Verification record · Build with Gemini XPRIZE

TradePass

Texas exam profiles match PSI’s current bulletin. Both official bulletins contradict the submitted “closed-book” description; business figures remain unverified.

sound.fan · Published September 17, 2026

Claim

TradePass offers electrician licensing-exam preparation run by Gemini agents, with practice tests built to state exam outlines. Its submission describes the Texas journeyman license as requiring “one closed-book exam” and reports five subscribers reconciled to Stripe.

What we tested

On 17 September 2026, sound.fan compared the public Texas journeyman and master exam profiles with PSI’s official candidate bulletin, updated 15 September, and the older TDLR-hosted bulletin linked from the journeyman page. We checked the homepage’s code-edition statement, inspected the public transparency page, and recomputed the submission’s sample and business arithmetic. This was a check of public profiles, documents and calculations, not an audit of the full practice-question bank or private operations.

Finding

Corroborated: the live Texas exam profiles match PSI’s current bulletin to the question and minute.

Exam	First portion	Second portion	Pass threshold
Texas journeyman	59 questions / 130 minutes	26 questions / 110 minutes	70% on each
Texas master	75 questions / 150 minutes	33 questions / 170 minutes	70% on each

The submission’s “closed-book” description is contradicted by both official bulletins, including the older one applicable at the submission deadline: candidates may bring the electrical code. The exam is open book.

What the evidence establishes

The later live exam profiles moved from Observed on participant-controlled pages to Corroborated against PSI. The NEC 2026 effective date of 1 September is corroborated; the claimed 28 August question switch is only Observed on the participant’s site. The journeyman page links an older 2023 TDLR bulletin with a single 80-item format. That is a stale source link, not an error in the current two-part profile.

Reproduced arithmetic: the sample feeder calculation is approximately 145.7 A; 3,407 of 7,770 rejected questions is 43.8%; $455 divided by 4,362 is about $0.104; the reported acquisition-cost/lifetime-value ratio is 2.31×. Recomputing those numbers does not establish the underlying spending, customer data or question quality.

Observed, later live surface: the transparency page reported two human interventions and three autonomous ad actions that month, consistent with the participant’s 9 September autopilot update. These are participant-controlled counts.

What remains unverified

Not verified: the five submission-time subscribers, the seventh subscriber announced on 11 September, revenue and usage, underlying business inputs, and the production operation of the Gemini agents. A statement that counts were reconciled to Stripe is still participant-reported until the reconciliation evidence can be checked independently. Matching the published exam structure does not establish the correctness of every practice question or improved exam outcomes.

What evidence would change that

A dated Stripe subscription export with hashed customer identifiers could reconcile the submitted subscriber count and later change against processor records. A versioned question sample with its official code references would allow a bounded question-accuracy check. Dated operational logs linking agent decisions to the reported ad actions would address the separate autonomy claim.