1,847 incidents. Twenty-one conditions. Three dimensions receive no recorded factor labels.
This is the shape of a delivered diagnostic — the same sheets, in the same order, with the same stated limits. Read it through and you will see how a messy export becomes scored patterns, one candidate control worth testing, and a list of things the record still cannot settle.
incidents reviewed
1,847incidents reviewedJan 2024 – Dec 2025 · 24 months
contributing conditions found
21contributing conditions foundfive of them repeat across locations
of five SPECS dimensions receive no recorded factor labels
3of five SPECS dimensions receive no recorded factor labelsthe export cannot tell us whether those options were absent or simply unused
01 · Where to begin
One bounded review path, and how the whole log narrows to it
Decision brief
Where to begin before you read the evidence
This page introduces a few bounded groups of records to review. The record review, condition maps, matrices, connected patterns, and full evidence pages that follow remain the body of the report.
Repeat-ID early-duty apparatus bay, garage, or carport records
Repeat-ID early-duty apparatus bay, garage, or carport records: 12 of 31 repeat-ID early-duty incident records.
Check whether this is a pull-out, changeover, staging, or facility-movement cohort.
Why look: It narrows a broad time-of-shift question to a concrete piece of work operations can describe in a sentence.
Next question: What movement, handoff, site layout, and vehicle-check work do these records represent?
Evidence available: repeat-ID early-duty incident records at a bay, garage, or carport: 12 of 31 repeat-ID early-duty incident records. This is a concentration in reported incidents, not a utilization-adjusted risk rate or evidence of cause.
96 incident records in the repeated-unit cohortof 218 incident records naming a unit
Repeated normalized unit IDs
24 normalized unit IDs recur in 96 of the 218 incident records that name a unit.
Why look: A bounded repeat cohort gives fleet and operations staff a smaller place to begin without labeling a unit unsafe.
Next question: Do these IDs represent stable physical units, and how do miles, calls, engine hours, assignment, and home station compare?
Vehicle IDs in this group
Reported-incident counts, not a ranking of vehicle safety. IDs use the normalization described below.
Vehicle ID
Reported incidents
Medic 18
4
Medic 22
4
Medic 30
4
Medic 41
4
Medic 52
4
Show all 24 vehicle IDs
Vehicle ID
Reported incidents
Medic 6
4
Medic 7
4
Medic 9
4
Rescue 18
4
Rescue 22
4
Rescue 30
4
Rescue 41
4
Rescue 52
4
Rescue 6
4
Rescue 7
4
Rescue 9
4
Squad 18
4
Squad 22
4
Squad 30
4
Squad 41
4
Squad 52
4
Squad 6
4
Squad 7
4
Squad 9
4
Showing 5 of 24 IDs, most frequently recorded first. Counts use the full repeated-unit group.
Evidence available: incident records in the repeated-unit cohort: 96 of 218 incident records naming a unit. This is a concentration in reported incidents, not a utilization-adjusted risk rate or evidence of cause.
A separate view: how the record was narrowed
Each band names the group it is counted against. Every band counts the same thing: incident records. Bar length shows the share of its named group. It is not utilization, exposure, or a rate.
narrows to
narrows to
Separate views of these 31 incident records. Groups may overlap.
Where the path stopped
Only 3 of the 31 repeat-ID early-duty incident records carry a recorded weather value — fewer than the 5 this report draws a stage on.
How this path was narrowed
Each stage is computed within its named denominator. Slices sharing a denominator may overlap. The records behind every count are kept with the report. Normalizing a unit ID means trimming it and collapsing runs of whitespace, including whitespace around hyphen separators — nothing else is merged, and no identifier is guessed. A record missing the value a stage tests is excluded from that subset. The named denominator states which records it is counted against.
What this brief cannot show
These are concentrations inside reported incidents. Without miles, calls, engine hours, headcount, or another exposure measure there is no denominator for a rate, so nothing here establishes that a unit, a shift hour, a location, or a person is riskier than any other. A recorded cause is the label a reporter chose, and reporting practice itself varies. Treat every cohort as somewhere to look.
Carry these into the evidence
What to validate next
Repeat-ID early-duty records recording backing
Which maneuvers, sites, vehicles, and spotting practices do these records actually represent?
