Why Uniformity Matters More Than Average Level

Why this matters

A retrofit raises the measured average in a pick aisle by a third and the pickers say it got worse. Both statements are true. An average is a statistic computed over a grid of points, and computing it discards the one thing the eye is actually responding to, which is how much the level changes between one point and the next one you walk to. A space can gain a third of its average and lose almost half of its darkest point in the same swap, and a report that prints only the average will show you the half that went up. This is not an edge case. It is the default outcome of a one-for-one fixture swap, because the replacement almost never has the same intensity distribution as the thing it replaced.

An average is only as good as the grid under it

An average illuminance is a number produced by a procedure, and the procedure has to travel with it. Change any of these and the same room yields a different average with nothing wrong anywhere:

• The plane and the height. Horizontal at 30 in above finished floor is a different quantity from horizontal at grade, and both differ from vertical at 5 ft facing the aisle. State it in the same clause as the number, every time.
• The grid spacing. A four-point grid in a bay lands wherever you put it. A grid on 8 ft centers samples the peaks and the valleys in roughly the proportion they occupy.
• Where the grid stops. A grid run to the walls includes the dark perimeter and reports a lower average than one bounded a fixture-spacing in from them. Neither is wrong; they are not the same statistic.
• Initial or maintained. A maintained average already contains its light loss factors, so it may not be compared against an initial one. That re-basing belongs to the light loss factor card.

Two contractors can survey the same warehouse on the same night, both honestly, and report averages a third apart purely on grid choice. The repeatable-reading card owns the procedure; this card is about what the resulting statistic can and cannot tell you.

The three ratios, and what each one catches

Three ratios get computed off the same grid, and they catch different failures.

Average to minimum. Divide the grid average by the lowest single reading on the grid. This is the usual acceptance ratio in a specification because it is stable: one dark point in twenty barely moves the average but moves this ratio immediately. It answers "is any part of this space substantially worse than the space as a whole."

Maximum to minimum. Divide the highest reading by the lowest. This is the harsh one and it catches the pattern the average cannot see at all: a space with bright pools under fixtures and dark valleys between them can hold a perfectly respectable average while the ratio runs into double figures. It answers "how far does a person's eye have to adapt as they walk this path."

The step between adjacent grid points. This one usually has no line in the specification, and it is the one occupants actually report. A person walking a 96 ft aisle sampled at twelve stations on 8 ft spacing crosses eleven transitions, and if each is a factor of two the complaint arrives whatever the average says, because the eye adapts over seconds and the walk is faster than that. Where the specification is silent, print the largest adjacent step anyway.

Note what none of the three contain: a code minimum. For construction work, 29 CFR 1926.56(a) and its Table D-3 set enforceable single-value floors, 5 foot-candles for general construction area lighting and 3 foot-candles for general construction areas including excavation, waste areas and accessways, with no uniformity ratio attached to either. Meeting an OSHA floor is a legal condition for the work to proceed and says nothing about whether the space is usable. An IES recommended illuminance, in the edition your specification or employer standard names, binds only through that document and is a design target rather than a floor, which is why it carries a plane, a height, a maintained basis and a ratio while the OSHA number carries none of that.

Why spacing over mounting height decides the ratio

Uniformity is mostly geometry, and the governing quantity is the fixture spacing divided by the mounting height above the work plane, not above the floor. That distinction matters: a 24 ft mounting height over a 30 in work plane gives 21.5 ft of throw, not 24.

  fixture fixture
     |                             |
     |<----------- S ------------->|
     |                             |
     H = mount height ABOVE the work plane
     |                             |
   __|_____________________________|__  work plane
    peak valley peak

  small S/H  ->  shallow valley, low max-to-min
  large S/H  ->  deep valley, high max-to-min

A luminaire's photometric report publishes a spacing criterion for this, the manufacturer's statement of the largest S over H at which their distribution still overlaps. It is derived for a flat, unobstructed plane with no racking, no partitions and no vertical surfaces in the way, so in an aisle walled by 20 ft of steel rack the real overlap is worse than the published criterion and the number is an approximation, not a calculation. Say that out loud when you use it.

The gate: a criterion with three slots

An acceptance criterion for illuminance has three slots: a level, a plane with a height and a basis, and at least one ratio. A criterion that fills only the first slot cannot be honestly passed or failed. The numbers in each slot come from the project specification, or from the IES recommended practice that specification names in its own edition, never from this card. What travels is the shape. Below, one criterion in that shape runs against two jobs: in the first the average improves and the job fails, in the second the average worsens and the job passes.

