DEV Community

Cover image for I asked for fire evidence. The API returned 152 thermal detections — and refused to draw a perimeter.
Alex
Alex

Posted on • Originally published at atlas.modelmarket.dev

I asked for fire evidence. The API returned 152 thermal detections — and refused to draw a perimeter.

#ai #webdev #showdev #opensource

ATLAS — fire + weather layers on planetary glass

Most wildfire dashboards imply a shape the sensor never gave.

A red polygon looks like a perimeter. FIRMS never sold you a perimeter. It sold you thermal anomalies — brightness, confidence, a point — and a lot of product UIs quietly upgrade that into “the fire.”

We build agent rails where a screenshot is not evidence. So we shipped a vertical product that stays honest:

atlas.fire.weather@v1 — FIRMS detections (and EFFIS when present) + bounded LIVE weather context + a signed content receipt.

Not a burn scar. Not a forecast. Not a parametric quote.

The thesis (one sentence)

If your fire map invents a perimeter, it is cinema — not evidence.

What we actually returned (this run)

Public demo bbox (roughly West Coast US). One live invoke of atlas.fire.weather@v1:

Field Value
LIVE FIRMS detections 152
Brightest 367 K near 37.66, −121.38 (confidence 90)
EFFIS polygons 0 in this frame
Weather LIVE nearest in range — SFO Airport, 15 °C
Price (public demo) ~$0.08

Hotspot list — LIVE FIRMS brightness and FRP
Count the detections. Do not invent a line around them.

Weather is on a leash — only inside a declared distance bound, not a global vibe:

Bounded weather pairing — nearest LIVE station
Dual attribution: thermal list + weather list. Same receipt discipline.

Open the glass

  1. atlas.modelmarket.dev — layers fire + weather, English UI
  2. Or the idea path: use.modelmarket.dev → Fire + Weather Evidence Snapshot
  3. Hub SKU: atlas.fire.weather@v1

Fire-only layer — thermal glass without the weather overlay
Humans get honesty on the map. Agents buy the artifact behind it.

Curl shape (ATLAS invoke):

curl -sS -X POST https://atlas.modelmarket.dev/ai-market/v2/invoke \
  -H 'Content-Type: application/json' \
  -d '{"capability_id":"atlas.fire.weather@v1","input":{"west":-125,"south":32,"east":-114,"north":42,"limit":24}}'
Enter fullscreen mode Exit fullscreen mode

What it is not (say it out loud)

We print the limits on the artifact on purpose:

Limitations — detections are not perimeters

  • FIRMS reports satellite thermal anomalies — not confirmed incident perimeters.
  • EFFIS current-fire polygons are a separate Copernicus product (CC BY 4.0) when present — not FIRMS pixels.
  • Weather / air context only within declared distance bounds.
  • Independent operational validation is still required.

That paragraph is the product. Dashboards that skip it are lying by omission.

Evidence with a receipt

The snapshot carries a content receipt — digest, signature, public key. Same grammar as our portable proofs post: the object can leave the tab.

Content receipt — signed digest on the fire-weather snapshot
Attribution ≠ physical truth. A valid receipt proves who signed these claims — not that the sky is on fire.

Sibling SKUs on the same rails: Verified Watchbox (atlas.watchbox.check@v1) for region queries; Fire Glass for the vertical evidence pack.

USE idea — Fire + Weather Evidence Snapshot
Live wedge. Not a slide.

Why this is the showdev version of “AI + wildfire”

Agents are about to buy physical-world context at micro-prices. If the first product they touch upgrades detections into perimeters, we train them to trust theater.

We would rather ship:

  1. cited detections
  2. bounded context
  3. a fail-closed empty box when LIVE is gone
  4. a receipt you can keep

Boring. Citeable. Useful.

Try it

  1. atlas.modelmarket.dev/?layers=fire,weather
  2. use.modelmarket.dev → Fire + Weather path
  3. Repo (MIT): github.com/alexar76/aicom

If it helped, a ⭐ is the softest thank-you. Brutal feedback on the limits copy is better.

—
AIMarket — verifiable provenance · agent economy · crypto optional

Top comments (0)