Skip to content
SentinelHSE
Near miss reporting

The injury you prevent never appears in your numbers.

Near misses are the only safety data you get before someone is hurt. SentinelHSE removes the friction that stops people reporting them — then turns what they report into owned, dated, evidenced action.

A low near-miss count is not safety. It is silence.

When a site reports two near misses a quarter, nobody concludes the site is safe. They conclude nobody is writing them down. The events still happen — the unchocked trailer, the defeated interlock, the harness clipped to nothing — they just arrive as information only after somebody is hurt.

That is the whole problem. Every serious injury was preceded by conditions somebody saw. The question is not whether your people notice. It is whether telling you costs them anything.

Capture

Reporting has to cost less than staying quiet

Four facts about the form, not four promises. Every one of them exists because a programme dies at the point where filing a report is more effort than walking past.

The whole form, every field of it, and the record it becomes. Hover a card to light that control on the screen — or hover the screenshots themselves to magnify them, and click to open one full size.

The New Near Miss form: name, an Open status chip, category, location, department, a pre-filled date, time, observer and likelihood, then description, immediate action, corrective action and the create button

Filing it, in under a minute

Near Miss #26 on the phone: pedestrian crossing the forklift aisle, ranked High, with share, download and edit controls, category, location, department, when it occurred, the observer, the likelihood, the description and the immediate action taken

What it becomes

Screenshots from a working build. Illustrative records.

Under a minute, one hand

Photo, category, what could have happened. On a cheap Android, on a weak signal, without going to find a laptop.

No blame surface

The form captures the condition, not a culprit. There is no field for whose fault it was, because that field is what empties a register.

The exact area, not the site

Locations resolve to Warehouse A, Aisle 4 rather than “Houston Plant”. A pattern you cannot locate is a pattern you cannot fix.

The reporter sees it close

They are told what happened to their report. Nothing kills a programme faster than reports disappearing into a void.

The shape

The base is where you still have a choice

One site, one month. The tiers are not a fixed ratio and this page will not pretend otherwise — what matters is which tier you are able to learn from.

  1. 4Recordable injuriesHarm already done
  2. 14IncidentsSomething happened
  3. 48Near missesThe outcome was still open
  4. Observations & hazardsA condition somebody noticed

Illustrative figures from the product demo. Near-miss to incident ratio 3.4 : 1. Every tier below the top is an event where the outcome was still open. A programme that only records the top tier is a programme that only learns from harm.

Observations sit at the base without a count on purpose. The ratios between these tiers are a property of one site in one month — not a law of nature, whatever the version of this triangle you were shown in training implied.

The register

One register. Every report. Ranked by what could have happened.

A live register from a working month, filterable by potential severity, status, site and category.

Near-miss register

One working month across four sites, ranked by what could have happened

Illustrative data — not a customer
Potential severity
  • NM-7
    CriticalClosed

    Employee reaching into a conveyor to clear a jam

    Date
    Jul 25
    Category
    Unsafe Behaviour
    Site
    Houston — Warehouse A
    Likelihood
    Possible
    Reported by
    Robert Pierce
  • NM-28
    CriticalClosed

    Guard interlock defeated with a cable tie

    Date
    Jul 11
    Category
    Machinery / Equipment
    Site
    Gary — Weld Bay 2
    Likelihood
    Possible
    Reported by
    Victor Ramos
  • NM-17
    CriticalUnder review

    Working at height with an unclipped harness

    Date
    Jul 25
    Category
    Working at Height
    Site
    Baton Rouge — Tank Farm B
    Likelihood
    Possible
    Reported by
    Derek Coleman
  • NM-23
    CriticalOpen

    Spark from an angle grinder near the solvent store

    Date
    Aug 04
    Category
    Fire / Explosion
    Site
    Houston — Chemical Yard
    Likelihood
    Unlikely
    Reported by
    Angela Reyes
  • NM-47
    CriticalClosed

