California Highway Traffic Speeds (MPH per Highway)

Purpose

Return the current real-time vehicle speed (MPH) for every California highway covered by ABC7 San Francisco's traffic map — broken down per road, per direction, per sensor — along with active incidents and traffic camera references. The data is sourced from the Sigalert / Total Traffic Network (TTN) backend that ABC7 embeds via the partner=kgo-tv iframe at abc7news.com/traffic/. Read-only; never posts or alters data.

When to Use

• "What's the current speed on I-580 eastbound?"
• "Which Bay Area freeways are running below 30 MPH right now?"
• A commute / routing agent that needs a regional MPH snapshot to score alternatives.
• A traffic-monitoring agent producing alerts when a highway segment drops below a threshold.
• Any task framed as "MPH for each highway" / "current freeway speeds" on a California ABC7 affiliate's traffic page.

Workflow

The ABC7 traffic page (abc7news.com/traffic/) is a thin shell that embeds an iframe to sigalert.com/Custom/Map.asp?partner=kgo-tv. The map is rendered client-side from three plain JSON endpoints with no auth, no cookies, no anti-bot, no proxy requirement. Lead with the JSON API — it returns the entire region's MPH-per-sensor table in a single GET. Browser-driving the embedded map is a ~100× slower fallback (panning + clicking each pin to read its tooltip), and browse snapshot of the rendered map exposes none of the speed values (they're drawn into canvas tiles, not DOM text).

Three regions are exposed by the same backend: NoCal (San Francisco Bay Area — what partner=kgo-tv shows), SoCal (Los Angeles — same backend serves KABC-TV's affiliate equivalent), and CenCal (Central Valley). Each region is independent — fetch one region's bundle to answer questions about it.

1. Resolve the current data path (cache-busting indirection)

GET https://www.sigalert.com/Data/{Region}/path.json

Returns:

{
  "path": "NoCal/3~j",
  "cacheBuster": 31811789,
  "updateTime": 1779168624001,
  "version": 29400181
}

path rotates every ~30s (the ~j / ~k / ~l suffix is the slot; 1~j, 2~j, 3~j form a small rotation). You must re-fetch path.json before each live-data pull — pinning a stale path returns the snapshot from the moment it was current, not "now".

2. Fetch the live speed/incident/camera bundle

GET https://www.sigalert.com/Data/{path}/{Region}Data.json?cb={cacheBuster}

E.g. https://www.sigalert.com/Data/NoCal/3~j/NoCalData.json?cb=31811789. Returns:

{
  "speeds":    [[mph, hovMph, items[]], ...],   // positional, aligned to sensorNames[]
  "incidents": [[severity, id, timeDisplay, location, type, ...], ...],
  "cameras":   [[cameraId, sensorIdx, roadId, ?, label, label, imageUrl, ...], ...]
}

• speeds[i][0] — current MPH at sensor i (integer; null if no reading).
• speeds[i][1] — current HOV-lane MPH (often null).
• speeds[i][2] — array of co-located items (camera/incident references).

The Content-Type header is misleadingly text/javascript but the body is plain JSON — parse with JSON.parse.

3. Fetch the static road metadata (cache this; changes rarely)

GET https://cdn-static.sigalert.com/240/Zip/RegionInfo/{Region}Static.json

Returns:

{
  "sensorNames":     ["Gorda Mt Rd (4.6 miles before)", "Gorda Mt Rd", ...],
  "sensorPositions": [[x, y, ...], ...],
  "roads":           { "100011": ["1", 0, 0, [[0,92,65],[93,100,35],...]], ... },
  "roadSections":    [[100011, "North", "1", 0, 332], [100012, "South", "1", 333, 667], ...]
}

• sensorNames[i] is aligned 1:1 with speeds[i] from step 2.
• roads[id] = [displayName, _, _, speedLimitSegments] — displayName is the bare route number ("1", "101", "580"); see Gotchas for the missing road-type prefix.
• roadSections[] is the join key — [roadId, direction, displayName, firstSensorIdx, lastSensorIdx] ties contiguous sensor ranges to a single direction of travel on a single highway. The 4th and 5th values are inclusive indices into sensorNames[] / speeds[].

