Two independent axes that solve feature-saturation and trust: depth (how much you see) and evidence (how much to trust it). Reference for the metric-registry rollout.
The segmented control and tiering mechanism already exist — .mode-bar / .mode-btn drive body[data-mode], and [data-tier] attributes gate visibility. The registry's depth field simply emits these existing attributes. Default Core; the expert opts up. Nothing is invented here — it's documented so the rollout reuses it.
click to preview tier filtering below ↓
Core
depth:'basic' → (no data-tier)
Readiness hero, 4–5 headline KPIs, one plain-language line. The default a normal user meets.
Experimental composites, DFA/SampEn, the full dump. The "everything" tier = today's UI.
02 — CLINICAL / EPISTEMIC HIERARCHY
Evidence badges ◆ net-new visual language
A metric's trustworthiness, encoded in a fill ladder on one neutral META ink — never a hue, because green/amber/red already mean status and teal/blue mean brand. Fill goes solid → half → hollow → dashed as confidence drops: an intuitive gradient that survives color-blindness and never collides with status color.
Measured
bullseye
Direct sensor reading — the strongest evidence, not computed. A raw statistic of the recorded signal (mean / min / max / nadir / duration / coverage).
Mean SpO₂ · Mean HR · duration · % recorded
Validated
solid fill
Established, externally validated, clinically meaningful derived metric. The gold standard of computed measures.
ODI-4 · T90 · hypoxic burden · RMSSD
Emerging
half fill
Published and promising, but less standardized or device-dependent. Use with context.
DFA α1 · CVHR screen · augmentation idx
Experimental
hollow ring
Plausible, node-computed composites without external validation. Directional only.
NSI · sleep-stability · SBII
Heuristic
dashed ring
Convenience estimates / population proxies. Useful as a trend, not a measurement.
ANS-age · VO₂max est · BP projection
EvidenceMeasuredValidatedEmergingExperimentalHeuristicone strip per view · cite on hover
03 — APPLIED TO OXYDEX
Both axes, composed
Status hue lives in the value + top-bar; evidence lives in a quiet badge in the card's bottom-right corner (inline + leading in dense tables; after the value on chips). They never touch. Toggle the depth selector above — KPIs gated above the current tier dim out. This is exactly how a node renders once the registry drives it.
The common trap is collapsing complexity and trust into one slider. They're independent: a Core metric can be Validated (ODI-4); a Research metric is usually Experimental — but not always. Keeping two fields is what lets an expert in Research mode still instantly see what's gold vs speculative.
depth ↓ / evidence →
Measured ◉
Validated ●
Emerging ◐
Experimental ○
Heuristic ◌
Core
Mean SpO₂
ODI-4
Mean-HR dip
— rare —
Readiness
Advanced
Mean HR · min HR
Hypoxic burden
CVHR · DFA α1
SBII
VO₂max est
Research
— rare —
— rare —
SampEn
NSI · stability
BP projection
◆
Both fields live in the registry and travel with the metric — into the screen, the crossnight envelope, and the Integrator longitudinal view. One declaration, three surfaces, zero drift. The badge renderer (MetricRegistry.badge) is shared so a Validated dot looks identical in every node.
Encoding rules (carry into the brief verbatim):
1 — Evidence uses fill/shape on ONE neutral ink (--ev-ink #aab8cc); never green/amber/red (status) or teal/blue (brand).
2 — Fill ladder = confidence ladder: bullseye ◉ (measured) → solid ● → half ◐ → hollow ○ → dashed ◌.
3 — Status hue stays on the value + KPI top-bar; evidence badge sits in the card's bottom-right corner (inline + leading in dense tables; after the value on chips). They never share a channel.
4 — depth → existing [data-tier] + body[data-mode]; evidence → new MetricRegistry.badge(). Two fields, one registry entry.
5 — Legend: one strip per view; cite string on hover/title of each badge.