c16 — Seat Composition Review (wf3)

Question: for each persona seat, is the collection of pages right — does it fully serve the seat's job, what pages are missing, and which are best combined? (This is an IA/membership review, distinct from seat-framing-review.md, which critiqued per-page content.)

Current seat map

14 unique deep pages + 5 seat homes + agent-profile (a drill target, not in any seat nav). Shared-by-design (each now lenses by ?seat): agent-inventory, findings, data-protection, security-posture, cost-center, fleet-overview, policy-enforcement.

Per-seat analysis

Security · CISO — richest seat; two real gaps

• Fit: strong. Inventory (who) → Findings (what needs a human) → Posture (how safe) → Identity (credentials) → Data Protection (what data) → Policy (what's allowed). A complete control loop.
• Missing:
  • Activity / Decision Log — a behavioral forensic trail (agent → task → tool calls → actions → outcome → timestamp). The data exists (activity families/timeline; per-agent data_access/mcp_servers) but there's no "what did this agent actually do" timeline. This is core CISO forensics and incident response; c15 had an audit-trail, c16 dropped it. Add.
  • Capability & Action Authorization — "what each agent may do" (tool/action allow-list matrix). Phase-1 net-new; today only hinted inside Policy & Enforcement. Recommend folding into policy-enforcement as its first-class matrix rather than a separate page (see Combine).
• Combine: agent-identity is a credential-focused slice of agent-inventory; consider making it a view/tab of inventory rather than a standalone page (or keep it — it's a distinct fleet-wide NHI surface; low urgency).

Finance · FinOps — thinnest seat; missing the forward-looking half

• Fit: partial. cost-center (spend) + showback (per-person attribution) cover visibility.
• Missing (the defining gap):
  • Budgets & Spend Controls — per-agent/per-team budgets, burn-down, forecast, spend-ceiling alerts, ceiling-breach policy. Phase-1 net-new #2. cost-center wf3 added budget-vs-actual inline, but Finance's control surface (set ceilings, get alerted, enforce) deserves its own page. The single most warranted add, and it fixes the seat being thin. (Pairs with policy-enforcement's spend-ceiling card — that's the enforcement side; this is the planning/owning side.)
  • Vendor / Provider Spend (procurement lens) — spend by provider, model $/token, commit/DPA status, negotiation levers. costs.by_provider exists but no surface. Optional add.
• Combine: fleet-overview in the Finance seat is largely redundant with finance-home + cost-center; it earns its place only as shared context. Low-value here — could drop from Finance (keep Exec/shared).

Executive — well-served, but home and board overlap heavily

• Fit: good. executive-board is the packaged read; posture/cost/fleet are drill-downs.
• Combine (clearest in the whole app): executive-home ≈ executive-board. Both lead with the same four signals (governance index, governed %, spend-at-risk, high findings). Differentiate or merge: make executive-home = the live drill hub (at-a-glance + cards to surfaces) and executive-board = the exportable, trended, narrative report (the thing you take to a board). Right now they duplicate.
• Missing: an ROI / value-realized angle (what governance prevented, $-at-risk avoided over time) would complete the exec story — aspirational; partially implied by the board's trend.

Engineering · Platform — operates the fleet; missing the outcome dimension

• Fit: good for operations. performance (latency/cost) + inventory (runtime) + policy (config)
  • integrations (pipes).
• Missing (the biggest conceptual gap in the product):
  • Agent Reliability / Task Outcomes — success rate, goal completion, output quality, hallucination/error rate, human-intervention rate. Phase-1 net-new #3; performance already flags "latency ≠ task success" and names this as the unmeasured dimension. Add (even as a largely-projected surface, it stakes the most important agent-native axis). Belongs to Engineering + Executive.
  • Agent Topology / Delegation — multi-agent who-delegates-to-whom / sub-agents. Agent-ecosystem- native, absent. Aspirational add (Engineering/architecture).
• Combine: keep performance and a future reliability page separate (latency/cost vs. outcomes are different questions) — they're siblings, not a merge.

Compliance · Privacy — evidence-strong; missing data lifecycle

• Fit: good. compliance (frameworks/evidence) + data-protection (exposure) + findings (raw behavioral evidence) + showback (accountability).
• Missing:
  • Agent Memory & Record Stores / Data Residency & Retention — where conversation history, agent memory, vector/RAG stores, and task records live, and how long they're retained. c15 had data-residency; c16 dropped it. A distinct privacy/records mandate beyond "what files were touched." Add.
• Combine / overlap to watch: compliance is findings-mapped-to-frameworks while findings is the raw triage queue — keep both, but ensure compliance reads as a conformance rollup, not a re-listing of findings. showback here is a borrowed Finance view; keep it (lensed to accountability) but it's the weakest member.
• Membership: Compliance should also see agent-identity (credential conformance is a compliance concern) and integrations' GRC/evidence-archive connectors. Add to seat.

Cross-seat structural findings

1. Seat balance is uneven. Security 6 / Finance 3 / Exec·Eng·Compliance 4. Finance is under-built relative to its mandate — and Phase 1 said Finance's gap is exactly the forward-looking control surface. The recommended adds rebalance toward Finance and Engineering.
2. Seat-home vs. seat-lensed fleet-overview overlap. Each seat home leads with the same metrics the seat-lensed fleet-overview now leads with. Decide their division of labor: home = thin launchpad (1-line takeaway + the "your surfaces" card grid, minimal KPIs) and fleet-overview = the full dashboard; or merge the home into the seat-lensed fleet-overview so the landing is the dashboard. Today the top of both duplicates.
3. Shared-by-design is working — the seat lenses (wf3) make a shared page read differently per seat, so duplication across seats is a feature, not debt. Keep extending it to new shared pages.
4. agent-profile correctly sits outside the seat nav as a drill target. Leave it.

Recommendations

A. Pages to ADD (prioritized)

B. Pages to COMBINE / sharpen

C. Seat-membership tweaks (reuse existing pages)

• Compliance += agent-identity (credential conformance) and integrations (GRC/evidence connectors).
• Finance = cost-center · showback · Budgets & Spend Controls (new) · (drop or keep fleet-overview as context).
• Engineering += Agent Reliability (new); keep performance/inventory/policy/integrations.
• Executive += Budgets & Spend Controls and Agent Reliability (both board-relevant); resolve home/board overlap.
• Security += Activity / Decision Log (new); fold Capability/Action Authorization into policy-enforcement.

Recommended target seat map (after P1 adds + combines)

Suggested next step

Build the three P1 pages — Budgets & Spend Controls, Agent Reliability / Task Outcomes, Activity / Decision Log — as a wf4 iteration (plus the executive home/board de-dup and the capability-matrix fold into policy-enforcement). That closes the last of the Phase-1 net-new gaps (#2 Budgets, #3 Task Outcomes) and gives Security its forensic trail. Then the page set is complete enough to graduate the lead surfaces to Phase 3 hi-fi mockups.