix (stakeholder goals quoted from STKE across all five cases)
• Inline goal referencing (pervasive STKE goal quotes throughout document body)
• Optimism bias adjustment (+40% for IT projects per Green Book)
• Sensitivity analysis with BCR for each scenario

Test plan

• Run /arckit.sobc 001 against a test repo with stakeholder analysis
• Verify all Green Book elements are present in output
• Verify traceability matrix quotes match STKE content
• Run python scripts/converter.py (already done — 7 files updated)
• Run markdownlint

🤖 Generated with Claude Code

[tractorjuice]

Superseded by #265 (combined autoresearch PR for all 48 commands)