Richard Cook’s unit mismatch. The Mars Climate Orbiter is lost.

When three AI explanations disagree about the same digital SAT mistake, compare each one with College Board guidance and the worked question, then identify the exact skill the student could not apply. The useful outcome is a diagnosis such as “misread a nonlinear expression” or “chose evidence that did not support the inference,” followed by targeted practice.

On a Saturday morning, the situation might look simple. A student pastes one missed question into three AI tools. One explanation blames careless arithmetic. Another says the student misunderstood the wording. The third introduces a method the student has never seen.

Three confident answers have created more uncertainty, not less.

What a spacecraft failure revealed about conflicting explanations

In 1999, NASA lost contact with the Mars Climate Orbiter as the spacecraft approached Mars. The mission’s outcome was still uncertain when controllers waited for a signal after the planned manoeuvre. The signal never came.

The investigation found that one part of the mission’s ground software produced data in United States customary units, while another part expected metric units. NASA’s Mars Climate Orbiter Mishap Investigation Board documented the mismatch and also identified weaknesses in the processes intended to detect it.

Richard Cook, the project manager for Mars exploration projects at NASA’s Jet Propulsion Laboratory in Pasadena, became one of the named figures associated with explaining the failure publicly. The loss was not caused by an unfamiliar branch of physics. Two systems had interpreted the same information differently, and the checks did not expose the disagreement before it mattered.

A digital SAT error can have the same shape on a much smaller scale. Several explanations may describe the visible wrong answer, while the real task is to find which interpretation matches the assessment and which underlying skill failed.

The lesson from the orbiter is direct: confidence does not resolve conflicting interpretations. A shared reference and a disciplined checking process do.

Use College Board materials as the reference point

Start with the official description of what the digital SAT measures. College Board groups Reading and Writing questions into domains such as Information and Ideas, Craft and Structure, Expression of Ideas, and Standard English Conventions. Mathematics questions also sit within defined content domains.

That framework helps a family move beyond “the AI said it was a grammar mistake.” The sharper questions are:

• Which domain does this question test?
• What must the student notice before calculating or selecting an answer?
• Which step in the student’s work first departed from the required reasoning?
• Does the AI explanation use a rule or method that applies to this exact question?

Then compare the responses against the official answer explanation when one is available through College Board practice materials or Bluebook preparation resources. An AI answer that reaches the correct option through unsupported reasoning remains unreliable. A long explanation may also obscure the single decision that caused the error.

This checking habit matters because AI tools can produce different explanations from the same prompt. They may infer missing context, choose different solution methods, or state an incorrect rule with polished language. The student needs a way to test each claim rather than selecting the answer that sounds most certain.

The same principle appears in Richard Cook’s unit mismatch. The Mars Climate Orbiter is lost.: agreement about the desired destination did not protect the mission from incompatible assumptions.

Trace the first wrong step

Once the official guidance establishes what the question measures, return to the student’s original work. Do not begin by replacing it with the neatest AI solution.

Suppose the correct explanation shows that an SAT mathematics question tests equivalent expressions. If the student formed the right equation but distributed a negative sign incorrectly, the skill gap is algebraic manipulation. More reading practice would miss the cause.

If the student completed the calculations correctly but answered for \(x\) when the question asked for \(2x\), the issue may be interpreting the requested quantity. Assigning another page of identical equations would provide activity without addressing the mistake.

For Reading and Writing, a student may know every word in a passage yet choose an answer that adds a claim the text never supports. The gap could lie in evaluating evidence, not vocabulary.

Record the diagnosis in a precise sentence:

“I can solve the equation, but I need to check which quantity the question asks me to report.”

That statement gives the next practice session a purpose. “I am bad at SAT maths” does not.

A tutor can help by watching how the student approaches the question, comparing the work with the relevant digital SAT domain, and selecting follow-up problems that test the suspected gap. Accelerate Tutors matches families with tutors according to curriculum, subjects, goals, availability, and learning support needs, including support for the PSAT and digital SAT. Live one-to-one lessons are available online worldwide.

Build a repeatable checking routine

For each disputed AI explanation, ask the student to mark three things: the rule being used, the evidence from the question, and the step that changes the answer. Any explanation that cannot support all three should remain untrusted.

Next, check the relevant College Board guidance or official practice explanation. Compare methods carefully. Different valid methods can reach the same result, but each step must still follow from the question.

Finally, choose two or three new questions that isolate the diagnosed skill. If the student succeeds, vary the wording or context. If the same error returns, revise the diagnosis.

NASA’s investigators did more than name a unit mismatch. Their report examined the safeguards and communication failures that allowed it to survive. A useful SAT review should do the same: identify the wrong answer, then examine why the student’s checking process did not catch it.

By the end of the session, keep one written skill diagnosis, one corrected example, and a small set of focused practice questions. That is more useful than saving three AI responses and hoping the longest one was right.