    Confined space barrier removed while the space was open

    Date
    Jul 14
    Category
    Unsafe Behaviour
    Site
    Houston — Chemical Yard
    Likelihood
    Unlikely
    Reported by
    Angela Reyes
  • NM-6
    HighClosed

    Missing guard on the bench grinder

    Date
    Jul 28
    Category
    Machinery / Equipment
    Site
    Houston — Maintenance Shop
    Likelihood
    Likely
    Reported by
    Anthony Russo
  • NM-26
    HighOpen

    Pedestrian crossing the forklift aisle without looking

    Date
    Aug 07
    Category
    Vehicle / Traffic
    Site
    Houston — Warehouse A
    Likelihood
    Likely
    Reported by
    Tanya Brooks
  • NM-5
    HighClosed

    Gas cylinder stored unsecured against the wall

    Date
    Jul 31
    Category
    Fire / Explosion
    Site
    Gary — Weld Bay 2
    Likelihood
    Possible
    Reported by
    Victor Ramos
  • NM-10
    HighClosed

    Eye wash station blocked by stacked cartons

    Date
    Jul 23
    Category
    Chemical / Spill
    Site
    Houston — Chemical Yard
    Likelihood
    Possible
    Reported by
    Angela Reyes
  • NM-18
    HighUnder review

    Damaged flexible hose on the caustic transfer line

    Date
    Jul 26
    Category
    Chemical / Spill
    Site
    Baton Rouge — Drum Storage
    Likelihood
    Possible
    Reported by
    Derek Coleman
  • NM-43
    MediumClosed

    Spill kit found empty when needed

    Date
    Jul 05
    Category
    Chemical / Spill
    Site
    Baton Rouge — Drum Storage
    Likelihood
    Likely
    Reported by
    Emily Carver
  • NM-19
    LowClosed

    Housekeeping — offcuts left in the walkway

    Date
    Jul 01
    Category
    Slip / Trip / Fall
    Site
    Gary — Weld Bay 2
    Likelihood
    Likely
    Reported by
    Victor Ramos

Showing 12 of 12 rows — twelve of 48 reports in the period. Nobody was hurt in any of them. Five had a credible worst outcome of a fatality.

Assessment

Rank by what could have happened, not what did

Every report gets a potential severity and a likelihood. The product multiplies them into a score, and the score decides whose week it belongs in.

The near-miss register in the web command centre: 48 total, 2 open, 16 under review, 30 closed, with severity and risk on every row

All 48 reports, by what could have happened

Potential severity down, likelihood across. Shading is severity × likelihood.

Illustrative
Near-miss reports plotted against potential severity and likelihood. Each cell gives the number of reports and its risk score.
PotentialRareUnlikelyPossibleLikelyAlmost certain
Critical0score 52score 103score 150score 200score 25
High0score 40score 87score 122score 160score 20
Medium0score 33score 615score 911score 120score 15
Low0score 20score 43score 62score 80score 10

A near miss and a fatality can be the same event with a different ending. The conveyor jam above hurt nobody; the same reach with the drive running is an amputation. Ranking by potential puts the five critical reports in front of the plant director this week, instead of after the event that proves the point.

Severity × likelihood

Each report carries both. The product multiplies them and prints the result on the record — Medium × Possible gives a score of 6. Nothing about the ranking is a matter of opinion once the two fields are set.

The scale is yours

Four severities by five likelihoods here. If your matrix is 5×5, or your labels differ, that is configuration an administrator changes — and the scores recalculate against your model, not ours.

The score sets the queue

A manager works down from the top-right corner. The five critical reports get looked at this week, rather than after the event that proves they mattered.

Indicators

The number to watch is how many people report

Lagging indicators count the harm you already did. Leading indicators are the ones you can still act on — and participation is the one that moves before any of the others.

Lagging

harm already done

TRIRTotal Recordable Incident Rate
recordable cases × 200,000 / hours worked
LTIFRLost Time Injury Frequency Rate
lost-time injuries × 1,000,000 / hours worked

Never quote either without its multiplier. Conventions differ, and a rate without one is not a rate.

