Five design languages rendering the same real Intelligence Objects
(10 creative-forensics facts + 3 whats-working opinions from the live nirvana-commons substrate,
plus 8 OTTO ROAS session extractions). Same information system, different expression.
Goal: pick a winner, hybridize the best elements. Scores are bossman's review-pass ratings
against the rubric (1–5); read the full reasoning in each variation's design-rationale footer.
Working lab, not a launch surface. Data is real; the substrate is the source of truth.
The trust ceiling. Scope-of-extract table (runs with verdicts, dual agents, filters, freshness) is the most complete lineage header of the set; OWID-style 'About this data' details per IO. Reads institutional — which is the point and the risk.
Strengths
Most complete scope/lineage table (runs + verdicts + both agents)
Per-IO 'About this data' expander (OWID pattern)
Index tables with stable anchors + ICD bands
Null-field muting keeps verbatim reports readable
Weaknesses
Long single page; low information density
Institutional register can read bureaucratic for a product
Breadcrumb parents are placeholders until catalog pages exist
Confidence-as-language done right. The numbered Key Findings deck (summaries → anchors) is the best human entry point of the set, and the banner scope bar stacks perfectly on mobile. The briefing genre carries authority but is the least web-native.
Strengths
Key Findings deck — best cold-start reading order
Confidence in ICD language first, number second, color-banded
Full-width scope band with honesty note as first-class element
Footnote-style evidence lines per claim
Weaknesses
Reports default-open make a longread
Serif-heavy register least familiar to dev/product audiences
Mid-density: neither scan-fast nor magazine-comfortable
The operator throughput winner. One aligned grid row per IO with tabular numerals makes cross-IO comparison instant; the terminal status bar compresses full lineage into six lines; the session table shows real ROAS/spend/clicks. Continuity with otto-boards.
Least 'public product' of the set — more internal tool
V4 · LEDGERdeveloper docs / API resource (Stripe/Context7)
The agents-first product surface. Four machine doors (MCP config, data island, anchors, quotable caveats) above the fold with copy buttons; every IO is a resource with a human panel beside its canonical GET address + raw record. Light+dark both pass AA. The strongest launch-genre candidate.
Strengths
Agent doors as first-class UI — the positioning made visible
Human panel + machine panel per IO (trust through transparency)
uuid sidebar TOC; light+dark; verified at 5 widths
Best extensibility to catalog/detail/home
Weaknesses
Two-panel layout = fewer IOs per screen than a grid
Machine panels are noise for pure-business readers
The readability ceiling and the best confidence UI of the set: labeled score bars with the four sub-dimensions as small multiples, rendering nulls honestly as 'not assessed'. As a flagship-analysis reading surface it wins; as a working catalog it is too slow.
Strengths
Confidence block: overall bar + sub-dimension small multiples with honest 'not assessed'
OWID source strip + 'Learn more about this data' overlay
Editorial exhibit treatment of creatives is the most engaging human read
Warm paper + serif display = highest polish impression
Weaknesses
Lowest density; 13 open reports = longread
Magazine register can undercut operator credibility