feedback_always_save_scorecards

When a decision is made by weighted scorecard in any INHERIT v2 / TT-strategy session, persist the scorecard as an auditable artefact in git, not just as an in-chat table that disappears into the transcript.

Why

Rich’s explicit directive on 2026-04-24:

“PLEASE CONTINUE TO SAVE ALL SCORECARDS: THEY ARE INVALUABLE. A GREAT CLAUDE FEATURE”

Scorecards are load-bearing evidence for high-consequence architectural decisions (Shape X angles, $ref coherence sub-decisions, module-authoring sequence, license-safety scaffolding). When a critic later asks “why did you pick vi-6?”, the answer must be an auditable file showing the 15 criteria, the weights, the per-option scores, the sensitivity analysis, and the honest critique — not a half-recalled summary.

This also supports the reframe-beats-re-weight discipline: if the scorecard is persisted, future readers (Rich, reviewers, me in next session) can see exactly which reframing produced the structurally-superior option, and verify it wasn’t weight-fudging.

How to apply

Every time a weighted scorecard is produced (even inline during a decision), save it to disk as a markdown file with:

1. Location: docs/superpowers/scoping/YYYY-MM-DD-scorecards/<topic>-scorecard.md

   • If the scoping directory already exists for that date, add to it
   • If not, create the directory + a README.md index file listing every scorecard chronologically

2. Frontmatter (required):

   ---
   title: "<angle/decision> scorecard — <winning option> wins at <%>"
   version: "1.0"
   status: approved
   date: YYYY-MM-DDTHH:MM
   lastmod: YYYY-MM-DDTHH:MM
   author: "Rich Davies"
   source: "docs/superpowers/scoping/YYYY-MM-DD-scorecards/<slug>.md"
   ---

3. Body structure (required sections):

   • Question — one-line framing of what’s being decided
   • Options — enumerated list with short descriptions (e.g. vi-1, vi-2, … vi-6)
   • Criteria + weights — 15-criterion framework summing to 100; justify any non-default weighting
   • Per-option scores — 1-5 per criterion; total = sum(weight × score) out of 500; percentage = total/500
   • Sensitivity analysis — what weight changes would flip the winner?
   • Honest critique — what’s weak about this scorecard? Were any options missing? Was the winner <95%? If so, was reframing considered?
   • Locked in — cross-reference to the architecture doc / amendment / commit where the decision is codified

4. Update the scorecards README.md index with the new row (ensure chronological + numbered)

5. Commit with message: docs: <decision> scorecard — <winning-option> at <%>

Anti-patterns

• In-chat tables only — once context compacts, the artefact is gone. Save to disk.
• Skipping honest critique — a scorecard without a critique section is a decoration, not an audit artefact. Include “what’s weak about this” every time.
• Reconstructing post-hoc without admitting it — if the per-criterion scoring wasn’t preserved, mark the file note: "PARTIALLY RECONSTRUCTED — structural verdict certain; per-criterion scoring detail not fully recalled." (exemplar: viii-OWL-fragment-multi-profile-scorecard.md)
• Single-option scorecards — if there’s only one option, it’s not a scorecard; document the rationale in an amendment instead.

Coverage to date (2026-04-24, 15 scorecards)

Located at docs/superpowers/scoping/2026-04-24-scorecards/:

#	Decision	Winner	Score
1	Angle iii upper-ontology	C (PROV-O + BFO peer)	87.2%
2	Angle i top-level class count	i-ζ (9 classes)	89%
3	A-20 RoleInstance promotion	1-γ (strong + equivClass)	86.2%
4	A-20.1 post-FIBO-spike correction	C1 (α split + 14-pair)	94.6%
5	Angle v T-box layering	α (3-layer)	~85-90% reconstructed
6	Angle viii OWL fragment	viii-3 (multi-profile via ROBOT)	~91% reconstructed
7	A-22 alignment-axiom strength	iv-3 (per-target matrix)	94.4%
8	A-23 faceted classification	vi-6 (OMG Commons 4-layer)	94.6%
9	A-21 extension/modularity	vii-3 (alignment-axiom-only)	96.8%
10	A-24 Phase-1 authoring sequence	Seq G1 (gap-discovery)	94.0%
11	LS-5 license safety scaffolding	LS-5 (full content + CI)	95.0%
12	D1 module import chain	D1-ε (β-source + γ-consumer)	96.6%
13	D3 role placement	R-1’ (intro-module + 3 refinements)	100.0%
14	D6 facet classifier placement	AC-1 (intro-module + Core-fallback)	100.0%
15	D4 Will/Codicil/LegacyLetter boundary	W-1’ (all Wills + abstract parent)	100.0%

Future scorecards extend this registry.

Cross-references

• README: docs/superpowers/scoping/2026-04-24-scorecards/README.md (v1.1 — index across all 15)
• feedback_reframe_beats_reweight.md — complementary discipline: when a scorecard stalls at 85-92%, question the framing before re-weighting
• feedback_introduction_module_rule.md — exemplar of a rule surfaced by scorecards (3 × 100% for R-1’/AC-1/W-1’)