Leading

harm not yet done

Near-miss reporting rate
Reports per person per period. The number that moves first when the friction goes.
Participation
How many people filed at least one. One enthusiast filing forty is not a programme.
High-potential reports
Ranked by what could have happened. These are the ones worth a manager’s week.
Actions closed on time
The percentage that did not run late — and the reason people keep reporting.

Illustrative — figures from the product demo

67%

participation — 10 of 15 staff filed at least one report

48

reports in the period

63%

already closed out

A rising near-miss count is a programme working, not a site getting worse — and the board needs to be told that before the first spike, not during it.

Follow-through

A report nobody acts on trains people to stop reporting

The second report is filed on the strength of what happened to the first one. This is the whole retention mechanism of a near-miss programme.

  1. 01

    Reported

    From the floor, on a phone, in under a minute.

  2. 02

    Ranked by potential

    Severity × likelihood gives a score, and the score sets the queue.

  3. 03

    Immediate action recorded

    What was done on the spot, captured while it is still true.

  4. 04

    Corrective action raised

    An owner and a due date, against the hierarchy of controls.

  5. 05

    Chased automatically

    Overdue actions escalate on their own. Nobody has to remember.

  6. 06

    Closed and verified

    Verified effective before the record closes.

  7. 07

    Briefed at the toolbox talk

    And the reporter is told what happened to their report.

The near-miss list on the phone, grouped by state: two open reports flagged for attention, sixteen in review and thirty closed

Grouped by state, so the two still open are the first thing anyone sees — including the person who filed them.

Evidence

What an auditor actually asks for

Buyers respond to clause numbers rather than feature lists, because the clause is the artefact they have to produce.

  • ISO 45001:20186.1.2

    Hazard identification & risk assessment

    Where the evidence comes from

    Near-miss and hazard reporting, ranked by potential severity × likelihood

  • ISO 45001:20185.4

    Consultation & participation of workers

    Where the evidence comes from

    Reporting from the field by anyone on site, and feedback to the reporter

  • ISO 45001:20189.1

    Monitoring, measurement & analysis

    Where the evidence comes from

    Leading indicators, participation, and trend by site and category

  • ISO 45001:201810.2

    Incident, nonconformity & corrective action

    Where the evidence comes from

    Corrective actions with owners, dates and verified closure

  • ISO 45001:20187.4

    Communication

    Where the evidence comes from

    What was learned, briefed at the next toolbox talk

SentinelHSE supports evidencing these requirements. Certification is an audit of your organisation, not of software — any vendor who tells you their product makes you compliant is selling you a problem.

Objections

The five things you are already thinking

Answered the way we would answer them in the room, including where the risk is real.

Yes, and that is the point. A rising near-miss count is a functioning programme, not a deteriorating site — but only if everyone has agreed that in advance. Settle with your board which direction is good before the first month’s spike, or the spike will be used against you and the programme will not survive it.

The form captures the condition, not a culprit — there is no field for who was at fault. That helps, and it is not sufficient. If a near-miss programme is used for blame it will die within a month whatever software you buy, so the first thing to fix is what happens to the second person who reports something awkward.

Often true, and usually the app’s fault. Adoption is the design constraint here: under a minute, one hand, a cheap Android, a weak signal. The honest test is not a demo — run one crew for a month and measure the reporting rate before and after. If it does not move, it has not worked, and you should not roll it out.

Yes. It can run in our cloud, inside your own cloud account under your policies and retention rules, or entirely on your own servers for sites that cannot send safety data outside at all. Everything exports, so leaving stays a decision you can make later.

Yes. Categories, potential-severity levels, likelihood scales and form fields are configuration, changed by an administrator without development work. If your matrix is 5×5 rather than 4×5, the platform should learn your model rather than the other way round.

See it on your own data

Bring a month of your near misses — even if there are only three. We'll load them and show you the register, the risk matrix and the participation rate your board would see.

45 minutes. Or run a four-week pilot with one crew and measure the reporting rate yourself.