Every contractor has a written safety policy, and most of them read alike. The difference between a safe site and a lucky one does not show up in the policy — it shows up in something far smaller: whether the near miss that hurt nobody gets recorded. A site logging twenty near misses a month is not a dangerous site, it is a site that can see itself. A site logging zero is not safe, it is blind.
Key takeaways
- A near miss is the only lesson you get for free — the same sequence of causes, with no injury, no cost and no stoppage.
- Zero reports is not a safety signal, it is a visibility signal: reporting falls when recording gets hard or when the reporter gets punished, not when the hazards go away.
- Typing and grading an incident is what turns a log into a triage tool, because the response to each combination is genuinely different.
- A root cause is not a person's name: "the worker was not paying attention" closes an investigation and changes nothing, while "no barriers around the slab opening" produces an action you can actually take.
- A corrective action with no owner and no due date is an intention, not an action — and the difference shows up three months later at the same location.
The near miss is the only free lesson
A scaffold plank falls from the fourth floor and hits the ground two metres from a worker. No injury, no stoppage, no investigation. That event is the single most valuable piece of information the site produced that day: the complete sequence that leads to a serious accident, which actually occurred, and stopped at the last step by coincidence alone. The coincidence may not be available next time.
And that information gets thrown away on most sites for a simple reason: recording it costs more than it appears to be worth. The foreman knows a report means a paper form, then an explanation to the project manager, then possibly a conversation about who is responsible. So he decides it "passed safely" and gets on with his day. From where he stands the decision is entirely rational, and its consequence is that the site has just discarded its only warning.
A report nobody fills in is a policy, not a control
Every safety system rests on one point: the minute somebody on site decides to record what they just saw. Everything after it — the investigation, the corrective action, the monthly report — does not exist unless that minute happens. Which makes the most important property of an incident form not its completeness, but its shortness.
- Filled in on a phone, at the location, not on a desktop hours later.
- Works with no signal — because the most hazardous places are very often exactly the places with no coverage: a basement, a concrete frame, a remote site.
- Takes a photo directly, because an image of the condition before it was fixed is worth a page of description.
- Never asks for what it can already infer: who reported, when, and on which project are all known already.
Type and severity: triage before investigation
An incident log with no classification is a list of paragraphs. Classifying on two axes — what happened, and how bad was it — is what makes it sortable and readable later:
| Axis | Values | What it decides |
|---|---|---|
| Type | Injury, near miss, property damage, environmental, fire, other | The nature of the investigation and who leads it |
| Severity | Low, medium, high, critical | How fast the response is and how far up it is escalated |
| Persons injured | A number | External reporting and regulatory obligations |
Conflating the two axes is the common error: "injury" gets logged as high severity by reflex, "near miss" as low. Reality inverts that regularly — a superficial cut on a finger is a low-severity injury, and a crane load dropping onto an area that happened to be empty is a critical near miss. The axes are independent because the response to them is independent.
A root cause is not a person's name
The easiest investigation is the one that stops at the nearest human: "the worker was not paying attention", "the supervisor was not wearing a helmet". That sentence closes the file and changes nothing — because the next worker at the same place will meet the same conditions, and the outcome will again depend on their attention alone.
| What went in the cause field | What action it produces |
|---|---|
| The worker was not paying attention | Nothing actionable — or a verbal warning |
| No barriers around the slab opening | Install barriers, and inspect every open penetration on site |
| The worker was not wearing PPE | Nothing — or a deduction |
| PPE is issued from a store 400 m from the site entrance | An issue point at the entrance, and stock monitoring |
A well-written root cause has one recognisable property: an action can be derived from it directly. If the recorded cause leads only to "remind the workers", the investigation stopped at the symptom and never reached the cause.
An action with no owner and no date is an intention
Every investigation ends with a list of what should be done. That list is where most safety systems die: written into minutes, agreed by everyone, and never revisited. Three months later the same incident happens, the old minutes get reopened, and the action proposed back then turns out to have been exactly right — and never executed.
A specific, checkable description
"Install barriers around all slab openings on the third floor", not "improve safety procedures on the upper floors".
One named owner
An action assigned to "the safety department" is assigned to nobody. A single name is what makes follow-up possible.
A due date
With no date an action cannot be late, and what cannot be late does not get chased.
A status that moves: open, in progress, completed
Status is what allows one genuinely useful question: which actions on this project are open and overdue right now?
Close the incident after its actions close
An incident marked closed with open actions beneath it means the file was tidied, not that the hazard was dealt with.
The one operational metric worth watching here is simple: how many corrective actions are open past their due date? That number describes a safety system more accurately than any incident count, because it measures what a site does after it knows — not how often it knew.
Numbers measured after the fact, and numbers that come before it
The best-known safety metrics — injury counts, lost days, incident frequency rates — are all lagging indicators: they measure what already happened. Useful for benchmarking and reporting, useless for prevention, since improving them requires the thing they measure to occur first.
- Ratio of near misses to injuries: a rising ratio is good news, because it means the site is seeing and recording.
- Average age of open corrective actions: the longer it runs, the weaker the preventive value of every investigation.
- Share of incidents with a usable root cause recorded: an investigation with no written cause will produce no action.
- Daily report consistency: a site with no daily record has no context to explain an incident when one occurs.
All of these derive from the same data the daily recording already produces — no separate system is needed. Their only precondition is that the recording actually happened at the time.
How this works in muqawil
The safety module in muqawil is built around the full cycle: a report from the site, classification, an investigation with a root cause, corrective actions with an owner and a date, then closure.
- Six incident types (injury, near miss, property damage, environmental, fire, other) and four severity levels, with persons injured, location and the date it happened.
- A report captured offline and synced later, with protection against a retry creating a duplicate.
- Photo attachments on the incident, and a root cause field independent of the description.
- Corrective actions with a bilingual description, an owner, a due date and a status (open, in progress, completed).
- A status on the incident itself — open, investigating, closed — and a safety report reading the log directly.
- A notification to the responsible managers the moment an incident is filed, not in a month-end summary.
Frequently asked questions
What is the difference between a near miss and an incident?
A near miss is an event that could have caused injury or damage and did not. Its chain of causes is identical to a real accident; only the outcome differs — which is why it deserves the same seriousness in investigation.
Does a rise in reports mean the site has become more dangerous?
Usually the opposite. Report counts measure visibility, not hazard. A rise after reporting is made easier means events that were already happening without a record have started to appear.
Should we record something that was fixed on the spot?
Yes. Fixing it solves the problem in that location; recording it is what stops it recurring on the project's other sites. The value of the log shows up the second time, not the first.
How do we keep safety reporting from becoming a blame tool?
By requiring the cause field to describe a condition rather than name a person, and by measuring the system on corrective actions closed rather than on violations recorded against individuals.
Are safety corrective actions the same as quality ones?
The logic is the same — description, owner, date, status — but the records are separate: a safety report concerns an event that happened to a person or property, while a quality non-conformance concerns work not built to specification.
Can an incident be filed on a site with no coverage?
Yes. The report is stored locally and sent when the connection returns — which is the most important practical requirement, because the most hazardous sites are frequently the worst covered.
Make recording a near miss easier than ignoring it
File safety reports from the site in muqawil — offline if you have to — and track corrective actions to closure with an owner and a date.