Christian Galo
81a11f8f5d
Re-run Phase A walkthrough against the seeded baseline (demo seeder + Keycloak ID pinning fix + person seeding). 14 slugs gain screenshot-v2.png; 10 previously-skipped slugs are now reachable. INDEX.md gains: - A Phase A v2 run-metadata block alongside the original v1 metadata. - Status upgrades for the 10 newly-reachable slugs (✗ → ✓ v2). - 10 cross-check findings (Phase A v2 findings section) comparing seeded evidence against the 7a.1 research doc — including the two-revoke-paths divergence in confirm-modal copy, per-org Issue Grant being plan-only, the empty-product-name interpolation defect on revoke-and-transition, and the rules form supporting only `limit` rules despite schema supporting boolean. Adds a README.md documenting Phase A's overwrite-in-place contract (non-LLM-keyed evidence reflecting current stack state).
Operator UX walkthrough evidence
Phase A captures of the member-console operator UI — screenshots, console
messages, network traces, accessibility trees, tab-order dumps, and empty-state
checks for each operator slug. The companion subjective walkthroughs live in
../operator-ux-walkthroughs/.
Contract
This directory is non-LLM-keyed and reflects the current stack state.
Phase A is mechanical: any agent running the briefing
(../operator-ux-walkthrough-task.md)
against the same stack should produce equivalent artifacts. Differences across
LLMs at this layer are noise (timestamps, viewport timing) and not worth
preserving in parallel trees.
When the stack state changes meaningfully (e.g., new seed data, schema migration that adds reachable slugs, route rename), re-run Phase A and overwrite this directory in place. Prior captures stay in git history if you need to compare.
Phase B (subjective walkthroughs) is per-LLM by design and lives in
../operator-ux-walkthroughs/ — one file per
LLM (e.g., claude-opus-4-7.md, gpt-5.md).
Current baseline
| Property | Value |
|---|---|
| Stack state | Seeded via ./test/seed-demo.sh (OpenSpec change 2026-05-10-member-console-demo-seeder) |
| Demo persons | bob, carlos, diana logged in once to populate identity.persons (Keycloak sub pinning is broken; see status/issues.md) |
| Capture agent | See INDEX.md for the run that produced the current artifacts |
| Slug coverage | See INDEX.md for the per-slug capture status table |
Git strategy
- Phase A re-captures: one commit per re-capture, message like
Re-capture Phase A against <stack-state>. Body lists what changed. - Phase B additions: one commit per LLM, message like
Add Phase B walkthrough by <llm>. - No branches per LLM, no parallel trees. The briefing in
../operator-ux-walkthrough-task.mdis the source of truth for "what Phase A captures"; if two agents disagree, fix the briefing.
Files in this directory
README.md— this file.INDEX.md— run metadata, per-slug capture status, findings, gaps.<slug>/— per-slug evidence (screenshot, console, network, a11y, etc.).<slug>/notes.md— for slugs intentionally skipped (with reason).