Skip to content

design: #170 v4 — grounded audit → request-origin IA (DESIGN.md v4 + golden ref + build plan) - #172

Merged
hellno merged 5 commits into
mainfrom
hellno/montreal-v1
Jul 2, 2026
Merged

design: #170 v4 — grounded audit → request-origin IA (DESIGN.md v4 + golden ref + build plan)#172
hellno merged 5 commits into
mainfrom
hellno/montreal-v1

Conversation

@hellno

@hellno hellno commented Jul 2, 2026

Copy link
Copy Markdown
Owner

Delivers the first step of #170 — a grounded audit of the running product on a funded just demo.

What

Ran the app live on macOS against a funded just demo (anvil fork of Sepolia @ block 10822990, verified-reads off) and drove every flow, screenshotting each as-built and grading it against the north star (DESIGN.md) + the versioned golden references (designs/deckard-editorial-v3.html, designs/deckard-agent-v4.html).

  • Report: docs/research/11-ia-flows-audit.md
  • Evidence: docs/screenshots/audit-170/ (17 curated shots: unlock, home, palette, send compose + clear-signing, shield, swap, receive, agent surface, activity empty/needs-you/approval, settings, onboarding welcome + secure)

Verdict

The bones are strong and on-brand — the overhaul is a consistency/trust pass, not a redesign. Fonts are bundled, the mono hero + hairline cockpit rows + amber-human/cyan-agent axis + ⌘↵ transaction-as-hero confirm + the see-and-stop Activity loop all render as designed. The Activity feed + agent-approval beat is the best-realized part of the product.

Four load-bearing (P1) gaps, where the "autopilot you can trust and stop" promise currently leaks:

  1. Identity collapses to "Personal / Wallet" for an unnamed wallet (breadcrumb says literal "Wallet"; page title falls back to the project name).
  2. Two different clear-signing screens — Send's transaction-as-hero vs a weaker boxed key/value card for the agent approval — breaking the one-shared-review contract.
  3. The agent surface is a read-only policy dump (no per-row edit, dimmed Pause/Rotate/Adjust, no "what Atlas did" feed) — the thing v2 said it removed from the home.
  4. Displayed cap ≠ enforced cap: the agent advertises "0.1 ETH per move" but a 0.15 ETH shield auto-approved and broadcast (shields are only gated by the daily cap).

Plus P2 (money loses its USD/verified context below the hero) and P3 consistency items — full list in the report. Fork-conditioned observations (no price feed) and synthetic-driving artifacts (GPUI eats synthetic keys) are separated out explicitly so they aren't mistaken for product defects.

Docs-only

No code, deps, or lockfile changes. .context/ (the throwaway driving harness + full 40-shot set) is gitignored.

Refs #170.

Ran the app on a funded `just demo` (Sepolia fork, verified-reads off) on macOS and
drove every flow live — onboarding, unlock, home, send/swap/shield/receive, clear-signing,
agent propose->approve, activity->STOP, settings — screenshotting each as-built and grading
against the north star + the versioned golden references.

Verdict: the bones are strong and on-brand; the overhaul is a consistency/trust pass, not a
redesign. Four load-bearing (P1) gaps: (1) unnamed wallet reads as 'Personal/Wallet'
everywhere; (2) two different clear-signing treatments (Send hero vs a boxed agent-approval
card) break the one-shared-review contract; (3) the agent surface is a read-only policy dump
(no per-row edit, dimmed controls, no 'what Atlas did' feed); (4) the agent advertises a
0.1 ETH per-move cap it doesn't enforce on shields (a 0.15 ETH shield auto-broadcast).
Plus P2 (money loses USD/verified context) and P3 consistency items. Fork-conditioned
observations and synthetic-driving artifacts are separated out explicitly.

Evidence: docs/screenshots/audit-170/ (17 curated shots).
hellno added 4 commits July 2, 2026 19:29
…170)

The design-consultation output on top of the flow audit (#170). One wallet, one shared
Review, N request origins (you / dapp / agent). Adds designs/deckard-v4.html (five views:
wallet home, shared Review, Activity, Transaction, rail-focus) built on the exact v3 system,
and docs/research/12-design-direction-v4.md (competitive research + two outside voices +
the locked v4 decisions). Kept the visual system; changed IA/flows: three-pane always-on
rail, request-origin model with auto-handles, dropped the projects layer, one simplified
clear-signing Review, a Transaction detail view, identity masthead (no literal 'Wallet'),
holdings with a mono $ column + decimal alignment, honest cap ledger. Refs #170 #174.
…cted state

design-review: the rail-focus view shows a selected request (.preq.sel) with a populated
detail rail, but the Needs-you head still read 'Nothing selected.' — a contradictory
empty-state string. Removed it from this view.
Codify the /design-consultation outcome. Visual system unchanged; structure reworked:
three-pane always-on rail; the request-origin model (you/dapp/agent, one shared clear-signing
Review, origin as identity + state-color trust badge, never a third signal); dropped the
Projects layer; auto-assigned renamable agent handles (retired 'Atlas'); a Transaction detail
view; identity masthead + the no-'Wallet'-label rule; holdings $ column + decimal alignment +
hero USD/verified meta; honest cap ledger; a platform-aware key_cap widget (⌘ mac / Ctrl Linux);
'This can't be undone.' danger copy. Updated IA, clear-signing, golden refs, the visual DoD, and
the decisions log. Golden ref: designs/deckard-v4.html. Refs #170 #174.
The agent-proof implementation plan for the v4 IA: an epic + 8 children (E0 driving recipe,
E1 widgets, E2 identity, E3 three-pane rail, E4 money-context, E5 trust-critical shared-review
+ enforced cap [merged], E6 activity/waiting, E7 transaction view, E8 palette coverage). Adds a
shared Implementer Contract (reuse-widgets convention, never-regress invariants, two-tier DoD)
and per-view fidelity checklists derived from designs/deckard-v4.html. Refs #170 #174.
@hellno hellno changed the title docs(design): grounded flow audit of the funded demo (#170 discovery) design: #170 v4 — grounded audit → request-origin IA (DESIGN.md v4 + golden ref + build plan) Jul 2, 2026
@hellno

hellno commented Jul 2, 2026

Copy link
Copy Markdown
Owner Author

Scope grew beyond the audit — this PR now carries the whole design pass for #170 (docs-only, no code/deps):

  1. Auditdocs/research/11-ia-flows-audit.md + docs/screenshots/audit-170/ (the grounded flow audit).
  2. Directiondocs/research/12-design-direction-v4.md (competitive research + two outside voices + locked decisions).
  3. Golden refdesigns/deckard-v4.html (the v4 pixel ground-truth; design-review pass applied).
  4. DESIGN.md v4 — the request-origin IA codified (three-pane rail, one shared Review, dropped Projects, identity/money/handle rules, honest cap ledger, platform-aware key-cap, This can't be undone. copy, updated visual DoD + decisions log).
  5. Build plandocs/research/13-v4-build-plan.md (the epic + 8 children with the Implementer Contract + per-view fidelity checklists — the backlog for the GPUI implementation).

Still docs-only. The cargo-deny-advisories red is the pre-existing advisory set already on main (this PR changes no dependency files).

@hellno
hellno merged commit 1c5b702 into main Jul 2, 2026
5 checks passed
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant