The RCM binder is wrong and nobody knows which page
Somewhere near you is an RCM binder. It was produced by a serious team over several months, it was reviewed and signed, and it directs the maintenance of equipment your operation depends on. Some of its decisions are now wrong.
Nobody knows which ones.
That's not a criticism of the team. It's the nature of the artifact. An RCM study is a snapshot of the evidence available on the day it was written: this failure mode looked random, so run to failure; that one wore out, so replace at 4,000 hours; this protective function is hidden, so test it quarterly. Then the study is filed, the equipment keeps operating, and the evidence keeps accumulating — new failures, new suspensions, new inspection data — none of which the binder can hear.
Decisions decay silently
The standard defence is the periodic review: revisit the study every few years, re-examine every decision. Reviews are better than nothing, and they're also why the binder problem persists — because a full review is so expensive that it happens rarely, and when it happens, most of its effort is spent re-confirming decisions that were still fine. The few decisions that had actually gone stale get the same attention as the many that hadn't. There's no triage, because nothing tells you which decisions the new data disagrees with.
But notice what an RCM decision actually is: a claim about a life distribution. "Run to failure" claims the failure mode is random — a Weibull shape parameter near 1, where scheduled replacement buys nothing. "Replace on a fixed interval" claims wear-out at a specific, cost-justified interval. "Monitor condition" claims degradation is observable and predictable enough to act on. These are statistical statements. They can be checked — by exactly the models you'd fit from the failure data you've collected since.
Worksheets with live evidence
That's how RCM works in Reliafy. A study is the classic decomposition — functions, functional failures, failure modes, decisions — but every decision cites the analysis that justifies it. A run-to-failure call links to the fitted life model whose β says failures are random. A fixed interval links to the cost-optimal replacement analysis that produced it. An on-condition task links to the degradation model; a failure-finding interval links to the availability calculation for the hidden function.
And the citations are live. When you refit that life model with another year of data and the β that justified run-to-failure has crept from 1.0 to 2.3, the study doesn't stay serenely confident. The decision is flagged contradicted, the moment the evidence turns — with the model one click away, showing exactly what changed. The study's dashboard rolls it up: so-many decisions supported, so-many awaiting evidence, so-many contradicted.
The periodic review doesn't disappear — it becomes triage instead of excavation. You walk straight to the contradicted decisions, and the review that took a week of workshops becomes an afternoon.
The units check, too
A small thing that caught us during development: a surprising number of stale RCM decisions aren't statistically wrong, they're dimensionally wrong — an interval in months justified by a model fitted in operating hours, drifted apart by a change in shift patterns. Reliafy checks the units of every cited analysis against the decision that cites it, and refuses to call a decision supported when they disagree. Pedantry, weaponised.
Evidence you already have
The prerequisite isn't exotic: fitted models from your own failure data, which is what the rest of Reliafy is for. Fit the life model with your suspensions counted, run the replacement-interval analysis, and the RCM study consumes them as evidence — same platform, same datasets, no export dance between a statistics tool and a worksheet tool.
There's a complete sample study in the app — a pump system with linked evidence, including one decision the sample data contradicts, so you can see the flag without waiting for your own data to turn on you. It's on the free tier, along with one study of your own with full live validation.
The binder was never the deliverable. The deliverable was maintenance that matches how the equipment actually fails — keeps matching it. Paper can't do that. A worksheet wired to the data can.