Evidence an agent can hand on.

Know what an endpoint declares before your wallet pays it.

Probe it once, the way a wallet or a directory does: the 402, the accepts, the price, the Bazaar extension, the declared method, Coinbase's validator. Eleven checks, a signed record. We probe the market's top hundred every day: the weather.

curl -s -X POST 'https://api.s2ar.dev/v1/x402/inspect' \
  -H "Authorization: Bearer $S2AR_KEY" \
  -H 'content-type: application/json' \
  -d '{"resource": "https://api.s2ar.dev/x402/v1/assert", "method": "GET"}'

→ HTTP 200 · 496 ms · list price $0.005 · record a2b099cddde1…

{
  "verdict": "pass",
  "passed": 11,
  "checks": 11,
  "summary.price_usd": 0.005,
  "summary.networks": [
    "base"
  ],
  "summary.declared_method": "GET",
  "record_sha256": "a2b099cddde1ee3304aadcc6c5e567a341381eca1f9c802242a445fd77a66b89",
  "charged_usd": 0.0
}

Read the function, not the 7,700-line file.

Map a repository by imports, skim a file's skeleton, pull one symbol's source, search by name or by meaning. Ten languages, tree-sitter, a receipt on every answer. /v1/sym/map $0.01 · /ls /read /find $0.005 · /where $0.01. Free and local too: cargo install starlab-sym.

curl -s -X POST 'https://api.s2ar.dev/v1/sym/read' \
  -H "Authorization: Bearer $S2AR_KEY" \
  -H 'content-type: application/json' \
  -d '{"repo": "BurntSushi/ripgrep", "ref": "14.1.1", "file": "crates/core/flags/defs.rs", "symbol": "Glob as Flag"}'

→ HTTP 200 · 190 ms · list price $0.005 · receipt sha256:598d5e19ec2b…

{
  "tokens_est": 565,
  "receipt.evidence_hash": "sha256:598d5e19ec2bace68778add369e74f054ca7dc43c2583e0e1e0c4cbefa58a832",
  "charged_usd": 0.0
}

impl Glob as Flag (crates/core/flags/defs.rs, lines 2459-2512)
 2459	impl Flag for Glob {
 2460	    fn is_switch(&self) -> bool {
 2461	        false
 2462	    }
 2463	    fn name_short(&self) -> Option<u8> {
 2464	        Some(b'g')
 2465	    }
 2466	    fn name_long(&self) -> &'static str {
 2467	        "glob"
… 45 more lines

Grade the haiku by its syllables, not by opinion.

Name the form and send the text. Meter comes from a versioned dictionary, then rhyme, then lines, then the gates: an exact line count, a content floor, no credit for identical end words. Same text, same answer, every time; each gate names its evidence; the answer is a signed record. GET /v1/score/forms publishes the forms and the policy, free.

curl -s -X POST 'https://api.s2ar.dev/v1/score' \
  -H "Authorization: Bearer $S2AR_KEY" \
  -H 'content-type: application/json' \
  -d '{"form": "haiku", "text": "An old silent pond\nA frog jumps into the pond—\nSplash! Silence again."}'

→ HTTP 200 · 112 ms · list price $0.005 · record 8df2a8705a89…

{
  "accepted": true,
  "composite": 1.0,
  "gates.0.gate": "line_count",
  "gates.0.ok": true,
  "gates.0.evidence": {
    "got": 3,
    "want": 3
  },
  "gates.1.gate": "content",
  "gates.1.ok": true,
  "record.outcome": "pass",
  "record_sha256": "8df2a8705a89d7f8b0d855deb31c26b484978f73d8832809285a3677c8a4baa8",
  "charged_usd": 0.0
}

Shrink the screenshot. Prove the model still reads it.

Send the image. Get back the smallest version a vision model reads the same way, measured by a deterministic witness and sealed per model. GET /v1/verify/{sha} re-checks the record free, forever. The calibrations behind the band: the lab.

curl -s -X GET 'https://api.s2ar.dev/v1/certify?image_url=https%3A%2F%2Fs2ar.dev%2Fexamples%2Fmedia%2Finvoice.png&probe=ocr' \
  -H "Authorization: Bearer $S2AR_KEY"

→ HTTP 200 · 3914 ms · list price $0.01 · record 924aed0a44f0…

{
  "outcome": "pass",
  "verdict": "isomorphic",
  "witnesses.0.name": "ocr",
  "payload.tokens.anthropic": {
    "provider": "anthropic",
    "before": 800,
    "after": 220,
    "saved": 580,
    "ratio": 0.725
  },
  "payload.fidelity.drift": 0.0,
  "payload.fidelity.tolerance": 0.05,
  "subject.1.w": 524,
  "subject.1.h": 314,
  "subject.1.bytes": 5966,
  "record_sha256": "924aed0a44f0e6754350a2cb7e02c8d640dbfd01057843172ea77259aa31df00",
  "charged_usd": 0.0
}

Diff the golden frame against today's. Find what changed.

Send two images. Three deterministic witnesses and no model: a luma difference hash for structure, an 8×8 colour signature for the palette regression a luma hash is blind to, the change in gradient energy for texture, each against a threshold you set. The record names both images by hash, never by pixels. The frames here are a StarLab render and the same render with red and blue swapped.

curl -s -X GET 'https://api.s2ar.dev/v1/diff?a=https%3A%2F%2Fs2ar.dev%2Fexamples%2Fmedia%2Fscene.png&b=https%3A%2F%2Fs2ar.dev%2Fexamples%2Fmedia%2Fscene-rb.png' \
  -H "Authorization: Bearer $S2AR_KEY"

→ HTTP 200 · 633 ms · list price $0.005 · record 74318bd1337c…

{
  "verdict": "regression",
  "outcome": "fail",
  "passed": 2,
  "checks": 3,
  "regressed": [
    "color_sig"
  ],
  "distances": {
    "dhash": 0.0625,
    "color_sig": 0.060417,
    "grad": 0.041722
  },
  "a.w": 729,
  "a.h": 381,
  "record_sha256": "74318bd1337cee42e3a8dd57086ea1bbdadab186d802bcec5bfd681a59d153ea",
  "charged_usd": 0.0
}