An independent, signed math and claim check for when someone else has to trust your numbers. Your agent sends numbers or draft text, we recompute the math on our server, and you get PASS, FAIL or FLAGS, plus a self-contained /check link and a badge anyone can open to confirm the result.
Why not let your coding agent write a quick script? A script written by the same model that wrote the design carries the same assumptions, and a second party has no proof it wasn't edited until it printed PASS.
Planning and math verification check only — not a licensed professional engineering review, stamp, or sign-off.
Toll booth rates. No payments are taken in 2026. Every beta call runs at $0 under daily caps and shows what it would cost at launch.
Discovery, quotes and /check receipt verification: always free
One check, slice tool or claims check: 5 credits ($0.50 at launch)
A chain of up to 8 checks, or a sealed bundle: 5 credits ($0.50 at launch)
Invalid input or a tool that isn't built yet: 0 credits
Planned launch packs: 150 credits for $15, 500 for $45
Callable now in the beta (38)
General math and claim checks (8)
Recompute a formula from your own numbers and compare it to a target
Worst-case tolerance stack against fit limits (RSS spread reported alongside, not used for the verdict)
Total load current against a supply rating with a margin
Sampling rate at least twice the highest signal frequency
Value inside a stated rated range
Formula recompute with trig functions in radians or degrees
Paste text; every inline "arithmetic = number" claim is recomputed at its stated digits
Check every number in a README, PR, spec or memo against the signed results from this service it relies on, and flag fixed-list overclaim phrases (deterministic, no AI)
Physics slices (30)
pulse-echo speed (m/s) [slice B-05 · Wave speed]
corrected temperature, rise over ambient [slice B-09 · Trend fit]
Three reports built from our own house slices with example inputs, not measurements, and signed by this server at startup: /samples. One is a tank-level slice result. One is a power budget sealed in a bundle with two other math checks. One is a claims check that catches an overclaim sentence. Open any check link inside them to verify it.
Get a free sandbox key
Or from a terminal: curl -s -X POST https://[domain]/kiosk/key
Paste these lines into your agent's instructions so it uses the booth:
Before finalizing any equation, tolerance stack, power budget or sensor spec:
1. Call verify_math_or_tolerance or run_slice_tool on the Checkbooth MCP server.
2. Call verify_claims_against_receipts on your summary and include the /check link or badge in your output.
Attach a photo or file without uploading it
Pick a file. Your browser computes its SHA-256 fingerprint on your device, and the file never leaves it. Pass the fingerprint as evidence_digests and it is sealed inside your receipt, so anyone holding the same file can confirm it matches.
Who runs this
Checkbooth is operated by Wesley Covert in Sacramento County, California. Contact: the form at /contact. Terms of Use and Privacy Notice (v1.1-beta): /terms.
No AI runs in this beta. Every callable tool is deterministic code on our server, and nothing you send goes to an outside AI model. Receipts hold hashes and verdict summaries, never your raw text. Run records are kept for no more than 30 days.