What Great Looks Like
v1.2 · 2026-07-11 (v1.2: the VP relationship — jobs done WITH and BY your VP, per-job delegation dial, symbiotic learning system · v1.1: Claude Design canonical, JTBD-fit screens) · the experience standard every slice is measured against — cited in specs alongside AC IDs · sources: po-vision-canon, tier-strategy, connections-design, Clay's own words (Otter, 2026-06-30), the design refs, the standing design rules
The north star, in Clay's words to a client: "I prefer reality to data… it's a real-time shared source of truth. Each node in the network is connected… all of that information is permissioned… it's not that you need to reach out for it — it's there, it's connected, it's live for you… and everything is an ETA until it happens, updating live with the actuals as it goes."
The bar: Bloomberg Terminal for cannabis operations. Mission-critical infrastructure a founder opens with calm confidence — not a dashboard someone has to double-check.
And the relationship that makes it VP (Clay, 2026-07-11): "JTBDs are jobs that you have your VP do, and do together with your VP — supporting the president as much or as little as they would like to be supported, and the president supports the VP. The symbiotic, learning, self-improving system and 'vice-president' they always wanted."
Every job carries a delegation dial: together (VP brings context, president drives) → VP drafts, president approves → VP executes and reports. The dial is set per job, per president — and it moves rightward as the VP earns trust, because every correction and approval teaches it. Support flows both ways: that is the product.
Values — how VP behaves (each earned the hard way)
- Reality over data. Every figure traces to a source, passes invariants, and reconciles with its own narrative. A number that can't be trusted doesn't render. Wrong numbers on screen are the cardinal sin — worse than blank.
- Everything is an ETA until it's an actual. Forward-looking values are estimates that update live and visibly firm up on arrival. Derived figures carry confidence; confirmed facts carry none.
- Absent, never fake. Missing data degrades the panel, not the screen — no €0k, no "—" pretending, no invented placeholders. Real, blank, or editable beats caveated.
- Every party sees their own leg; nobody sees yours. Permissioning is the product. Hidden means physically absent. Sharing is explicit, logged, revocable-instantly.
- AI suggests, humans approve. VP speaks as "I", shows its per-signal reasoning, is honest about weighting, and always invites override. It drafts; people commit.
- Don't make people hunt. The chase is the product's job: every status has a responsible party and a working path to them. If a user has to go find something, the screen failed.
The experience standard — with exemplars (the pixels, not just the words)
| Standard | The exemplar to match |
|---|
| Screen ambition — screens are instruments, not lists | The Connections industry graph (/connections in prod — 3D/2D, stakeholder groups, search-everything). This is the ceiling to aim at, and the floor for "done" is its design ref. |
| One world; screens are JTBD-fit — the shared source of truth is universal, screens are not automatically. The test: does this stakeholder's job on the object differ, or only their permissions? Permissions only → the SAME screen through the lens (PO Tracker). Different job → a tailored screen on the same world objects (producer cultivation, processor QP release, retailer assortment) — tailored per stakeholder and per role within it. Never fork the data; never rebuild what's shared. | PO Tracker via viewer switcher (projection case) · vervana-pro's stakeholder modules + the role-specific JTBDs in the .dc refs (tailored case) |
| Real data discipline | The BCP cockpit period-engine (real daily aggregates, May €74,088 — a number with provenance and an audit trail behind it). |
| Design fidelity | The .dc.html refs in docs/vp-suite/ are canonical per screen (Choreography, Entity Profile, Messages, Data…). Match verbatim; deviations get flagged, never shipped. Design rules: defused reds, no navy-filled cards, brand gates on tenant surfaces. |
| Copy & voice | VP speaks first person, plainly, with reasoning ("2 of 6 POs need expediting — 2042 leads at 5d late"). Narrative and table must be the same fact. English for people, DE conventions on German tenant surfaces. |
| The demo world | The fictional cast (Steinbach, Atlas Verde, Rheinwerk…) is the single canonical world for every demo, fixture, screenshot, walkthrough — deterministic and versioned. |
The reference shelf — where everything canonical lives
| Question | Canonical source |
|---|
| Why does VP exist / the network thesis? | po-vision-canon/docs/vision/ (platform · data · tier-strategy) + connections-design spec + VIRAL_ADOPTION_STRATEGY |
| What must be true? (testable) | AC-CANON v1.3+ — SPINE/RBAC/ING/LENS/TIER/ARCH/TP |
| What does each screen look like? | Distributor: Claude Design is canonical — the claude.ai design project, synced two-way via /design-sync; its changes flow into Claude Code as canon. Repo-side artifacts: docs/vp-suite/*.dc.html refs. Plus pro.vervana.co (frozen concept, 84 pages) and shipped prod screens. |
| What does each NON-distributor stakeholder screen look like? | pro.vervana.co is the canonical reference — the real built screens for all 12 stakeholders (broker, lab, logistics, customs, market-releaser, remediator, trade-finance, processor, producer, retailer, insurance) live in ~/dev/vervana-pro/app/<stakeholder>/ (on the Hetzner box at ~/vervana-pro-ref/). PORT from these real screens — never paraphrase. PRECEDENCE (Clay 2026-07-12): vervanav4 is AHEAD of pro.vervana.co in many areas (esp. distributor). So: (1) for any SHARED screen, reuse vervanav4's advanced screen via RBAC projection — the distributor's advances propagate up to every stakeholder automatically; (2) fall to pro.vervana.co ONLY for stakeholder-specific tailored screens vervanav4 lacks (CoA workflow, cultivation, QP release, remediation, genetics). Never re-derive an advanced screen from the older concept. They also encode how each stakeholder connects to the spine + the vision. /demo becomes the clean, unified, comprehensive port of all of them. Clay 2026-07-12 |
| What remains to be done? | The Completion Board — 86 rows × 5 gates, regenerated, can't lie. Phases: distributor → processor → producer → every other stakeholder lens (ARCH-3) → seams (producer-observer, shop) → clients self-connect sources → live. |
| How do we ingest & organize data? | Ingestion Plan v1.2 (Allan's field-event pattern; FE/BE seam drawn) |
| What already exists in code? | The Atlas (3,074 exports, 71 routes) + BACKEND-CONTRACT.html + FE-SCHEMA-MAP.html |
| How do we work & improve? | The process contract + docs/process/process-log.md — violations flagged in the moment, both sides, Friday trend review. Canon amendments dated, via accept/redo only. |
What can be improved (current, honest)
- Spec-with-exemplar discipline: today's two inventions traced to specs citing ACs without exemplars. Standing fix: every codex spec names the exemplar screen/file to match — this document makes that citable.
- Gate depth: the sweep passed while real gaps existed twice; it gains assertions every time reality beats it. The parity pass (screenshots vs exemplar) is now mandatory and mine.
- Board truthfulness: gates only flip with evidence — still 0/86 until HITL verdicts land against real screens; that honesty is the feature.
- This document: v1, written by Claude from Clay's artifacts. Clay's redo direction sharpens it — especially the exemplar column, which should grow one row per standard as screens reach "great."
Cited in every codex spec alongside AC IDs from now on. Amend via "amend great: …". · Clay ⇄ Claude · 2026-07-11