Glossary
The words, as this product uses them.
Several of these exist to head off a specific misreading rather than to define a word — so they say what the thing is not, as well as what it is.
- Contributing condition
- Something that was true of the situation around an incident, rather than something a person did. A spotter being unavailable, time pressure, an unfamiliar vehicle, a shift in its eleventh hour. The working test is whether it could be true again tomorrow with entirely different people on shift — if so it is a condition, and it will recur. Contributing conditions vs. root cause →
- Root cause
- A single originating failure, found by following one causal thread back from an outcome. The technique comes from manufacturing, where the surrounding conditions are held constant by design. chatIR does not report root causes: a single thread through an EMS incident reliably terminates at whoever was closest to the outcome, which is why the method records conditions instead. Why the chain ends at a person →
- Population
- The group of incidents a figure was measured over — the denominator. Every count in a chatIR report is printed over its own population, because a numerator alone says nothing. Two figures scored over different populations are not comparable even when they count the same condition. How to read a pattern report →
- Denominator
- The size of the population a count is measured against. Reading a report well starts by finding the denominator before the numerator; a figure printed without one should be treated as decoration until somebody produces it.
- Stratum
- A subset of the record that a finding is scored inside — a location, an incident type, a shift. Scoring a pattern within its own stratum rather than across the whole log is what stops a pattern specific to one kind of work being diluted by everything else.
- SPECS
- The five domains a contributing condition can sit in: System, Process, Equipment, Crew, and Scene, in that order. System leads because it underlies everything; Scene trails because it is the most surface-level. It is a classification of where a described condition belongs — useful, but not itself a causal model or a control-structure analysis. The method →
- Display floor
- The minimum number of incidents a condition needs before it is drawn in the condition map. It is a legibility rule and nothing else: it was not derived from any safety standard, and a condition below it has not been assessed and found unimportant — it has not been assessed. Conditions below the floor still appear in the full inventory table, and a condition that crosses several groups is drawn regardless.
- Crossing (repeat)
- A condition that reaches a minimum share of the incidents in more than one group — several locations rather than one. Crossings are marked because a condition appearing across contexts is usually structural. A condition that concentrates in a single place is not a lesser finding; it is a different one, and often more actionable because it has a boundary you can put a control around.
- Confidence grade
- How much the record supports a finding — Confident, Established, or Directional. It is not a statement about how dangerous the finding is. A directional finding can concern something lethal and a confident one something trivial; support and importance are separate axes and the report keeps them separate deliberately. What a grade is claiming →
- Established
- A finding that is real and pervasive — enough incidents carry it — but whose severity was never recorded. It is a coverage finding and must never be read as a severity claim. A record with no severity column cannot produce anything stronger, and a report full of Established findings is an accurate description of what that instrument can see.
- Potential severity
- Exposure to harm, computed from a rubric over the hazard conditions described in a narrative — how hazardous the setup was, not a record of what actually happened. It is always labelled potential, never priced, and never substituted for recorded severity. See it in the sample report →
- Co-occurrence
- Two or more conditions appearing together in the same incidents more often than they appear apart. It is an observation about a record, not a claim that one produces the other. The most useful question to ask of any co-occurrence is whether a third thing would explain both.
- Near miss
- An event that carried the same conditions as an incident but produced no injury or damage. The cheapest evidence a safety program can get, and the most reliably missing — nothing forces a near miss into the record, so the count usually measures willingness to report rather than how often they happen. Building a near-miss reporting culture →
- Normalization of deviance
- Practice drifting from the standard and becoming accepted a little more each time, precisely because nothing has gone wrong yet. Named by Diane Vaughan in her work on the Challenger accident. In an interview it surfaces as language rather than admission: we usually just, normally we, it has not been a problem before.
- Just culture
- An approach that separates accountability from blame — accountability being able to give an account of what happened and why it made sense at the time, blame being what happens when that account is used against the person giving it. The common failure is announcing the policy while keeping a process shaped for blame.
- Candidate control
- A proposed change treated as a claim to be tested rather than a decision already justified. The report writes each one with the operational step, a slot for the owner your team still has to name, and the question of how it could be observed — including when it is unavailable. Turning those into an owner and an agreed check is work a working session does, not something a report can do on your behalf. Calling it a candidate is not hedging: the record establishes that a condition recurs, never that a particular control will address it. Corrective actions that survive 90 days →
- Observable check
- The thing you could look at that would show a control was actually available and used — not whether a policy exists, but whether it operated. If none can be named, there is no control yet, only an intention.
- Reporting instrument
- The form itself, as distinct from the records it produces. Auditing the instrument asks what the form is capable of capturing; auditing the record asks what it happened to capture. The first is answerable with no data at all, which is why it is the free rung. Audit your instrument →
- Vocabulary fragmentation
- One condition recorded under several different labels, so it is counted several times as something smaller. It is worst for the condition people have had the most occasions to describe, which is usually the most common one — so a factor count taken straight from a dropdown tends to understate exactly the thing you most need counted.
- Evidence limit
- A statement of what the record could not settle: where the population was too small, where absence cannot be distinguished from a missing field, where a finding may be an artefact of wording. A report that states its limits can be checked; one that only makes claims cannot. The method and its limits →