8 of the 31 repeat-ID early-duty incident records name backing as the recorded cause.
Why look: A named maneuver inside an already-bounded cohort is something a supervisor can watch once and describe.
Evidence available: repeat-ID early-duty incident records recording backing: 8 of 31 repeat-ID early-duty incident records. This is a concentration in reported incidents, not a utilization-adjusted risk rate or evidence of cause.
Keep in view
Other things worth checking
Reported incidents involving personnel in their first six months137 reported incidents involve personnel in their first six months.
Treat this as a review cohort, not evidence that newer personnel are less safe.
Why look: It gives operations a simple place to inspect what work these reports represent before drawing any conclusion about people.
Next question: What assignment, vehicle type, shift, route, facility movement, supervision, and reporting practice do these records represent?
Evidence available: reported incidents involving personnel in their first six months: 137 of 1,203 reported incidents with parsable tenure. This is a concentration in reported incidents, not a utilization-adjusted risk rate or evidence of cause.
02 · Your log as it stands
Where the record already is, before anything new
Your log as it stands
Start where you already are
Before anything new: here is your incident history as your current process describes it.
1,847 incidents over roughly 24 months.
32.7% are Vehicle/Rig incidents — 604 of 1,847.
“No spotter” — 144 incidents (23.8% of incidents Vehicle/Rig incidents), your most-recorded contributing factor.
“Human error” — 131 incidents (21.7% of incidents Vehicle/Rig incidents).
Compare factor labels with the context already present in the supplied records.
Before you read the charts
What do these categories mean?
SPECS is chatIR’s way of organizing conditions described in an incident record: System, Process, Equipment, Crew, and Scene. The charts show how often those conditions were documented. They are not safety grades.
Scene
The conditions surrounding the work
The setting in which the incident happened, including layout, visibility, weather, space, and other activity nearby.
System
The operation doing the work
Process
How work is organized and carried out: procedures, training, checks, and handoffs.
Equipment
The vehicles, tools, and devices used, including their condition, availability, and suitability for the task.
Crew
The people doing the work: their experience, actions, communication, and coordination.
Read Scene and System together to see the circumstances of the work and the operation involved. These counts alone cannot tell us whether the operation was equipped to meet those circumstances.
How we got these numbers
Your form’s factor labels We group the contributing-factor labels already recorded on your forms into these categories.
What chatIR extracted We identify conditions in the supplied narratives and structured fields, then group the supported observations into these categories.
Each bar counts incidents with at least one matching observation included in that category. Its percentage uses the report’s incident total. An incident can appear in several categories, so the bars do not add up to 100%. A low or zero count means few or no matching observations were included. Information may be missing, ambiguous, or excluded from this comparison; it does not establish that a condition was absent or that the operation performed well.
Your form’s factor labels
S
Scene
65%1204 incidents
System
P
Process0%0 incidents
E
Equipment0%0 incidents
C
Crew
27%498 incidents
What chatIR extracted
S
Scene
38%702 incidents
System
P
Process
31%573 incidents
E
Equipment
12%221 incidents
C
Crew
16%296 incidents
Organization-wide observations appear in 168 incidents. These may overlap with Process, Equipment, and Crew counts.
Layers aren’t mutually exclusive: an incident can carry conditions in two or three of them at once, which is why the extracted shares above add to more than 100%. Your recorded factor labels are close to one-per-incident, so under that multi-label counting every layer would be expected to rise, not fall.
What these views measure. Your recorded factor labels land in Scene most often. Not one of them lands in Process — protocol, training, documentation, policy. Across the same 1847 incidents, accepted narrative and structured-field observations describe process-level context in 573. Context such as tenure describes the work; it does not establish a process deficiency.
Two different measurements, not one re-weighted: the factor labels recorded on each incident, against every condition present in that incident’s own narrative or form fields. The recorded counts come from the factor labels your records actually used, and a label that names no layer clearly is counted in none of them. They do not say whether your form offers any given layer as an option: one that exists but went unused looks the same here.
03 · The map
What repeats across locations, and where it concentrates
THE MAP
Find what repeats across locations. Of 21 compared conditions, five do — and that’s the finding.
