Predictive Maintenance Field Frameworks: The Evidence Chain

Predictive Maintenance Part 12 of 12

Six Predictive Maintenance Field Cards

Six one-page cards for the records a predictive programme keeps skipping: the detection evidence record, the criticality queue, the failure-mode sentence, the measurement contract, the alert anatomy and its ladder, and the pilot registration sheet. Questions and record fields only.

Lokesh Chennuru

Aug 11, 2026 5 min read

Six one-page cards for the records a predictive programme keeps meaning to write down and never does: the detection evidence record, the criticality queue, the failure-mode sentence, the measurement contract, the alert anatomy and its ladder, and the pilot registration sheet.

Each card carries questions and record fields, and nothing else. No thresholds, no intervals, no scales, no authorisation — because those come from your own plant, and a number borrowed from somewhere else is the fastest way to lose an audit. Every card also carries the one-line boundary the underlying reference states at that point, and the section it came from.

No signup, no email, no form — download what you need.

01 · The detection evidence record

A P–F interval is a claim about one failure mode, one functional boundary, one measurement method, one operating context and one decision process. Six fields hold the claim — function and mode, measurement contract, detection criterion, warning evidence, response path, governance — and four questions test it. A warning window carried in from another site fills none of them.

Download card 01 (PDF) · Pairs with The P–F Interval Is a Measurement Hypothesis

Reference card: the detection evidence record. Six numbered fields — function and failure mode, measurement contract, detection criterion, warning evidence, response path, governance — above four evidence questions covering what warning evidence exists, when a valid measurement can occur, what the response path demands, and which uncertainty and consequence controls apply. — The six fields are what makes a P–F claim reviewable; the four questions are what stops a mean or an anecdote standing in for a lower bound.

02 · Ordering the queue

Two questions per machine — how much does it hurt when this fails, and how often does that happen — answered against written scales and applied to every asset with the same yardstick. The card records the slot rather than the machine, the consequence criterion by criterion, the likelihood and its evidence source, the combination rule fixed before scoring starts, the overrides that sit outside the ranking entirely, and the assumptions the score will decay with.

Download card 02 (PDF) · Pairs with From “Everything Is Important” to an Ordered List

Reference card: ordering the queue. Six record fields covering functional location, consequence criteria, likelihood with its evidence source, the combination rule, overrides and the dated assumptions, above five questions asking whether the rule was fixed before the result was visible, who was in the room, worst credible versus worst imaginable, whether the standby was demonstrated, and which scores rest on evidence nobody can produce. — Criticality is a cross-functional judgement exercise dressed as arithmetic — so the card records the rule, the room and the date alongside the score.

03 · Four words, one sentence

“It broke” aims no sensor. Mode, mechanism, cause and effect — plus the two things that are not modes, symptom and detectable condition — turn the same event into a sentence that tells you where to look, what might be observable in the data, what to fix so it stops recurring, and why anyone should pay for the watching.

Download card 03 (PDF) · Pairs with Bad Actors

Reference card: four words, one sentence. Seven record fields — functional failure, failure mode at maintainable-item level, failure mechanism, failure cause, effect at local, next and end levels, symptom which is not a mode, and detectable condition — above four questions testing whether the row reads as a sentence, has a mechanism, merges two modes, and scores severity at the end effect. — A row without a mechanism cannot be monitored, only mourned — which is why the mechanism gets its own field rather than being implied by the mode.

04 · The measurement-evidence record

A sensor can work exactly as designed and still support no defensible maintenance decision. Nine fields decide whether a condition signal is evidence about a named failure mechanism or a datasheet feature that happens to be trending — and the record resolves to one of three states, one of which is not demonstrated. Silence does not promote a signal: a quiet trend can equally be an insensitive measurement, an invalid operating state, a missed observation, or a channel that died last Tuesday.

Download card 04 (PDF) · Pairs with A Data-Quality Problem First

Reference card: the measurement-evidence record. Nine numbered fields from required function and failure mode through candidate signal, measurement contract, data-quality controls, detection and confirmation rule, warning evidence, failure and uncertainty modes, response path and governance, above the three evidence states — supported for this context, controlled trial, not demonstrated — and the question of whether a signal has been quiet or blind. — Nine fields, three states, and the reminder that a flat trend is not the same thing as a healthy machine.

05 · Seven fields, five rungs

The consumer is a planner or technician who was not watching the screen and has eleven other things due. Seven fields — identity, finding, meaning, severity, confidence, time, action and owner — fit one screen, and a candidate that cannot fill meaning is a threshold event for an analyst’s queue rather than a dispatch. Beneath them, the five-rung ladder that names how far the evidence has actually climbed.

Download card 05 (PDF) · Pairs with The Anatomy of an Alert Worth Acting On

Reference card: seven fields, five rungs. The seven alert fields — identity, finding, meaning, severity, confidence, time, action and owner — above the five-rung ladder running anomaly, suspected failure mode, confirmed defect, work recommendation, and authorised operating decision, with the fifth rung marked as belonging outside the monitoring system. — A monitoring system climbs to rung four on its own evidence; rung five is a different kind of decision, made by people who answer for the consequence.

