Source: c78 fleet-overview-reports — mini-doc iteration 3 (after report-card-minidocs), 2026-08-24
New ruling: no greek lines. The gray bars evoked a document but said nothing — and a card this small can't afford decoration. Every piece of the mini report must communicate: each slot is a real, derived figure chosen because it helps someone understand or address an issue or inefficiency in the AI estate. The content audit below picks the slots per report; two compositions then set the same content. Hierarchy rules carry forward — the card's own title on top, weight marks the most informative line, color stays reserved for genuine tone (one device per card).
Content audit — what earns a slot, per report
Each report answers one question at card altitude; a slot is kept only if it serves that question. What's cut is on the report page one click away — cutting it here is the design.
Compliance
"Will we pass the audit, and what breaks it?"
- ·Score + delta + spark — audit posture and whether the needle is moving
- ·Failing controls, with owner — the blockers, routed to the person who addresses them
- ·Framework split — which certification is exposed — SOC 2 and NIST fail independently
- ·Attention count — the noted exceptions an auditor will ask about next
- ·Evidence volume + run id — is the trail deep and current — the auditor cadence
Cut: control names (labels, not signals), evidence kinds, snapshot history.
SLA Insights
"Are we resolving on time, and where is debt building?"
- ·% within SLA + delta + spark — the resolution promise, kept or slipping
- ·Open by severity — the workload, weighted — 1 high ≠ 3 medium
- ·Nearest-to-target finding — the next breach if nothing happens — early warning beats breach count
- ·Quarter found vs resolved — throughput debt — backlog grows before any SLA breaks
Cut: the target rubric (static config), resolved history, unresolved reasons (report-level detail).
Inventory
"Do we see the whole estate, and what entered it unvetted?"
- ·Coverage + delta + spark — the visibility promise itself
- ·Monitored / installs — the % made concrete — 5 blind installs is actionable, 87% is not
- ·Shadow + personal-account split — AI use outside enterprise controls — the data-drift signal
- ·Newest shadow install — the next pickup, named — tool, host, account
- ·Sanction exceptions — unvetted models and MCP servers entering the supply chain
Cut: the harness-by-harness table, model list, estate totals (machines, repos).
Board Report
"One posture number — and which part moved it?"
- ·Composite + delta + spark — the single number the board retains
- ·Three source registers — each with its own state — the composite decomposed, one line per report
- ·Packet id — the board cadence — quarterly, filed
Cut: the headline sentence — it restates the three source rows verbatim; redundancy earns no slot.
Briefing — hero number, alarm lines, quiet register
The headline + lede merge from the last iteration, with the greek replaced by the audited slots: score and spark carry state and movement, alarm lines carry the items someone must address (with an owner where the fixture knows one), the register carries the figures that explain the score. Weight marks the watch line; tone appears once.
✗CC6.1 — secrets.yaml read in an agent session · sofia
✗SI-4 — Adapter error on one machine — events stopped for 2 agents · kai
▲5 shadow installs · 4 personal accounts — newest Windsurf on chen-mba
Notes: roughly 3× the information of the greeked cards, and it reads calmer — real figures in a quiet register beat decorative texture. The document identity survives on typography alone (the paper, the ruled masthead, the filing footer); the greek was costume. Cost: the card grew 168 → 200px; below that the registers wrap.
Ledger — same content, no hero
The control question: is the big numeral even earning its 24px? Identical audited content as a uniform ledger — score is just the first row in bold, alarms are toned rows, nothing is enlarged. Tests whether hierarchy-by-size beats hierarchy-by-weight alone.
Notes: the most honest pixels on the page — every row a fact — and the auditor's aesthetic. But the four-up scan flattens: without the hero numeral you must read all four cards to rank them, and the landing's first job is triage.
Provisional verdict — awaiting Mark's pick
- The greek is gone for good. Real content carries the document identity by itself; nothing in either composition gestures at a report — it is one, compressed.
- Slot selection is the design act. The only editorial freedom left is which derived figures earn a place — recorded per report in the audit. Everything on a card traces to report-fixtures.ts; the board headline lost its slot because it restates the three source rows.
- Two emphasis channels, one alarm. Weight now marks "most informative" (SLA's Q3 throughput — 12 found, 6 resolved — is debt building without a single breach), while color remains the single genuine-tone device. Urgency and importance are no longer the same channel.
- Lean: briefing. The hero numeral is what makes four cards scannable as a dashboard; the ledger is the right form one level down (hover, or the snapshot shelf). If the briefing wins, it replaces the merged treatment in the fleet-overview-v3 spine.
- Iterated: the attention device gets its own bake-off in report-card-minidocs-v3 — ten ways to call out the areas of concern on these cards.