Optimisation Loop Pilot 001 — homepage/team warmth and crossing system

[iainherd]

Goal

Run Optimisation Loop Pilot 001 using the branch-scored delegated design improvement process: capture the current deployed/main baseline, scout bounded high-value opportunities, generate candidate branches with delegated implementation agents, score candidates with expert and audience-agent review, and promote only a material improvement.

Scope

Pilot focus: homepage/team transition, visual warmth, editorial authority, and the ownable “crossing” system.

The pilot may improve composition, rhythm, hierarchy, interaction feel, placeholder discipline, and continuity between homepage and team page. It must not add fake photography, stock/AI imagery, new regulated claims, analytics/tracking, or broad copy rewrites.

Orchestration model

Morwenna is master orchestrator. Implementation and reviews are delegated to bounded agent profiles/subagents. Candidate branches are isolated; losing candidates are not merged.

Required process

• Use docs/workflow/branch-scored-design-loop.md.
• Use templates under docs/workflow/templates/.
• Capture baseline evidence under docs/design/loops/20260628-pilot-001-home-team-warmth/.
• Pre-declare scoring rules before candidate implementation.
• Dispatch design/brand opportunity scouts and UX researcher scenario setup.
• Generate 1–3 bounded candidate branches.
• Score trend/marginal gain and audience-agent findings.
• Promote only if a candidate clears the pre-declared materiality threshold with no hard blockers.
• No-promotion is acceptable.

Pre-declared scoring rules

• Formula: (candidate weighted score - baseline weighted score) / 75 × 100.
• Default promotion gate: +3 percentage points of max score, equivalent to +2.25 weighted points.
• Priority dimensions: composition and hierarchy; institutional credibility; human warmth; AI-template smell/originality; ownable crossing idea.
• Critical dimensions: accessibility; compliance/security risk; institutional credibility; imagery/placeholder discipline; performance.
• Regression guard: no critical dimension may regress by more than 0.25 weighted points; no non-critical dimension may regress by more than 1.0 weighted point unless explicitly accepted.

Audience-agent testing

Use agent:ux-researcher to define scenarios for:

• audience:institutional-investor
• audience:founder-dealmaker
• audience:intermediary-adviser
• audience:compliance-conscious-reviewer

Simulated audience findings inform score dimensions but do not replace real human/client testing and cannot bypass hard blockers.

Hard blockers

• Accessibility/WCAG blocker or hidden focus state.
• Secret/auth/deploy/security regression.
• Compliance-risky claims or regulated phrasing without approved source.
• Stock/AI/fake imagery or placeholder leakage beyond approved pre-prod usage.
• Performance budget failure.
• Broken pre-prod password gate or route smoke.

Non-goals

• Do not launch production.
• Do not add analytics/tracking/instrumentation.
• Do not claim simulated audience agents are real market proof.
• Do not solve production photography in this pilot.

[iainherd]

Pilot 001 restarted under production-design craft gate

Pilot 001 is now restarted after #63/#64 added the production-design / pixel-perfect craft QA gate.

Baseline and process state

• Current main: 15bfdc439e58f83352f66d6202de6cb0287662f5
• Candidate branches/worktrees reset to current origin/main before implementation.
• Candidate scoring remains blocked until production-design review is PASS or focused changes are fixed and re-reviewed.

Candidate implementation dispatch

Delegation: deleg_eb70e1c5

Candidate	Branch / worktree	Bounded hypothesis	Craft requirements baked into prompt
A	design-loop/20260628-pilot-001-a-crossing / /root/lircap/worktrees/pilot001-a-crossing	Carry crossing/route/fix system through homepage → team	Fix/avoid nav wrapping, mobile nav heaviness, homepage illustration order/scale, heading rhythm, divider inconsistency, affected responsive craft
B	design-loop/20260628-pilot-001-b-provenance-team / /root/lircap/worktrees/pilot001-b-provenance-team	Provenance-first team warmth / editorial judgement profiles	Fix /team/ 1180px overflow, portrait-placeholder dominance, team heading rhythm, card density, placeholder discipline
C	design-loop/20260628-pilot-001-c-judgement-map / /root/lircap/worktrees/pilot001-c-judgement-map	Judgement map / route ledger bridge	Improve route/proof clarity without claims; preserve/repair nav rhythm, heading system, divider hierarchy, responsive spacing

Non-negotiables

• No stock/AI/fake imagery.
• No new regulated/proof/access claims.
• No analytics/tracking.
• Preserve accessibility, performance, placeholder discipline, and compliance constraints.
• Expanded craft breakpoint attention required: 320, 390, 768, 980, 981, 1024, 1180, 1181, 1280, 1440, and 1728/1920 where practical.

