Composition notes

• Question: what is this person's AI footprint, and is it accounted for? The footprint is the chain (devices → installs → accounts → sessions → surface); the accounting is whether each link is covered, monitored, enterprise, and producing telemetry. The verdict answers the accounting in one sentence; the body lays the footprint out.
• Verdict line from verdicts.userDetail(world, id), priority no-sensor → personal account (warn) → stalled joiner → detected-only → clear, wrapped by userDetailVerdict which adds one rung above clear: an account with no agreement on file (xAI's team id) is signal, not clear. The personal account outranks everything but "we cannot see her at all" — the c21 §5 ruling.
• The chain is the hero (c21 §2.9), at two altitudes: the strip of Phase1EntityChips right under the verdict (subs carry exceptions only: no sensor · detected-only · personal · seat only), then the sectioned body walking the same links at reading altitude. Two pages, one graph — this is the people lens; "view in Inventory" carries ?person= for the technology lens.
• Headline stats are four Phase1StatBlocks (tokens · spend · installs · sessions) with one attention device per state, on the sub line: Dana's is the sessions block (n on a personal account, warn), Wes's the sessions block (install seen, never run), Jonah's the installs block (detected-only). Alexis's coverage gap is the verdict's and the chain strip's to say, not a fifth block. Tokens are measured, dollars estimated; intensity rides the spend sub (thresholds: heavy ≥ 40M · moderate ≥ 15M · light > 0 tokens · 30d).
• Personal accounts unmissable (c21 §2.6, §5): the account row is bordered in the error register and names the device and the enterprise org it sits beside, with the personal-session count; the sessions table flags those rows; the strip's account chip says personal. Observation only — the block-personal-accounts control is pointed at, not built.
• Coverage honesty: an uncovered device is a dashed amber card that says which kind of gap it is (Windows platform gap vs deployable today) and that the empty rows are a statement about the sensor, not the person. Seat held with no sensor → the investigate line on the account row. Detected-only installs get the three-absences note.
• Sessions attach to users (c21 §2.7) and stay a list — DataTable with per-column slots, row click and "receipt →" both to Session detail. Labelled honestly: the 7-day sample plus older scenario sessions, 30-day tokens measured separately and never summed from the list. Sensitive-file and policy-event counts come from sessionTotals (events when present, file links otherwise).
• Deliberate omissions (c67 v8): credential severity math → /identity (quiet link), spend history → Cost Center (quiet sub on the $ stat), per-agent config → Agent detail. The LLM work summary in the world is not rendered — no designed treatment yet; it stays a fixture field.
• Derivation: everything comes from personFootprint(world, id) in derived/user-detail.ts over the world's walkers; nothing hand-typed, nothing fabricated. First seen reads from the earliest install, so a person with no installs prints the honest absence.