health-report-dashboard

name: health-report-dashboard description: Build a self-contained, Power-BI/Tableau-style HTML portfolio report from partner-health data — KPI cards (NRR/GRR/avg-health/red-count/renewal-risk), a health-band donut, a 12-week trend line, a peer-cohort range chart, score-component drill-downs, red-flag surfacing, and a sortable/filterable per-partner table. Used by learning-analytics-analyst (primary) + edtech-partner-success-manager. Ships with a ready-to-open demo (sample data) AND regenerates from real data — replace bi-report/data.json (same shape) and re-run the generator. Invoke when someone asks for a partner-health dashboard, a "report I can show leadership", a QBR data view, or "how's my whole book doing in one picture".

Skill: health-report-dashboard

Purpose: turn the partner-health numbers the PSM team already tracks into a single, self-contained HTML report that looks like an advanced Power BI / Tableau page — but needs no BI license, no server, and no internet. It opens by double-clicking the file. This is the visual companion to the health-score-dashboard spec and the partner-health-scoring skill (which defines the score itself).

It exists in two halves, so you get both a thing to look at now and a real tool:

1. Demo (ships in the plugin). plugins/edtech-partner-success/bi-report/data.json holds realistic synthetic data; report.html is the rendered demo. Open it to see the whole layout immediately.
2. Generator (the capability). Replace data.json with a real export in the same shape and re-run — the report rebuilds from your numbers.

When to use

• "Show me how my whole book is doing" / "I need a portfolio health view."
• Leadership / EBR wants a one-glance report of every partner's health, trend, and risk.
• QBR prep — drop one partner's drill-down (components + red flags + dates) into the deck.
• You have partner-health numbers in a spreadsheet and want a clean, shareable page.

What the report shows

Every metric carries a "?" explainer in everyday words, and the whole report reads at roughly a 5th-grade level — the jargon (NRR, GRR, IQR) is the small print.

How to (re)build it

# rebuild every plugin report that has bi-report/data.json
python3 scripts/generate-bi-report.py

# just this plugin
python3 scripts/generate-bi-report.py --plugin edtech-partner-success

# CI / pre-commit: fail if the committed report.html is stale
python3 scripts/generate-bi-report.py --check

The generator is plugin-agnostic: any plugin that ships a bi-report/data.json gets a report.html. The charts are hand-rolled inline SVG and the sort/filter/ drill-down is vanilla JS, so the output stays fully self-contained (no CDN, no charting library) — consistent with the rest of the marketplace's static surfaces.

The data shape (what to export from real systems)

bi-report/data.json (see the shipped sample for a complete, filled example):

{
  "report":  { "title": "...", "subtitle": "...", "refreshed": "YYYY-MM-DD",
               "synthetic": true, "owner": "..." },     // set synthetic:false for real data
  "bands":   { "green": [70,100], "yellow": [50,69], "red": [0,49] },
  "components": [ { "key": "adoption", "name": "Adoption depth", "weight": 25,
                    "half_life_days": 90, "plain": "everyday explanation" }, ... ],
  "kpis":    [ { "key": "nrr", "label": "Net revenue kept + grown", "short": "NRR",
                 "value": 108, "unit": "%", "delta": 2.0, "good": "up",
                 "plain": "everyday explanation" }, ... ],
  "cohort":  { "label": "K-12 · Enterprise · Year 2+", "median": 68, "p25": 54, "p75": 81, "size": 11 },
  "trend_weeks": ["12 wk ago", ..., "This wk"],
  "portfolio_trend": [62, 63, ..., 68],                  // 12 portfolio-average scores
  "partners": [ {
    "name": "Riverside Unified", "segment": "k12", "psm": "Dana L.",
    "score": 81, "delta": 3, "band": "green",            // band optional — derived if omitted
    "components": { "adoption": 84, "touchpoint": 88, ... },
    "spark": [72, 73, ..., 81],                          // last 12 weeks of the composite
    "flags": ["plain-language red-flag sentence", ...],  // empty = none
    "play": "Maintain", "last_touch": "YYYY-MM-DD",
    "next_qbr": "YYYY-MM-DD", "renewal": "YYYY-MM-DD" } ]
}

The component keys/weights/half-lives mirror partner-health-scoring and the health-score-dashboard spec; the red-flag sentences mirror that spec's six triggers; the metric vocabulary mirrors psm-metrics-glossary.md.

Recolouring the report (theme override)

Two levels of colour control, both from data.json — no code:

1. Per-element — every chart element takes a color: a token name (teal, gold, ok, warn, danger, muted) or a raw value ("#1f8ded"). Status bands map to ok/warn/danger.
2. Whole-report theme — an optional top-level theme block re-skins the entire report (canvas, cards, accent, status colours) by overriding the shared --rc-* tokens. Absent ⇒ the marketplace defaults (Intercom cool-grey
   • teal). It's injected as a :root block after the inlined tokens, so it wins; values are validated to a safe colour grammar (hex / rgb() / hsl() / named) so a data file can never inject arbitrary CSS.

"theme": {
  "accent":  "#1f8ded",   // links, trend line, the chosen-accent chart colour
  "bg":      "#f5f6f8",   // page canvas
  "surface": "#ffffff",   // cards
  "border":  "#e3e6ea",
  "text":    "#1c1f23",   "muted": "#5b6166",   "faint": "#8e959c",
  "ok": "#1a7f4b", "warn": "#b7791f", "danger": "#d14343",
  "ink": "#1c1f23"        // charcoal primary-button fill
}

Keys are all optional (set only what you want to change). A11y: if you pick a new accent, verify it still passes WCAG AA on your bg before shipping — the accent carries links and text, not just decoration.

FERPA / privacy (load-bearing)

• Synthetic identifiers only in anything that ships or is shared. The demo data.json uses invented districts and PSM initials — never real partner or student data. Keep report.html out of any repo that isn't access-controlled.
• The report is partner-level, never student-level. Do not add student PII columns. Student-level analysis routes through security-reviewer first (plugin constitution §2).
• A real export should set report.synthetic to false and live next to the data it came from, not in a public location.

Adapting this to other plugins ("apply where applicable")

Any plugin with quantitative, per-entity data can ship a bi-report/data.json and get a report for free. Natural candidates and the data that would feed them:

For a domain whose shape differs a lot from partner-health (e.g. a P&L waterfall), either extend the generator's chart helpers or add a domain-specific renderer that follows the same self-contained, token-inlined pattern.

Hand-off

• Score design / weights / decay → partner-health-scoring (learning-analytics-analyst).
• Turning a drill-down into a QBR slide → qbr-composition (qbr-composer).
• Which play a red flag triggers → the recovery / renewal / expansion play skills (success-playbook-designer).
• Anything student-level → ravenclaude-core/security-reviewer (mandatory).

References

• Spec: templates/health-score-dashboard.md
• Metric definitions: knowledge/psm-metrics-glossary.md
• Generator: scripts/generate-bi-report.py
• Demo data + output: bi-report/data.json → report.html