4. Join + aggregate to per-highway MPH

const byRoad = new Map();
for (const [roadId, dir, name, first, last] of stat.roadSections) {
  const rec = byRoad.get(roadId) || { roadId, name, sections: [] };
  const mphValues = [];
  for (let i = first; i <= last; i++) {
    const s = data.speeds[i];
    if (s && typeof s[0] === "number") mphValues.push(s[0]);
  }
  if (mphValues.length) {
    rec.sections.push({
      direction: dir,                                      // "North" | "South" | "East" | "West"
      sensorRange: [first, last],
      sensors: mphValues.length,
      avgMph: +(mphValues.reduce((a,b)=>a+b,0) / mphValues.length).toFixed(1),
      minMph: Math.min(...mphValues),
      maxMph: Math.max(...mphValues),
      perSensor: mphValues                                  // optional — same order as sensor indices
    });
  }
  byRoad.set(roadId, rec);
}

Yields ~74 distinct highway-direction entries in NoCal, each with current MPH stats.

5. (Optional) Cross-reference incidents to enrich highway entries

incidents[i][3] is a free-text location string like "CA-1 South at Scott Creek" or "US-101 North at Marsh Rd" — the only place in the dataset where the road type prefix (I-, US-, CA-) is present. Match by the bare route number ("1", "101") plus direction substring to attach incident lists to your per-highway rows. incidents[i][2] is the human-readable time ("1:01 PM") and incidents[i][8]/[9] are ISO timestamps (creation / last-update).

Browser fallback

Only reach for the browser if the JSON endpoints start returning 4xx (none observed in 4 successive fetches across both NoCal and SoCal). The fallback is:

sid=$(browse cloud sessions create --keep-alive | node -e "...id...")
export BROWSE_SESSION="$sid"
browse open "https://www.sigalert.com/Custom/Map.asp?partner=kgo-tv&sp=p&th=blue&z=2" --remote
browse eval --remote "performance.getEntriesByType('resource').map(e=>e.name).filter(n=>/NoCalData|NoCalStatic|path\\.json/.test(n))"

Then read the same three JSON URLs from the resource log. Stealth (--verified) and residential proxies (--proxies) are not required — the embedded iframe loads fine with a bare session and the JSON CDN has no anti-bot layer.

Site-Specific Gotchas

• abc7news.com/traffic/ itself contains no speed values. The page is an article shell — the entire traffic display is one nested iframe (www.sigalert.com/Custom/Map.asp?partner=kgo-tv). Scraping the article's DOM or running text extraction on abc7news.com/traffic/ returns only the "TRAFFIC NEWS" sidebar headlines, not MPH. Don't waste a turn there — go straight to the Sigalert endpoints.
• The Sigalert map renders speeds into canvas tiles, not DOM text. browse snapshot of the loaded iframe returns ~20 a11y refs (UI chrome only) and zero MPH values. Clicking individual pushpins exposes a single sensor at a time via a tooltip — extremely expensive vs. the bulk JSON pull.
• path.json rotates roughly every 30 seconds. The path field cycles through 3–4 slots (1~j, 2~j, 3~j, ...) within the current build version (~j). Always re-fetch path.json before each {Region}Data.json pull; do not hard-code a path. If you get a stale-looking response, the updateTime ms-epoch field will tell you exactly when that snapshot was published.
• cb cache-buster is mandatory. Without ?cb=... matching the value from path.json, the CDN may serve a cached older version. The cache-buster monotonically increases; treat it as opaque.
• Content-Type: text/javascript on JSON responses is a JSONP-era artifact. The body is valid JSON — JSON.parse it directly. Do not eval.
• browse cloud fetch and base64 encoding. browse cloud fetch returns small JSON responses as raw text in the content field, but when written to --output it sometimes base64-encodes the file body (observed for the ~1 MB JS bundle; not observed for the 408 KB NoCalData.json). If your output looks like base64, decode with base64 -d. The 200 statusCode and sizeBytes fields are reliable either way.
• Road displayName strips the road-type prefix. roads[id][0] is just "1" / "101" / "580" / "880" — there's no "I-" / "US-" / "CA-" qualifier. To know whether route 1 is "CA-1 / PCH" vs. "Highway 1", inspect incidents[].location strings (they include the prefix) or fall back to public route-classification data. Within California: 5/15/40/80/205/210/215/238/280/380/405/505/580/605/680/710/780/805/880/980 are Interstates; 50/97/101/199/395 are US routes; everything else (1, 4, 12, 13, 17, 24, 25, 33, 35, 37, 41, 46, 52, 60, 67, 84, 85, 87, 92, 99, 109, 113, 116, 121, 128, 129, 132, 152, 156, 160, 162, 174, 175, 183, 220, 237, 242, 280-prefixed mins, 4-digit local) is a CA- state route.
• Per-direction sections. roadSections lists each highway as 2 (or sometimes more — for highways that turn) rows: one per direction-of-travel. Aggregating sections together loses the directional asymmetry that's usually the whole point of a traffic query (e.g. northbound 101 at 25 MPH vs. southbound at 60 MPH at 5 PM). Keep the rows separate in output.
• HOV lane speeds available separately. speeds[i][1] is the HOV/carpool-lane MPH and is often higher than the general-purpose lanes during peak hours. It's frequently null outside HOV-equipped segments. Decide upfront whether your output reports general-purpose only, HOV separately, or merged.
• Three regions on one backend. Replace NoCal with SoCal (Greater Los Angeles) or CenCal (Central Valley) in all three URLs to query other regions. Same schema, same indirection, same JSON shape. Other ABC affiliates (KABC-TV in LA, KFSN-TV in Fresno) use the same partner-keyed iframe with a different region prefilled, but the underlying data API is region-keyed not partner-keyed — so you can directly query SoCal data even from an "abc7news.com" task context if needed.
• Custom/Map.asp partner shells. partner=kgo-tv is hard-wired to NoCal; partner=kabc-tv, partner=kfsn-tv, etc. exist but return HTTP 500 when fetched through browse cloud fetch (the .asp server-side render checks Referer/User-Agent). This does not affect the JSON data endpoints, which are independent and need no partner key. Don't try to use the partner= URL as a data source.
• Sandboxed curl will not resolve www.sigalert.com. The Vercel Sandbox has restricted DNS — curl fails with "Could not resolve host". Use browse cloud fetch (which proxies via Browserbase's network) for all three JSON URLs. From an agent runtime that has plain internet (a hosted Lambda, a user's laptop), vanilla fetch / curl works fine — no auth, no CORS, no rate limit observed.
• Region naming. Probed region tokens: NoCal ✓, SoCal ✓, CenCal ✓, Sacto ✗ (404), SD / SDiego ✗ (404). San Diego is rolled into SoCal. Sacramento is rolled into NoCal.

