Web Feature Usage & Prioritization Report#

A standalone, engine-agnostic dataset and report that prioritizes web platform features by real-world usage. Useful for any browser engine project deciding what to implement next.

Quick start#

npm run usage-json     # → data/web-feature-usage.json
npm run usage-report   # → web-feature-usage.html

Data sources#

All sources are treated as parallel signals — no waterfall or hierarchy. Each source tells a different part of the story.

Total: 383 of 415 Baseline Widely Available (BWA) features have usage data from at least one source. 32 features have no data.

Weighted composite scoring methodology#

Philosophy#

• Keep all features in the list — never drop a feature, never drop a data point
• All sources are parallel signals, not a hierarchy — each tells a different part of the story
• Firefox is equally weighted to Chrome — it's an independent browser signal, and Fenix provides our only mobile data
• Annotate confidence, don't filter — experts can evaluate the backing data themselves
• The composite score is one view among many, not the final word

How weights work#

Weights are assigned by measurement class, not per-browser:

• Content analysis (specific): HA custom_metrics with specific observable = 0.35. Ground truth — the feature is actually on the page. Site-weighted means every site counts equally.
• Content analysis (generic): HA custom_metrics with generic observable = 0.10. Informative upper bound but noisy — heavily discounted vs specific.
• Chrome runtime (site-weighted): HA blink_features = 0.20. Runtime detection on crawled pages, same denominator as HA. Discounted for detection artifacts.
• Chrome telemetry (traffic-weighted): ChromeStatus = 0.15. Real user telemetry, but traffic-weighted (high-traffic sites dominate).
• Firefox telemetry (desktop, traffic-weighted): Firefox desktop = 0.15. Independent browser signal.
• Firefox telemetry (mobile, traffic-weighted): Firefox Fenix = 0.15. Only mobile signal available. Unique value.

Weight redistribution#

When a source is missing for a feature, its weight is redistributed proportionally among available sources. Example: if a feature only has ChromeStatus + Firefox desktop, the effective weights become ChromeStatus 0.50 and Firefox desktop 0.50.

Blink artifact discount#

Blink features values >45% from canvas/webgl/media-related interfaces receive a 0.5 discount multiplier to account for suspected detection artifacts (suspicious clustering around ~53.5%).

Generic observable recovery#

Features previously dropped by the specificity filter (e.g., subgrid via grid-template-columns) are now included with lower weight (0.10 vs 0.35) and flagged as generic_observable_only. The value represents an informative upper bound.

Confidence levels#

Flags#

Non-exclusive flags that annotate data quality:

How Firefox use-counter data is collected#

API endpoints (public, no auth required):

• Desktop: https://public-data.telemetry.mozilla.org/api/v1/tables/firefox_desktop_derived/firefox_desktop_use_counters/v2/files
• Fenix (Android): https://public-data.telemetry.mozilla.org/api/v1/tables/fenix_derived/fenix_use_counters/v2/files

Each endpoint returns a list of JSON file URLs (one per day). Each file contains records keyed by (submission_date, version_major, country, metric) with pre-computed rate and raw cnt fields.

Metric naming conventions:

• CSS: use.counter.css.page.css_grid_template_columns → strip prefix → strip css_ → replace _ with - → grid-template-columns → BCD css.properties.grid-template-columns
• API: use.counter.page.window_intersectionobserver → strip prefix → strip window_ → case-match to known BCD interface → IntersectionObserver → BCD api.IntersectionObserver

Aggregation: Records are per-(country, version). Global rate = sum(cnt) / sum(use_counter_top_level_content_documents_destroyed) across all countries and versions.

Coverage: ~682 CSS metrics and ~367 API metrics in Firefox telemetry. After mapping to web-features BCD keys, ~82 BWA features have data. CSS mapping is mechanical and reliable. API mapping requires case normalization (lowercase → PascalCase) which is lossy — some APIs don't match.

Limitations:

• Firefox-only user base (~3-4% global market share)
• Traffic-weighted like ChromeStatus (not site-weighted like HTTP Archive)
• No HTML element counters, no CSS selectors, no CSS at-rules
• Fenix files are very large (~765MB uncompressed) — collected via streaming

node data/collect-firefox.mjs                    # collect latest data
node data/collect-firefox.mjs --dry-run           # show file info, no download
node data/collect-firefox.mjs --reprocess         # reprocess cached desktop data
node data/collect-firefox.mjs --desktop-only      # skip Fenix data

How chromestatus data maps to web-features#

The mapping pipeline:

chromestatus.com/data/webfeaturepopularity
  → [{property_name: "Flexbox", day_percentage: 0.826, ...}]
       │
       ▼
AreWeBrowserYet/servo-bcd-scripts/feature-tag.ts
  1. Converts property_name to kebab-case (e.g. "Canvas2d" → "canvas-2d")
  2. Matches against the `web-features` npm package, which defines
     canonical feature-id → BCD key mappings
  3. Attaches BCD collector results (Servo pass/fail per BCD key)
  4. Writes popularityBcdMap.json
       │
       ▼  (copied into this repo)
data/collector-results/popularityBcdMap.json
  { "flexbox": { day_percentage: 0.826, bcd_entries: [...] },
    "avif": null,  ← no chromestatus BCD association
    ... }
       │
       ├──▶ data/build-httparchive-mapping.mjs
       │      Maps BCD keys → HTTP Archive observable paths
       │      Generates BigQuery SQL for the HA crawl queries
       │      Writes: httparchive-mapping.json
       │                    │
       │                    ▼
       │           data/collect-httparchive.mjs  (npm run collect → BigQuery)
       │                    │
       │                    ▼
       │           data/httparchive-results/combined-usage.json
       │
       └──▶ [all sources merged as parallel signals]
                             │
                             ▼
              data/build-usage-json.mjs   → data/web-feature-usage.json
              data/build-usage-report.mjs → web-feature-usage.html
              data/integrate-usage.mjs    → patches index.html