Next step after implementers return: parent verification, candidate browser/craft evidence capture, production-design QA, design/brand review, audience-agent UX testing, scorecards, then promote/no-promote decision.

[iainherd]

Pilot 001 candidate implementation returned; parent verification started

Delegation deleg_eb70e1c5 returned with:

• Candidate A: implementer timed out, but left a concrete local diff in its isolated worktree.
• Candidate B: completed with implementation summary.
• Candidate C: implementer timed out, but left a concrete local diff in its isolated worktree.

The timed-out A/C worktrees were not trusted from agent summary. Parent inspected them directly and ran verification.

Worktree status

Candidate	Worktree	Changed files	Parent verification
A	/root/lircap/worktrees/pilot001-a-crossing	src/components/CrossingHero.astro, src/pages/team.astro, src/styles/global.css	passed command suite; screenshots/metrics captured
B	/root/lircap/worktrees/pilot001-b-provenance-team	src/pages/team.astro, src/styles/global.css	passed command suite; screenshots/metrics captured
C	/root/lircap/worktrees/pilot001-c-judgement-map	src/pages/index.astro, src/pages/team.astro, src/styles/global.css	passed command suite; screenshots/metrics captured

Parent verification commands

All three candidates passed:

npm test -- --run
npm run typecheck
npm run lint
npm run build
npm run perf:budget
python3 scripts/check-placeholders.py

Objective browser/craft metrics captured

Candidate screenshots and browser-qa.json were captured in each worktree under:

docs/design/loops/20260628-pilot-001-home-team-warmth/candidates/<a|b|c>/

Routes captured:

/
/team/
/contact/

Widths captured:

320, 390, 768, 980, 981, 1024, 1180, 1181, 1280, 1440, 1728

Parent metric summary:

• A/B/C: overflowX = 0 at all captured route/width combinations.
• A/B/C: no hero chart/H1 intersections detected.
• A/B/C: no console/page errors detected in captured matrix.
• A: no nav label wrapping detected in captured matrix.
• B/C: still preserve For investors nav wrapping at some tablet/intermediate widths, so this is a likely craft-gate issue unless explicitly accepted as out of candidate scope.

Review delegation dispatched

Delegation deleg_8c7732f4 is now reviewing the candidates:

1. agent:production-design-reviewer — craft QA across screenshots/metrics.
2. agent:design-director + agent:brand-strategist — direction/brand/score implications.
3. agent:code-security-reviewer — compliance/security/process safety.

No candidate will be scored or promoted until production-design review passes or focused changes are fixed and re-reviewed.

[iainherd]

Pilot 001 specialist review verdicts

Delegation deleg_8c7732f4 completed specialist reviews for candidates A/B/C.

Production-design craft QA

Candidate	Verdict	Key reason
A	PASS	Best craft response: no overflow, no chart/H1 collision, no nav wrapping in captured matrix, improved mobile hero/fold hierarchy, lighter divider/header system, crossing motif better integrated.
B	NEEDS FOCUSED CHANGES	Team cards improved, but global nav wrapping remains and homepage mobile/tablet hero hierarchy remains baseline-like.
C	REJECT	Does not resolve major baseline blockers and adds more rule/card density; team remains framework-heavy rather than warmer.

Design-director / brand-strategist

Candidate	Verdict summary
A	Strongest overall for the full pilot: best ownable crossing system, strongest anti-template move, good composition/hierarchy, acceptable risk if restrained.
B	Best conservative/team-warmth candidate, but under-delivers on homepage/crossing-system ambition.
C	Strategically intelligent but dense, framework-heavy, least warm; useful ideas but not a final design candidate.

Code-security / compliance-risk

Candidate	Verdict	Key reason
A	PASS	No secrets/tracking/external assets/deploy changes; abstract route/fix language does not create concrete access/mandate/performance claims.
B	PASS	Provenance note reduces claims risk; no tracking/fake imagery/security concerns.
C	REQUIRED CHANGES	Judgement-map language and elevated focus areas may imply a formal process/access/capability model beyond approved copy.

Current promotion eligibility

Candidate	Promotion/scoring status
A	Eligible for audience testing and final scorecard.
B	Not promotable this round without focused changes and production-design re-review.
C	Rejected / not promotable this round.

Next step: run audience-agent UX testing and final scorecard for Candidate A against baseline. B can be mined later for team-card improvements if A needs a subsequent focused iteration.