Agent Fleet Truth-Probe Review

Specification as of 2026-06-13: you send 3 inputs, you get back 4 sections.

An adversarial review of up to twelve existing sanitized Python or Bash operational probes, with claim-to-source mapping, forced-failure tests, mutation evidence, and unapplied correction diffs.

$457

one-time

Request this →

Checkout for this service opens by emailed invoice link. Tell us where to send it and we reply the same working day.

What you get

Claim-to-source and six-condition failure-semantics matrices
Expected-good and forced-failure synthetic cases for every accepted probe
Mutation results showing whether existing guards reach both sides
BOUND, UNBOUND, or UNKNOWN verdicts carrying the reviewed fleet count
Unapplied correction diffs and a prioritized remediation brief
Delivery within seven business days after written scope and payment

Free specimen and paid-delivery boundary

A filled review of our own two target configurations using the same sealed battery, with every per-target and per-category total balanced across passed, failed, and unknown.

Specimen provenance: Captured: 2026-06-13T00:58:38Z · Target: our own sealed Nemotron baseline and hardened target configurations · Evidence method: Open the named audit database mode=ro, verify every row and batch seal, then count the latest sealed run per target with source errors published as UNKNOWN.

Proof scope: method boundary only. This free specimen does not demonstrate the full paid delivery.

What this free specimen actually shows

Two owned target configurations under the same sealed probe battery
Per-target and per-category passed, failed, and UNKNOWN arithmetic
Recomputed result-row and batch seals for every retained source row
Source-clock freshness and explicit UNKNOWN semantics

Paid delivery contract

You send:

One accepted, sanitized probe or behavioral-test surface
A versioned battery and explicit expected-good / forced-failure fixtures
A target fingerprint that excludes credentials and binds configuration

You get back:

Per-target probes / passed / failed / UNKNOWN matrix
Per-category arithmetic and deterministic-versus-model-assisted coverage
Row and batch seal verification for the evidence reviewed
An aggregate evidence table that keeps unavailable evidence out of pass/fail

Limits: This specimen reviews our own sealed behavioral audit evidence; it is not a BOUND/UNBOUND verdict on a buyer's probe code, a production-health statement, an audit opinion, or a certification. Only the latest sealed run for each target is included in the target scorecard; older runs remain counted in source coverage.

The database is opened read-only without immutable mode so a WAL cannot be skipped. Every result's canonical row hash and every run's sorted-hash batch seal are recomputed before counting. Manifest and detail counts must agree. Source pass maps to passed, source fail maps to failed, and source error maps only to UNKNOWN; all three visible columns must sum to probes_total. No transcript body, local endpoint, credential, or synthetic canary is exposed. This is bounded evidence of the dated method stated here; it does not establish affiliation, endorsement, authorization, or any current outcome beyond that scope.

View the full machine-readable specimen and evidence rows

See the other services

US Tech Automations is a data and automation company. These are our own service terms, not a government fee, filing cost, or permit charge. Nothing on this page is legal, engineering, or financial advice.

US Tech Automations · All services