Expected Output

A single snapshot for one region, returned as JSON. Shape:

{
  "region": "NoCal",
  "regionLabel": "San Francisco Bay Area",
  "snapshotTime": "2026-05-19T05:30:24.001Z",
  "version": 29400181,
  "highways": [
    {
      "roadId": 105803,
      "name": "580",
      "direction": "East",
      "sensors": 24,
      "avgMph": 58.4,
      "minMph": 22,
      "maxMph": 71,
      "freeFlowSpeedLimit": 65,
      "congested": false
    },
    {
      "roadId": 101011,
      "name": "101",
      "direction": "North",
      "sensors": 67,
      "avgMph": 27.3,
      "minMph": 9,
      "maxMph": 58,
      "freeFlowSpeedLimit": 65,
      "congested": true
    }
  ],
  "incidents": [
    {
      "id": 48449051,
      "severity": 519,
      "time": "1:01 PM",
      "createdAt": "2026-05-18T20:01:06",
      "updatedAt": "2026-05-18T23:09:08",
      "location": "CA-1 South at Scott Creek",
      "type": "Accident. Shoulder blocked",
      "road": "1",
      "direction": "South"
    }
  ],
  "cameraCount": 819
}

Optional richer shape — per-sensor MPH for downstream segment-level analysis:

{
  "region": "NoCal",
  "highways": [
    {
      "roadId": 108801,
      "name": "880",
      "direction": "North",
      "sections": [
        { "sensorIndex": 4601, "name": "Marina Blvd", "mph": 62, "hovMph": 65 },
        { "sensorIndex": 4602, "name": "23rd Ave",    "mph": 48, "hovMph": 58 }
      ]
    }
  ]
}

Failure / edge shapes:

{ "success": false, "reason": "region_not_supported", "region": "SD", "supported": ["NoCal", "SoCal", "CenCal"] }

{ "success": false, "reason": "data_endpoint_unavailable", "lastPathFetchStatus": 503 }

In observed runs (4 successive fetches across NoCal + SoCal), every endpoint returned HTTP 200 with the expected schema; no anti-bot, captcha, or rate-limit response was encountered.