The missing event is evidence

Records draw attention toward what they contain. A timestamp says an action occurred. A row says a rule fired. A signature says someone accepted responsibility. Reviewers follow these marks because they offer something concrete to inspect.

The blank spaces are harder. A notice was expected but no notice event exists. A review should have occurred, but the record moves directly from recommendation to action. A policy requires approval, yet no approval appears. The common instinct is to say that the record tells us nothing. Perhaps the event happened and was not captured. Perhaps it never happened. Without an entry, certainty seems impossible.

That caution is right, but “nothing can be known” goes too far. An expected event that is missing is itself evidence. It may be evidence that the event did not occur, that the recording system failed, that the process bypassed a required step, or that the record was altered. Those possibilities are different, but they share an important fact: the account is incomplete exactly where completeness was expected.

Absence becomes evidence through expectation

Not every missing event matters. A record cannot contain everything, and the absence of an event no one expected says very little. Negative evidence begins with a justified expectation that a mark should exist if the underlying event occurred.

If every completed review is required to produce an approval record, then a completed action without one creates a meaningful discrepancy. If a notification service produces a delivery event for every successful notice, the absence of that event weighs against the claim that delivery succeeded. The strength of the inference depends on how reliable and universal the recording rule is.

This is familiar outside software. A locked door with no sign of forced entry does not prove no one entered; it changes the field of plausible explanations. A bank statement without an expected payment does not prove the payer never initiated it; it tells us the account and the claim are not yet reconciled. The blank matters because a process was supposed to leave a trace.

A missing event is strongest evidence when the system can first prove that the event had nowhere legitimate to hide.

That last condition is essential. Organizations cannot treat their own unreliable logging as proof against the person affected by it. If the record frequently drops events, if different paths write to different stores, or if retention is inconsistent, absence is weak evidence about the underlying action and strong evidence about the quality of the system’s memory.

Distinguish “did not happen” from “was not recorded”

A trustworthy account preserves this distinction instead of collapsing it. “No approval occurred” and “no approval record was found” are not interchangeable statements. The first is about the world. The second is about the available evidence.

The difference matters most when the record is used to assign responsibility. Suppose an operator says they approved an action, but no approval event appears. A system designed to record every approval atomically with the action gives the missing event considerable weight. A system in which approval and logging occur separately cannot make the same claim. The log may have failed after the approval succeeded.

This is why recording should be part of the action’s conditions, not a decorative step afterward. When the consequential act and its required record can diverge, every later review inherits an avoidable ambiguity. The system has created two histories: what may have happened and what it can prove happened.

Some designs respond by failing closed when the record cannot be written. That can be appropriate when acting without an account would create an indefensible result. It is not universally harmless. Refusing an urgent or protective action because a logging service is unavailable can impose its own cost. The governing question is again substantive: which actions require a durable trace before they may proceed, and which may proceed temporarily under a separately recorded exception?

The answer should be established before failure. Otherwise the missing event becomes an invitation to reconstruct policy after the outcome is known.

Record expectations as well as events

Most logs describe observed events. They do not describe what events were required. A later reviewer sees what happened but lacks the grammar needed to identify what is missing.

A fuller record should carry the expected sequence for the decision path: evidence gathered, authority checked, review completed, action taken, notice sent, and any opportunity to contest opened. The sequence need not be rigid in every case, but required events and legitimate exceptions should be identifiable. Only then can an absence be evaluated against a declared expectation rather than an auditor’s intuition.

This makes gaps visible in real time. If an action arrives without the approval it depends on, the system can stop or mark the decision incomplete. If a notice lacks delivery confirmation, it can remain unresolved rather than being silently treated as delivered. If an exception removes a normal step, the exception becomes an event of its own, with an author and a reason.

A Decision Receipt should therefore preserve both positive and negative findings. It can say that a required event was present, that it was absent, or that its status could not be established because the recording channel was impaired. Those statements carry different levels of confidence. Honesty requires keeping them different.

The same discipline improves appeals. A person should not be forced to prove that an invisible step never happened. If the institution claims a review occurred, the institution should produce the record its own process required. When it cannot, the missing event should count against the claim in proportion to the reliability of the system that was supposed to preserve it.

Records are often imagined as collections of facts. In practice, they are also arrangements of expectations. Their silences can reveal bypassed steps, failed memory, and unsupported claims—but only when the system has made clear what sound it promised to make.

The blank is not a verdict. It is evidence that the account has work left to do.

— Dispatches · Summit Cognitive