Spec compliance
Checked against internal and customer specs, instantly.
Compliance is two checks, not one: is the data valid enough to judge, and does the valid result land inside its spec window. Niobia gates on data validity first (a fit on bad data is a confident wrong answer), then reports conformance by lot and by limit.
What it measures
Spec compliance answers a deceptively simple question, pass or fail, and gets it wrong whenever it skips the validity gate that has to come first:
- Data validity: before a result can be compared to a spec, the measurement has to be sound. For EIS that means Kramers-Kronig validation: residuals under 0.5% are excellent, systematic residuals above 2% mean the data is non-stationary and a fit would be meaningless. Other methods have their own validity checks. A pass against spec computed from invalid data is worse than no answer, because it carries false confidence.
- Conformance: the valid result placed against its limits, single or double sided, with the margin reported, not just the binary. A value barely inside spec is a different signal than one centered in the window.
- Resolution: by lot and by limit, so a compliance summary points at which lots failed and which limit they failed, rather than a single aggregate pass rate that hides the pattern.
How to read the output
Read the validity gate before the verdict. A clean compliance report says, for each measurement, that the data passed its quality check and then where it landed against the limits. The dangerous report is the one that skipped straight to pass or fail: a green checkmark on a measurement whose underlying data never passed validation is a false negative waiting to escape. On the conformance itself, watch margins and patterns, not just the rate: a cluster of barely-passing results on one lot or one limit is a drift heading for a failure, and it is invisible in an aggregate pass percentage.
A real use case
Incoming cells are released against an internal-resistance spec measured by EIS. The naive workflow fits every spectrum and checks the resistance against the limit. With the validity gate first, two lots that would have passed are caught upstream: their spectra fail Kramers-Kronig validation with systematic residuals above 2%, meaning the cells were not stationary during measurement and the fitted resistance is unreliable, pass or fail. Those lots go back for re-measurement instead of being released on a number that did not mean anything. The rest pass the validity gate and are reported by lot against the limit, with one lot flagged at the low-margin edge for a closer look. Compliance caught both the invalid data and the drifting lot.
Common mistakes
- Comparing to spec without validating the data first. A fit on non-stationary or low-quality data produces a confident number that should not exist.
- Reporting only the binary. A pass at the edge of the window and a pass in the center are different risk signals; the margin matters.
- Aggregating away the pattern. A single pass rate hides which lots and which limits are drifting; resolve by lot and by limit.
- Treating a green result as proof of quality when the validity check was skipped, the most expensive kind of false confidence.
- Confusing internal specs with customer specs. The same measurement may pass one and fail the other; both windows have to be checked.
Validity first, then conformance, by lot and by limit
Niobia gates compliance on data validity before it compares anything to a spec. For EIS it runs Kramers-Kronig (Lin-KK) validation and refuses to fit when systematic residuals exceed 2%, so a resistance is never reported against a limit unless the measurement was sound, and the same data-quality discipline applies across methods. Once a result is valid, conformance is immediate: each measurement lands inside or outside its window, reported by lot and by limit against both internal and customer specs, with the margin visible rather than collapsed into a single rate. This is the same data-quality logic that runs at ingestion, carried through to the release decision so a green result always means a valid one.
Frequently asked
Why check data validity before checking the spec?
Because a spec comparison assumes the measurement is sound. Fit non-stationary or low-quality data and you get a precise number that means nothing, and if it happens to land inside the window it ships as a false pass. Validity gating (like Kramers-Kronig for EIS) prevents that.
What does 'by lot and by limit' add over a pass rate?
Direction. A single pass percentage tells you something is wrong but not what; resolving conformance by lot and by limit points at exactly which lots and which limit are drifting, which is the difference between an alert and an investigation lead.
Can it check both internal and customer specs?
Yes. The same valid measurement is compared against each window, because a result can pass an internal limit and fail a tighter customer one. Both verdicts are reported with their margins.