Find out what your sampling plan cannot see, before you collect the data.

Where this goes wrong

• Audit samples. Picking every Nth record from a population with a weekly or monthly rhythm can keep landing on the same day, clerk, or batch. Auditing texts warn about this; the plan rarely gets checked against it.
• Multi-cavity molds and multi-head lines. When the inspection interval lines up with the cavity or head count, inspection can keep returning to the same cavity while a bad one ships.
• Rotating machinery. Vibration sampled in step with the shaft, or sensors placed at evenly spaced angles, can leave some vibration orders invisible.
• Fixed-time monitoring. Readings taken at the same clock time every day, like satellites that pass overhead at the same local time, never see what happens during the rest of the day.

The evidence

We tested the check before selling it. Across ten plans in three settings, the catch rate it predicted from each plan alone matched simulation to within a percentage point, with every verdict frozen before the test ran. On a robot joint, the plans it admitted learned a change they had never seen and the plan it refused did not, in two runs made in opposite order.

Download the white paper (PDF)

What you send

The plan, not the data. Nothing measured has to leave your organisation.

1. How you sample: the interval, rate, or sample times; or the sensor positions; or the selection rule for records or parts, with the number of samples.
2. What you need to catch: the cycles, frequencies, orders, or repeating patterns that matter, such as a weekly cycle, a cavity count, or engine orders 1 to 12.
3. How small a change must show up, and, if you have one, a rough noise level and timing uncertainty.

What you get

1. A verdict for the plan: admitted, meaning every declared pattern is visible at the size you asked for, or refused.
2. For a refusal, the named blind spots: which declared patterns the plan cannot see, or cannot see at the size you asked for.
3. A suggested adjustment to the interval, positions, or count that removes the blind spots, checked the same way, where one exists.
4. A signed receipt binding the plan you sent to the verdict, re-checkable later with the public keys published on codemani.com.

Limits

• The answer covers the patterns you declare, at the noise and timing you declare. A pattern outside that list is outside the answer.
• It assumes the plan is carried out as written. If your sample times drift or jitter more than you told us, say so; it changes the answer.
• It is a check of the plan, not an engineering sign-off, a statistical audit opinion, or a compliance certification.

Price

Early access: we check your first plan at no charge. After that, each plan is quoted up front, before any work starts.

Send a plan to check

Email the three items above to [email protected]. A plain description is fine; we reply by email with questions if something is ambiguous.