Test 0: a broken QA cockpit cannot certify anything.

The step

Validate the cockpit before judging any production workbook.

Test 0 opens the dropdown and output audit tab and confirms that the QA workbook's own machinery behaves: selector lists populate correctly, Matrix ID lookups resolve, formula zones compute, expected-output linkage joins, error counters count, retest status propagates, and Master Control readiness reflects the state beneath it. Only then does the workbook get used on anything. The doctrine's own phrasing is the argument in full — a broken QA cockpit cannot certify MARS. Every verdict the workbook issues is a function of its own correctness, so a fault in the instrument is not one error among many; it is a multiplier applied to every result.

Why it is numbered zero

It is not the first step of the audit; it is the precondition for having one.

The numbering carries real meaning. Steps one through fourteen are the formal pass — choose the workbook version, select the scenario, confirm contracts, log the run, inspect, register, repair, retest, certify. Test 0 is not part of that sequence and does not run per-scenario; it establishes that the sequence is capable of meaning anything. Numbering it one would have implied it belongs to each pass and can be traded off against time pressure like any other step. Numbering it zero says the pass does not begin until it is done, which is a different kind of statement about a step that will always feel like overhead.

The recursion, and where it stops

Who audits the auditor's auditor is a real question with a practical answer.

The obvious objection is that Test 0 is itself performed using the QA workbook, which invites an infinite regress. The regress is broken the way it is broken in every real inspection system: by making the self-check simple enough to verify by direct observation rather than by inference. Confirming that a dropdown lists the expected values, that a lookup returns the row it should, and that a counter matches a manual count are all checks an operator performs by looking, not by trusting another formula. The self-audit is deliberately shallow and mechanical for exactly this reason — a sophisticated self-check would need its own verification, and a trivially observable one does not.

• Test 0 checks things that can be confirmed by eye, which is what stops the regress.
• It is not a test of QA judgement; it is a test of QA plumbing.
• Its shallowness is a feature — depth here would reintroduce the problem it exists to close.

The general principle

Every measuring instrument in MARS gets measured.

Test 0 is one instance of a pattern that runs through the system. The Monte Carlo benchmark is the ruler against which live results are compared, and it carries a regeneration log recording which ruler graded which weeks, because a ruler that is silently re-cut invalidates every comparison made against it. The volatility layer's sample status ladder exists to say whether a percentile has enough readings behind it to be a percentile. The efficiency lab stamps a data QA status on every row before grading it. In each case the system declines to trust an instrument merely because it produced a number, and Test 0 is that principle applied to the instrument whose failure would be most expensive.

The key idea

The most dangerous fault is in the thing that decides whether faults exist.

A broken production workbook produces wrong answers that a working QA layer will eventually catch. A broken QA layer produces confident certifications of whatever it is pointed at, and there is nothing above it to notice. That asymmetry is why the cheapest, dullest step in the whole process is the one that runs first.