How Checkride107 works: the 90-second tour

Ninety seconds, five stages, no algorithms.

1. The bank

464 questions live as static, version-controlled data. Each carries the FAA section its answer comes from: 14 CFR Part 107, Part 89, Part 48, the AIM, an advisory circular, a handbook chapter, or the chart legend. No question is generated by a model, ever. The AI in the product explains a stored answer; it cannot write, change or re-key one.

In October 2026 the whole bank was reviewed question by question against the text of the FAA documents it cites, and each verdict quotes the passage it rests on. The fixes it called for shipped as dated change sets, 2026.10.5 to 2026.10.7. A person has read 364 of the 464 questions against the source.

Read the source. Under the explanation, the student can open what the question cites, wherever the library holds it. For the 274 questions that cite a regulation, that is the section text itself, with the date the text is current to. For a testing-supplement figure, it is the FAA figure page. For other FAA documents, it is the document's title and edition, and the documents are listed in the free resource library.

2. The guards

Before any deploy, the build runs a set of checks and fails on any of them. Every question must cite a source. The bank size must be rendered from the bank, never typed into a page. The blueprint shares must match PSI’s bulletin. Three shape tells — answer length, answer position, and absolutes appearing only in wrong answers — are measured and the build fails if any drifts. A claim of “unlimited” sims may only ship if the bank supports it. The retired ACS percentage bands may not appear anywhere. Alongside the build, a source audit runs on every change to the bank and lists any question whose content no longer matches a review quoted from its source. The regulation text a student reads is copied byte for byte from a dated eCFR snapshot at build time, never edited by hand. Each guard exists because the thing it prevents once happened.

3. The sims

A timed simulation draws 60 questions across the five content areas at the current blueprint’s weighting — 48 / 25 / 20 / 5 / 2 — with the five unscored validation items and the 120-minute clock. It excludes every question the student has already been shown, and the number of clean sims the bank supports (5 today) is computed from the bank rather than claimed.

4. Readiness

Every answer a student gives, in drills or sims, feeds a per-topic accuracy that weights recent answers more than old ones. Each topic shows its score and the number of answers behind it; under five answers it shows nothing. The gate is 85 percent on every topic plus two timed sims at 80 percent or better. Only the summary leaves the browser — score, sample size, gate status. No answers, no question ids.

5. The instructor view

A cohort is a list of students. The view is per student and per topic, with sample sizes, and one line naming the topic to reteach first, computed from the whole class rather than one student’s bad week. Seats are either a monthly licence an instructor reassigns freely, or prepaid, consumed once, with the ninety days starting when the student activates.

---

What is not in the tour: a pass rate, because we do not publish one; a prediction, because the gate does not make one; and a vendor comparison, because we do not name any. What is in it is checkable from the product in about four minutes.

The standard we measure ourselves against →