This matrix draws 8 of your 21 conditions across 5 locations — those carried by 5 or more incidents, plus every condition that repeats. Each cell shows how many incidents in that population carry the condition and the share of that population. 5 conditions repeat across at least two locations at or above the 5% display floor. Highlighted rows are those repeated conditions; the remaining rows show where each condition is concentrated.
Condition
All
Post moven=214
Station bayn=163
Residencen=142
Hospitaln=98
Roadwayn=71
a spotter was unavailablerepeats
144
6831.8%
4427%
1812.7%
1212.2%
—
time pressurerepeats
131
6128.5%
2213.5%
1913.4%
2121.4%
811.3%
backing the vehiclerepeats
118
5425.2%
4125.2%
1611.3%
77.1%
—
solo task completionrepeats
96
2210.3%
148.6%
4330.3%
1111.2%
68.5%
rails or restraints left downrepeats
74
—
95.5%
3826.8%
2424.5%
—
overtime or extended shift
61
3817.8%
84.9%
64.2%
—
34.2%
stretcher lock or latch failure
29
—
84.9%
128.5%
—
—
practice described as normal
22
104.7%
—
85.6%
—
—
Each cell shows the incident count, then that condition’s share of the incidents in that column. Highlighted rows are the conditions that repeat — present in at least 5% of the incidents in two or more columns. That is the rule for the highlight, not a ranking of importance.
How to read this comparison. Rows are conditions and columns are locations.
13 conditions below the 5-incident display floor are also true isolates — not drawn above, but worth knowing about: an unfamiliar vehicle (4 incidents), poor lighting (4 incidents), an uncooperative patient (4 incidents), an interrupted handover (3 incidents), training not recent (3 incidents), a bariatric load (3 incidents), a radio dead zone (3 incidents), a supervisor unreachable (2 incidents), a skipped pre-shift check (2 incidents), narrow stairs (2 incidents), a second call already pending (2 incidents), a missing restraint (1 incident), and an ambiguous policy (1 incident) — each sharing no condition with anything else in the record. See the full inventory below.
All conditions
The figure above draws 8 of these 21 conditions. Every condition your record actually carries is in this table, floor or no floor.
04 · The patterns
What keeps happening, scored inside its own population
01 THE PATTERNS
Conditions appearing together
These patterns describe conditions found together in the supplied records. Each has its own supporting incidents and comparison population. The decision brief above also includes descriptive groups selected for record review.
Two different measurements, not one re-weighted — the full disclosure and its limits are on “Your log as it stands.”
What keeps happening
A pattern is a pair of conditions that appears together on the same incident often enough to clear the floor below. Patterns shown here recur in at least 5 incidents eligible for pattern scoring. Recurrence establishes that they are present in the record — not that they are causal or severe. 604 of the 1,847 incidents this report reviews are eligible for pattern scoring; the other 1,243 are not. Eligibility needs a condition the pattern engine can compare against other incidents, which is a narrower test than having a condition captured at all. Every share in this section is taken over the population named with it, never over all 1,847. Counts are of 604 incidents eligible for pattern scoring.
This group collects pairs with overlapping conditions. Each row counts its own pair; the heading does not describe all conditions occurring together. Counts are not added because the same incident can appear in more than one pair.
Pattern group 1 of 1
Pairs involving time pressure
backing the vehicle · a spotter was unavailable · time pressure
54 Vehicle/Rig severity not recorded on this dataset
time pressure · overtime or extended shift
38 Vehicle/Rig severity not recorded on this dataset
Review the records behind each pair before choosing a response. Severity comparisons are unavailable for some pairs; their counts alone do not establish greater risk.
05 · What the file can't see
The limits, stated where you cannot miss them
02 WHAT THE FILE CAN’T SEE
Your record captures what happened. It never captures how badly.
No incident carries a recorded severity
None of your 1847 incidents carry a severity level in this export. Every finding in this report counts incidents; none of them can yet weigh one incident against another. If severity lives elsewhere — a claims system, an OSHA 300 log, a field that wasn’t shared — connect it, and every finding here can be weighed by what actually happened instead of only counted: the consequence axis takes a measured value.
