Radar Treasury is Poor to agents.
Discry independently scored how well an AI agent can discover and understand the Radar Treasury API from what’s public — not whether it’s usable. Below: every signal we checked, what’s costing the score, and what to change.
SCORED UNDER RUBRIC 1.2 · A full re-launch under Discry Score 2.5 — a new behavioral instrument, not comparable to these scores — is in progress.
Discovery
45% of score · 40/100Comprehension
55% of score · 71/100What we found
- The docs domain (justradar.readme.io) serves 'User-agent: * Disallow: /' — a robots.txt-respecting agent or crawler is forbidden from reading every documentation page, so discovery depends entirely on the single 'Documentación' link on the Spanish-language marketing site.
- A genuinely good llms.txt exists on the docs domain — API-focused, right-sized (2.2KB), with described links to markdown versions of every reference page — but it points into the same domain robots.txt blocks, undercutting its value.
- An agent consuming the markdown exports of the core payout endpoints (login, payment instruction, payment status, account balance) never learns the HTTP method, URL, or request schema — that content only renders in the JavaScript API explorer. The two newest endpoints (2026 account validation) embed full OpenAPI 3.1 fragments with error examples, showing the fix is already known.
- Error documentation is a bright spot: per-parameter validation tables map each rule to its exact 409/422 error with valid-value guidance (RUT format, SBIF bank codes, amount limits), and webhook failures carry a documented errorType taxonomy.
- There is no standalone OpenAPI download, no public GitHub presence (the justradar org has zero public repos), no MCP registry listing, and sandbox access requires emailing the team — an agent cannot self-serve its way into a first API call.
What to change
Prioritized by impact on discoverability. You (or your docs platform) deploy these — Discry never touches your API.
- 01Remove 'Disallow: /' from justradar.readme.io/robots.txt (or allow AI crawlers explicitly) — the single highest-impact fix; today every documentation page is invisible to compliant agents and search engines.
- 02Publish the complete OpenAPI spec at a stable public URL (e.g. /openapi.json). ReadMe already generates the reference from it — the validation pages prove the spec exists — and the placeholder tokenUrl (example.com) should be corrected while doing so.
- 03Regenerate the four core payout endpoint pages to the 2026 standard so their markdown exports include the OpenAPI definition (method, URL, request/response schemas), matching the new validation endpoints.
- 04Add llms-full.txt so agents can ingest the full API surface in one fetch instead of nine separate .md requests.
- 05Document rate limits and provide self-serve sandbox credentials — currently an agent must email demo@somosradar.com before it can attempt a single call.
Execution coverage · INFORMATIONAL, UNSCORED
Whether an agent can actually complete a call and recover from errors is the deeper Audit layer — documented here, but not part of the Discry Score.
Auth is a two-step user/apiToken login that returns a JWT bearer token; errors are machine-parseable JSON ({message, error}) with per-parameter validation tables mapped to 409/422 codes and a webhook errorType taxonomy. Client-supplied unique payment IDs (409 on duplicate) provide documented idempotency protection. Rate limits and pagination are not documented anywhere.