Case one: the average went up and the aisle failed

Warehouse pick aisle, 96 ft long, six existing high-intensity discharge low bays on 16 ft centers at 22 ft mounting height. Twelve stations at 8 ft spacing, horizontal at 30 in above finished floor, all readings maintained, taken three years into service. Before any of it the aisle was blocked at both ends and lift traffic rerouted with the supervisor's sign-off under the powered industrial truck rules at 29 CFR 1910.178, with high-visibility apparel worn throughout, because a person walking a grid with their eyes on a meter is a pedestrian in a truck aisle.

Specification: maintained average not below 20 fc horizontal at 30 in above finished floor, maximum to minimum not above 6 to 1.

The average rose from 18.6 to 24.8 fc, which is 24.8 / 18.6 = a factor of 1.33, up 33 percent. The minimum fell from 11.2 to 6.1 fc, which is 6.1 / 11.2 = 0.54, down 46 percent. Maximum to minimum went from 2.36 to 9.51, worse by a factor of 9.51 / 2.36 = 4.03.

Verdict against the criterion: the existing installation failed the level at 18.6 against 20 fc and passed the ratio at 2.36 against 6 to 1. The retrofit passed the level at 24.8 and failed the ratio at 9.51. The job was sold on the level and delivered on the level, and the aisle got worse.

The mechanism is not mysterious. The replacement had a narrower distribution, so more of its flux landed directly under itself and less reached the midpoint between fixtures. At an unchanged 16 ft spacing over 19.5 ft of throw above the 30 in plane, the layout was already relying on wide overlap and the new optic removed it.

Case two: the average went down and the lot passed

Employee parking area, existing four poles, resurveyed on a 24-station grid, horizontal at grade, maintained. The survey ran after close with the lot coned off and the crew in high-visibility apparel, because a person crouching at grade in a dark lot is invisible to a driver until the headlights are on them.

Specification: minimum not below 0.4 fc horizontal at grade maintained, average to minimum not above 4 to 1.

The average fell from 1.9 to 1.5 fc, which is 1.5 / 1.9 = 0.79, down 21 percent. The minimum rose from 0.2 to 0.5 fc, a factor of 2.5. Average to minimum improved from 9.5 to 3.0.

Verdict: the existing lot failed the minimum at 0.2 against 0.4 fc and failed the ratio at 9.5 against 4 to 1. The rebuild passed both, at 0.5 fc and 3.0 to 1, on a lower average than the thing it replaced. Had the specification filled only the level slot, the rebuild would have read as a downgrade.

What flips this

The ratio does not always govern. Where the task is a single fixed position rather than a path - a bench, one machine station, an inspection table - a person adapts once and stays adapted, and the level at that position is the whole question. Uniformity is what matters over a traversed area, and a tight ratio specified across a room whose occupants never leave their chairs buys nothing. The other flip is daylight: a perimeter zone with glazing runs an enormous max-to-min at noon and nobody complains, because the bright end is the window and the eye reads it as a source rather than as a task surface being unevenly lit.

Checking your own figures

• Level and ratio evaluated separately, both printed. Aisle: level 24.8 fc against 20 fc, passes; ratio 9.51 against 6 to 1, fails. Lot: minimum 0.5 fc against 0.4 fc, passes; ratio 3.0 against 4 to 1, passes.
• Direction words check against the printed series. Aisle average rose 18.6 to 24.8; aisle minimum fell 11.2 to 6.1. Lot average fell 1.9 to 1.5; lot minimum rose 0.2 to 0.5. Two of the four moved against the average.
• Percentages carry their base. 24.8 / 18.6 = 1.33 on a base of 18.6 fc; 6.1 / 11.2 = 0.54 on a base of 11.2 fc.
• One basis throughout. Every value on both jobs is maintained, so no light loss factor is applied anywhere in this card.
• Plane and height printed on every value. Aisle horizontal at 30 in above finished floor; lot horizontal at grade. The two are never compared to each other.
• Throw height used, not mounting height. 22 ft mounting over a 30 in plane is 19.5 ft of throw, and the S over H discussion uses 19.5, not 22.
• Three ratios named, three ratios enumerated. Average to minimum, maximum to minimum, and the largest adjacent-station step, which is printed for the aisle at 1.4x and 3.2x and is absent from the lot table because the lot criterion did not ask for it.

References

• 29 CFR 1926.56(a) and Table D-3, minimum illumination intensities for construction work areas, which are single-value floors carrying no uniformity requirement
• 29 CFR 1910.178, powered industrial trucks, for pedestrian and traffic control while surveying an active aisle
• IES recommended practice for the space type, in the edition your specification or employer standard names, which is where target levels and acceptance ratios live and which binds only through that document
• Manufacturer photometric test report for the specific luminaire, which is the only source for its distribution and its published spacing criterion
• See related: How to Take a Light Level Reading Somebody Else Can Repeat; What a Light Loss Factor Is Accounting For; What the Inverse Square Law Does to a Lighting Layout