Firefox use-counter pipeline (independent, parallel signal):

  Mozilla public telemetry API
    → data/collect-firefox.mjs (node data/collect-firefox.mjs)
      Downloads desktop + Fenix JSON files
      Streams large files, aggregates per-metric rates
      Maps metric names → BCD keys → web-feature IDs
        │
        ▼
      data/firefox-results/combined-usage.json
        │
        ▼  (loaded by build-usage-json.mjs)
      Desktop and Fenix treated as separate parallel signals
      (firefox_desktop and firefox_fenix in output)

The join is name-based: chromestatus property_name → kebab-case normalization → lookup against web-features package IDs.

Mapping quality & coverage gaps#

32 features with no usage data from any source#

Draft features (7) — not yet shipped, expected gap:

• draft_crash-report-storage-apiinitialize, draft_meta-text-scale, draft_reference-target, draft_wasm-branch-hinting, draft_wasm-custom-descriptors, draft_web-app-manifest-update, draft_web-install-api

Obsolete features (13) — retired chromestatus groupings, expected gap:

• obsolete_canvas-2d, obsolete_canvas-alpha, obsolete_canvas-color-management, obsolete_canvas-desynchronized, obsolete_canvas-element, obsolete_canvas-fill-text, obsolete_canvas-measure-text, obsolete_canvas-text-baselines, obsolete_element-check-visibility, obsolete_hidden-until-found-attribute, obsolete_locale-info-obsoleted-getters, obsolete_popover, obsolete_will-read-frequently

Regular features with no match (12) — the substantive gap:

• avif, webp, http3 — high-usage features; chromestatus likely tracks these under different IDs or has no BCD associations for them
• jpegxl — removed from Chrome; chromestatus entry may lack BCD
• intersection-observer-v2 — v1 matched; v2 may not have a distinct chromestatus entry
• view-transitions-element-scoped, uint8-array-base64-hex, float16-array — newer features, may not have BCD associations in chromestatus yet
• function — ambiguous name, unlikely to map to any single chromestatus feature
• prompt — generic name; beforeinstallprompt matched separately
• text-detect, translation-api — experimental/new APIs

Output format#

data/web-feature-usage.json:

{
  "generated": "2026-02-26T...",
  "description": "Web feature usage with weighted composite scoring...",
  "sources": {
    "ha_custom_metrics": { "weight": 0.35, "feature_count": 107 },
    "ha_custom_metrics_generic": { "weight": 0.10, "feature_count": 61 },
    "ha_blink_features": { "weight": 0.20, "feature_count": 44 },
    "chromestatus": { "weight": 0.15, "feature_count": 383 },
    "firefox_desktop": { "weight": 0.15, "feature_count": 82 },
    "firefox_fenix": { "weight": 0.15, "feature_count": 81 }
  },
  "methodology": { "weighting": "...", "confidence_levels": {...}, "flags": {...} },
  "summary": {
    "total_features": 415,
    "with_composite_score": 383,
    "without_data": 32,
    "tiers": { "above_50_pct": 20, "10_to_50_pct": 103, ... },
    "by_confidence": { "high": 67, "medium": 150, "low": 166, "none": 32 },
    "cross_validation_pairs": 83
  },
  "features": [
    {
      "name": "flexbox",
      "signals": {
        "ha_custom_metrics": { "pct": 0.877, "observable": "align-items", "type": "css_property" },
        "ha_blink_features": null,
        "chromestatus": { "pct": 0.826 },
        "firefox_desktop": { "pct": 0.584, "observable": "css.properties.justify-content" },
        "firefox_fenix": { "pct": 0.743, "observable": "css.properties.justify-content" }
      },
      "recovered": {
        "ha_generic_observable": null
      },
      "composite": {
        "score": 0.787,
        "confidence": "medium",
        "source_count": 4,
        "flags": ["chrome_firefox_divergent", "mobile_divergent"]
      },
      "bcd_key_count": 15,
      "primary_type": "css_property"
    }
  ],
  "cross_validation": [
    {
      "name": "remote-playback",
      "chrome_value": 0.473,
      "chrome_source": "chromestatus",
      "firefox_desktop": 0.006,
      "firefox_fenix": 0.004,
      "delta": 0.467,
      "agreement": "strongly_diverge"
    }
  ]
}

Regenerating#

To update the data:

1. Get fresh popularityBcdMap.json from the AreWeBrowserYet collector pipeline
2. Run HTTP Archive queries: npm run collect (requires BigQuery credentials)
3. Collect Firefox data: npm run collect:firefox (no credentials needed, downloads ~870MB)
4. Reprocess HA results (if re-running without new queries): npm run collect:reprocess
5. Rebuild outputs: npm run usage-json && npm run usage-report