Why Systems Built to Find Fault Cannot Reward What Goes Right, and How the SMV Cuts Through
Ask any safety manager who has tried to balance an observation programme: the request for positive observations goes out, and what comes back is a trickle. "Crew wearing PPE correctly." "Housekeeping good in Bay 4." Compliance restated as praise, filed to make a quota, teaching nobody anything.
Meanwhile the negative side of the programme hums along, finding faults by the hundred. The struggle to find good observations is close to universal, and it is routinely misdiagnosed as a workforce problem, or a motivation problem. It is a design problem.
Most observation systems descend from unsafe-act and unsafe-condition auditing, and the ancestry shows. The form leads with deviation. The observer training centres on spotting what is wrong. The metrics count problems found, and a diligent observer is one who finds many.
Inside that design, "good" has no natural home. There is no field for it beyond a comments box, no classification for it beyond the absence of fault, and no follow-up path for it at all: a fault generates an action, an owner and a due date, while a good observation generates a nod.
The asymmetry teaches. Observers learn that finding fault reads as rigour while reporting good reads as softness, and crews learn that the observer only ever arrives to subtract. The same tuning failure undermines observation reporting programmes at the data level: a record built from faults describes everything except the best work being done.
Behaviour that is never noticed is never rewarded, and behaviour that is never rewarded fades. The crew that quietly rearranged a task to keep hands clear of pinch points gets nothing; the crew that got caught taking the same shortcut everyone takes gets a finding. Over time the programme extinguishes exactly the behaviour it claims to want.
Worse, the improvements go underground. A better method disclosed is a deviation admitted, so crews keep their good ideas out of sight, unassessed and unshared. The organisation keeps paying for innovation it never receives.
The instinctive fix is to demand more "good" observations to balance the "bad" ones, and it fails because the axis is wrong. Good and bad are judgements, and judgements invite performance: observers manufacture praise to hit the ratio, and the trickle of PPE-compliance platitudes is the result.
An observation is not good or bad. It is a difference, between work as planned and work as done, and the difference carries information in whichever direction it runs. The productive question is never "was this good or bad?" but "what varied, and what is the effect of the variation?"
That question is precisely what the Safe Method Variation asks. The SMV records the difference (SMV = WAD − WAP) and classifies it by effect: Increased Opportunity and Enhanced Resilience where the variation improves on the method, Increased Vulnerability and Decreased Reliability where it degrades the task or the system.
The dichotomy dissolves because one instrument captures everything. The observer is no longer choosing between fault-finding and praise; every variation goes through the same gate and comes out signed by evidence. A "good observation" stops being a compliment and becomes a rigorous object: a validated variation that outperforms the approved method.
And reward stops being a poster campaign. A validated positive is adopted into the method and credited to the crew that found it, which is the one reward that proves the programme means it: the organisation visibly changed how it works because of what a crew showed it. It is also worker participation of the ISO 45001:2018 Clause 5.4 kind, evidenced rather than asserted.
The struggle to find good observations ends when the programme stops looking for "good" and starts measuring difference. The good was always there, being done quietly on every shift. The system just had no way to see it.
Because most systems descend from unsafe-act auditing: the forms lead with deviation, observer training centres on fault, and the metrics count problems found. "Good" has no field, no classification and no follow-up path, so what comes back is compliance restated as praise rather than anything worth learning from.
Because good and bad are judgements, and demanded judgements get performed: observers manufacture praise to hit the ratio, which is why forced-positive programmes fill with PPE-compliance platitudes. The fix is changing the axis from judgement to difference, recording what varied from the planned method and classifying its effect.
Rigorously: a validated variation that outperforms the approved method, classified as Increased Opportunity (an opportunity for a better system, beyond the task) or Enhanced Resilience (reinforcing the current system at the same cost). That definition is earned through risk assessment rather than granted as praise, which is what makes rewarding it defensible.
Adoption with credit: the validated improvement becomes the approved method, and the crew that found it is named. It outperforms prizes and points because it proves the programme means it, the organisation visibly changed how it works because of what a crew showed it, and that is what keeps disclosure flowing.