No severity
1847 incidents 100%
Hatched = unrecorded.
The hazard was in your narratives — even where severity wasn’t
You recorded severity on 0 of 1847 — but the hazard conditions were in the records all along. Reading the narratives, we recovered a potential-severity profile for 1592 of your 1847 incidents: 335 carried critical- or high-potential hazard exposure.
This is potential, not recorded. Potential severity is exposure to harm computed from a rubric over the hazard conditions in each narrative — how hazardous the setup was, not a record of what outcome occurred. It does not replace the recorded-severity line above (still unrecorded on this dataset), and it is never priced. Recording realized severity is how you confirm which of these potential-high incidents actually landed — and 255 incidents carried too little in the narrative to grade either way, shown honestly above.
Your top factor is bigger than your system thinks it is
Free-text capture splits one problem into many names. This is from your actual data:
“No spotter”144 incidents
VS
“Spotter n/a”26 incidents
One small spelling or naming difference, and your top contributing factor is potentially undercounted by 18% — worth confirming with the people who wrote the labels. Nobody did anything wrong — this is what unstructured capture does to everyone. It also means every percentage in your current reporting is a floor, not a fact.
06 · Monday
Where to look next, and what to ask before changing anything
03 MONDAY
Where to look next
Use the decision brief to choose records for an operational review. The scored patterns offer another starting point: conditions shared across incidents. Bring both to the people who know the work.
The next two conversations
1
Walk the backing the vehicle · a spotter was unavailable · time pressure pattern with your ops lead.
54 incidents. We bring the incident-level drill-down; you bring the person who knows what those calls actually look like on shift.
2
Decide what your next 1847 incidents will know.
The System, Process, and Equipment dimensions carry no recorded factor label anywhere in this export. Review what is already recorded about shift time, supervision contact, and prior events, then decide which details need more consistent capture.
Questions from the decision brief
Each of these comes from the decision brief at the front of the report. The reason to ask is the record itself; the answer is not in it.
What movement, handoff, site layout, and vehicle-check work do these records represent?
Why it follows. 12 of the 31 repeat-ID early-duty incident records list an apparatus bay, garage, or carport.
Evidence that would answer it. shift starts and vehicle movements, facility or site identifiers and the current pull-out and changeover workflow.
What a next cycle could clarify. With that evidence, a later reading could weigh more facility movements early in the shift, location-labeling differences and mixed site control against what this cohort shows today. It would still be a description of the same records, more precisely conditioned.
Which maneuvers, sites, vehicles, and spotting practices do these records actually represent?
Why it follows. 8 of the 31 repeat-ID early-duty incident records name backing as the recorded cause.
Evidence that would answer it. stable site identifiers, the spotting practice in use at each site and vehicle and maneuver configuration.
What a next cycle could clarify. With that evidence, a later reading could weigh site mix, how the cause list is used and vehicle or assignment mix against what this cohort shows today. It would still be a description of the same records, more precisely conditioned.
What assignment, vehicle type, shift, route, facility movement, supervision, and reporting practice do these records represent?
Why it follows. 137 reported incidents involve personnel in their first six months.
Evidence that would answer it. headcount and assignment history by tenure, work and movement exposure by tenure and workflow and supervision context.
What a next cycle could clarify. With that evidence, a later reading could weigh different assignments or vehicle access, different reporting patterns and cohort size and staffing mix against what this cohort shows today. It would still be a description of the same records, more precisely conditioned.
What to check in the next record
Check which of these fields are already collected and which need more detail. This list does not establish that a field is missing.
Examples in this record
Fields to review for completeness
“No spotter”
Time on shift, and the last scheduled break before it
Free-text description of the event
Whether a second crew member was actually free at that moment
Equipment named, if mentioned
Equipment condition and last inspection date
Short factor labels
Whether the practice is usual, and whether anyone had raised it before
The patterns in Section 01 were found in your data. More consistent capture could make a later review easier; this report cannot establish when those patterns would have become visible.
This shape, run against your own export.
One approved, deidentified historical export goes in. These sheets, your incidents, and a live decision review with your team come back — fixed scope, no platform commitment.