06 · The registration sheet

A verdict negotiated after the evidence arrives is not a verdict but a settlement. Six lines, filed before day zero: the primary outcome and how it will be computed, the question the window is too short to settle, the stop line from your own committed spend and your own baseline, the secondaries, what explicitly does not count, and the date with the names of the jury.

Download card 06 (PDF) · Pairs with Pre-Register the Pilot or Don’t Run It

Reference card: the registration sheet. Six record lines covering primary outcome, the question declared undecidable in advance, the stop line computed from committed spend over horizon and baseline, secondary outcomes, explicit non-outcomes, and the date and jury — above four registration rules about registering the analysis, pre-committing the readings, saying what the window cannot settle, and deriving the length from the failure. — The stop line is arithmetic on two numbers you already own — which is exactly why it has to be written down before anyone can see the result.

From the course toolkit: two further cards

The Practitioners Academy toolkit for this reference adds two more cards in the same format — same rule, same boundary panels, questions and record fields only.

07 · The data-readiness page

Whether the data can carry the decision is a question you can answer before anyone budgets a model — per consumer of the data, not per warehouse. One page for the six dimensions a readiness claim has to survive, and the record each one leaves behind.

Download card 07 (PDF) · Pairs with Predictive Maintenance Is a Data-Quality Problem First

Reference card: the data-readiness page. Record fields and evidence questions for the six data-quality dimensions, assessed per decision consumer, with the boundary panel stating what a readiness record does not authorise. — Readiness is claimed per consumer of the data, not per warehouse — and the record says who checked, and how.

08 · The five denominators

Every evaluation number is a fraction, and the honest part is the denominator. One page for the five denominators an evaluation claim has to declare before precision, recall or a false-alarm rate means anything at all.

Download card 08 (PDF) · Pairs with Remaining Useful Life: A Window, Not a Date

Reference card: the five denominators. Record fields and questions for declaring the denominators behind precision, recall and false-alarm claims, with the boundary panel on what an evaluation number cannot establish. — A precision figure with an undeclared denominator is a rumour with a decimal point.

What these cards are, and what they aren’t

They are records: what to write down, what to ask of it, and what the answer cannot establish. They are deliberately not fillable forms, scoring calculators or threshold tables.

Every card stops exactly where your own plant’s numbers begin. Card 02 names the consequence criteria and refuses to supply the scale. Card 06 gives the stop-line arithmetic and refuses to supply the share. Card 01 lists the fields a warning claim has to fill and declines to name an interval. That is not caution for its own sake — a scale, a share or an interval that came from another plant’s cost base, duty and failure history is not evidence about yours, and it is the first thing a reviewer will find.

Each card also carries the boundary its source section states, in one line, at the point where a reader might otherwise act. Where the record cannot be completed, the honest entry on all six is the same: NOT DEMONSTRATED — a verdict in its own right, not a failure of nerve.

Print one. Fill it in by hand against a machine you already argue about. You will know very quickly which of the six your programme is missing.

Which of the six cards describes a record your plant does not currently keep?

The record fields, the evidence questions, and the one-line boundaries printed on each card are drawn from the foundations, criticality, failure-mode, monitoring-technology, alerting and pilot-governance parts of Predictive Maintenance: Practitioner Reference Frameworks and Planning Guide (sections 1.4, 2.4, 3.1, 4.1, 10.1 and 12.5).

Lokesh Chennuru

Industry Digits Author

Lokesh Chennuru writes Industry Digits field notes for industrial decision makers, focused on automation, IIoT, condition monitoring, predictive maintenance, and industrial AI.

Connect on LinkedIn

Frequently asked

Questions industrial leaders ask about this

Are the field cards free to download?

Yes. All six cards and the combined six-card PDF download directly, with no signup, no email and no form. They are method references — what to record and what to ask — and they are yours to print, post on a cabinet door or hand to a shift.

What is on each card?

Questions and record fields, and nothing else. Each card carries a RECORD block listing the fields a decision has to fill, an ASK block listing the questions that test them, the one-line boundary the underlying reference states at that point, and the section it was drawn from. There are no thresholds, no intervals, no scales and no authorisation on any of the six.

Why do the cards contain no numbers?

Because a number that came from one plant's cost base, duty and failure history is not evidence about another's. The cards carry the method that produces your numbers — the fields to record, the arithmetic to compute from your own baseline — and deliberately stop where a borrowed threshold would begin.

Which article does each card pair with?

Card 01 pairs with the P–F interval article, card 02 with the criticality article, card 03 with the bad actors article, card 04 with the data-quality article, card 05 with the alert anatomy article, and card 06 with the pilot pre-registration article. Each card also names the section of the underlying reference it was drawn from.

Go deeper

Predictive Maintenance — Practitioner Reference Frameworks and Planning Guide

The twelve-part reference this series draws on: foundations and the value case, asset criticality and strategy, failure modes and degradation, the monitoring technologies, asset-class playbooks, sensors and IIoT architecture, data foundations, signal processing, analytics and prediction models, alerts and diagnosis, work management and CMMS integration, and pilot execution through rollout and governance — 126 sections with 46 technical figures.

View the course Book a consultation