Skip to main content

Clarity Spec — Feedback for Mockups

Date: 2026-02-27 Spec reviewed: Clarity: so what? / what do I do now? Purpose: Technical feedback before Sergey starts mockups in Stitch.


Overall Verdict: Good to Build — With 4 Notes

The three-step flow, the execution-determined framing, and the CISO/Analyst layer separation are all sound and implementable. Sergey can start mockups. These notes are things to be aware of during design, not blockers.


What's Good — Build As-Is

The three-step mental model works. Overview → Cluster Exposure Brief → Authority Paths → Path Evidence maps cleanly to existing screens. No new top-level pages needed.

"Functional cluster name" over attribute labels is right. "Orphaned + Sensitive" as a primary label forces the CISO to decode two dimensions simultaneously. A functional name like "Unowned HR Automation" is immediately comprehensible.

Section D remediation bullets are the right scope. 1-3 advisory bullets, no enforcement logic. This is deliverable now.

"View Authority Paths (N)" button is smart UX. Hiding the table behind a click prevents cognitive overload on the Exposure Brief page. Good decision.

Observed vs. Potential separation in Path Detail (Step 3) is minimal work. The current page already shows authority. Adding a collapsed "Standing Authority (Not Exercised)" panel is a minor UI change.


4 Notes for Mockup Design

1. The verdict sentence needs "Endpoint Type" — but it's available

The template:

<N> autonomous paths exercised <Domain>-scoped authority and invoked <Endpoint Type> <X> times in the last 30 days.

"Endpoint Type" is not a current entity, but it maps to data we already collect:

  • Foundry: connection_type (e.g., AzureOpenAI, AzureBlob) + egress_category (llm, cloud, external)
  • Entra: resource_display_name from sign-in logs (e.g., "Microsoft Graph")
  • ServiceNow: connection_subtype (rest_message, soap_message, http_connection)

For mockups: Use concrete examples like "invoked LLM API endpoints" or "invoked Microsoft Graph" rather than a generic placeholder. This will look right in the final product.

Full analysis: Endpoint Type Feasibility

2. Evidence grades should be visible — the spec doesn't mention them

The spec frames everything as "observed authority" but doesn't address the reality that not all paths have the same quality of evidence. Per the Execution Evidence Feasibility Study, we recommend Evidence-Graded Authority Paths:

  • Grade A — Confirmed execution (deterministic proof)
  • Grade B — Inferred execution (temporal correlation, partial evidence)
  • Grade C — Standing authority (configured but no proof of exercise)

Why this matters for mockups:

  • The Overview (Step 1) risk badge should reflect evidence grade, not just governance condition.
  • The Exposure Brief (Step 2) Section B ("Am I Exposed?") should show the evidence grade distribution — e.g., "14 paths confirmed, 38 inferred, 90 unconfirmed."
  • The Authority Path table should have a visible grade column.

Suggestion for Sergey: Add a small evidence grade indicator (A/B/C or a colored dot) next to path counts wherever they appear. This is SV0's differentiator — no competitor grades evidence quality.

3. Section C governance items have mixed availability

The spec lists four governance conditions:

ConditionStatusNotes
Orphaned?Available nowownership_status already evaluated
Long-lived?Available nowToken/credential age tracked via created_at
Drift?Not built yetRequires temporal diff engine — DRAFT feature
Identity reuse?Not built yetRequires cross-path identity correlation — not in current evaluator

For mockups: Design all four in the layout (they're all valid concepts), but Sergey should know that Drift and Identity Reuse will show "Not yet analyzed" or similar placeholder states in the near-term build. Orphaned and Long-lived are ready to populate.

4. The "Standing Authority (Not Exercised)" panel in Step 3 needs path materializer awareness

The spec says: "Add collapsed panel: Additional Standing Authority (Not Exercised)"

Today the path materializer (authority-path-materializer.ts) is structural-only — it traverses the permission graph (HAS_ROLE → GRANTS → APPLIES_TO) without distinguishing observed from potential. To split paths into "exercised" and "not exercised" panels, we need the materializer to join against execution evidence.

For mockups: Design the collapsed panel as shown. The backend work to split observed vs. standing is scoped at ~1-2 Claude Code sessions and is already on the roadmap.


Summary Table

Spec ElementBuild StatusMockup Guidance
Three-step flowReadyBuild as-is
Functional cluster namesReadyBuild as-is
Verdict sentenceReady (endpoint type composable)Use concrete examples ("LLM API", "Microsoft Graph")
Risk badgeReady, but needs evidence gradeAdd evidence grade indicator
Section A — What HappenedReadyBuild as-is
Section B — Am I Exposed?Ready, needs grade distributionAdd grade breakdown (14 confirmed / 38 inferred / 90 standing)
Section C — Governance2 of 4 conditions readyDesign all 4; Drift + Identity Reuse show placeholder
Section D — RemediationReadyBuild as-is
Authority Path tableReadyAdd evidence grade column
Standing Authority panelBackend ~1-2 sessionsDesign the collapsed panel as-is
Observed vs Potential separationConceptually readyBuild as-is

Bottom line: The spec is well-structured and implementable. The biggest design-level addition is making evidence grades visible — this is what differentiates SV0 from every competitor that shows paths flat without evidence quality.