How to Work Out Whether a Lighting Complaint Is Level or Quality

Why this matters

"It is too dark in here" is a sensation, not a diagnosis, and it comes out the same way whether the cause is a dimming scene somebody set two years ago, five dead luminaires, four years of uncleaned lenses, or a distribution that was never right for the task. Those four have nothing in common except the sentence. A level problem and a quality problem are different jobs and different parts, and quoting the wrong one is how a shop ends up replacing sixty luminaires to fix a setting. This is the one-visit reading set that separates them before anybody prices anything.

The order is a dependency order, not a ranking

Six readings, and they run in this sequence because each one contaminates every reading after it if you take them out of order. That is the only claim being made about the order. It is not a ranking by cost or by likelihood.

1. Control state, because every illuminance you measure is a measurement of whatever the controls were doing at that instant.
2. Outage count, because dark units move the average and the minimum by different amounts.
3. Horizontal illuminance on the work plane, which needs both of the above settled.
4. Vertical illuminance at the task, which is only interpretable as a ratio against a horizontal reading from the same visit.
5. The largest adjacent-point step, computed off the grid you already took.
6. A cleaned-fixture probe, which tells you how much of any shortfall is recoverable.

Before any of it, ask what the complaint actually names. Write down the words. "The shelf", "the bench", "the screen" and "the aisle" are four different planes, and if you measure a plane the complaint never mentioned you will get a number that is true and useless.

1. Record the control state before you change it

Read the controller, the wall station, the schedule and any sensor setpoints, and write them down at their found state. Then, and only then, take a grid reading at found state and a second at full output.

Skip this and the visit is void, because a grid taken at a dimmed setting measures the setting rather than the installation, and what you carry into the rest of the diagnosis is a number about a decision somebody made. What a control system is being asked to do is its own card's subject; here you need the found state, the full-output state, and the difference.

Driving output to full is an instruction that moves equipment, so confirm nothing else rides on that scene or relay before you send it: a lighting controller that also drives a shade, an exhaust interlock or a process signal is not a place to experiment during production.

2. Count the dark units before you trust any average

Walk the space with the lights at full output and count units that are out, strobing or visibly dim, and note where they are. A dispersed outage scales the average roughly in proportion to the fraction lost. A cluster does something else entirely: it barely touches the average and puts a hole in the minimum, so the same count produces a level finding in one case and a uniformity finding in the other. Record the count and the pattern, not just the count.

3. Take the horizontal grid on a stated basis

Grid spacing, plane, height above finished floor, extent, meter, time of day and daylight state all travel with the number or it is not a measurement. The repeatable-reading card owns that procedure and it is worth following exactly, because this reading is the one that will be compared against a design figure written by somebody else.

A field reading is a maintained reading. You are measuring the installation in whatever loss state it is actually in, which is what "maintained" means. It only approximates an initial figure when the source is new or the optics have just been cleaned. So when you compare against a design, move the design onto the reading's basis rather than moving the reading; the light loss factor card owns that operation and the direction it runs.

4. Take the vertical reading at the task, on the same visit

Sensor turned through ninety degrees, at task height, facing the surface the complaint named. Take it on the same visit with the same meter as step 3, because the useful output is the ratio of the two, and a ratio divides out any fixed multiplicative error the instrument carries as well as the installation's whole maintenance state. Two readings on two visits do not have that property.

5. Compute the largest adjacent step

Off the grid from step 3, take every pair of neighbouring points and find the largest ratio between them. Most specifications are silent on this and it is the number occupants report, because the eye adapts over seconds and a person crosses the transition faster than that. The uniformity card owns what the standard ratios catch; this step costs nothing once the grid exists.

6. Probe the dirt term

Clean one luminaire's optic, re-read the station directly under it before and after, and take the quotient. That single number tells you how much of a shortfall a cleaning program recovers, and it is the difference between a maintenance quote and a design quote. Reaching the optic means a lift rated for the height with the fall protection the equipment requires under 29 CFR 1910 Subpart D in general industry or 29 CFR 1926 Subpart M in construction, and where settled dust on the optic comes off a process, check the material's safety data sheet and damp-wipe rather than dry-brush, because putting that dust into the breathing zone is an inhalation route that needs a respiratory control under a 29 CFR 1910.134 program, not a glove.

The visit: a light industrial floor, sixty luminaires

The owner reports the floor has been getting darker for months and is holding a quote to re-lamp all sixty units. Design basis on the drawings: 40 fc maintained, horizontal at 30 in above finished floor, light loss factor 0.75.

Step 1, control state. The controller is found on a 65 percent scene, undated, no commissioning record. Found-state grid average 24 fc. Full-output grid average 36 fc. The quotient is 24 / 36 = 0.667 against a 65 percent setpoint, a 2.6 percent difference that the driver's own dimming curve accounts for, so the setting explains the found-state reading and no second cause is needed for it.

Step 2, outages. Five of sixty units dark, dispersed across the bays, no two adjacent. Fraction working 55 / 60 = 0.917.

Step 3, horizontal. 36 fc at full output with five out, on a nine-point grid per bay, horizontal at 30 in above finished floor, maintained basis. Because the outages are dispersed, correcting to a full complement is defensible: 36 / 0.917 = 39.3 fc. That correction would not hold if the five were clustered, in which case the average barely moves and the minimum carries the loss instead.

Compare on one basis. The design is 40 fc maintained and the corrected reading is 39.3 fc maintained, so the installation as designed is delivering 98 percent of its target. The 0.75 light loss factor is not applied to either side, because both figures are already maintained.

Step 4, vertical. Task face at 4 ft above finished floor, same visit, same meter, same five units out: 16 fc. Ratio against the uncorrected horizontal from the same visit, 16 / 36 = 0.44, which is unremarkable for this layout and does not point anywhere. Correct one side of a ratio and you lose the shared instrument and shared installation state that made it worth taking.

Step 5, adjacent step. Largest neighbouring pair on the grid, 1.4 to 1. No finding.

Step 6, dirt probe. One optic cleaned: 34 fc before, 37 fc after at the same station, a dirt term of 34 / 37 = 0.919. Real, worth a cleaning interval, and about 8 percent of the light, which is not the complaint.

The separation. The gap the owner lives with is 40 - 24 = 16.0 fc, and it splits in the order the fixes would be applied, each measured from where the previous one left off. Restoring the scene, with the five units still out, moves 24 to 36 fc, so the setting owns 12.0 fc. Replacing the five dispersed units moves 36 to 39.3 fc, so outages own 3.3 fc. What remains against design is 40 - 39.3 = 0.7 fc. Those add to 16.0. Apply them in the other order and the split shifts slightly, because the two are multiplicative rather than additive; what does not shift is that it is a level complaint, that it is entirely recoverable, and that none of it is a design problem. Steps 4 and 5 came back clean, which is what allows the quote to be a maintenance quote.

What was done. The scene was restored to full and documented with a date and a name so the next person does not rediscover it, and the five dark units were replaced. Replacing them is electrical work: de-energize the branch circuit, lock and tag, and prove dead live-dead-live per NFPA 70E-2021, 120.5, in the edition the employer's electrical safety program adopts, under 29 CFR 1910.333(b)(2), before a hand enters the housing. A cleaning interval went on the customer's calendar rather than into the invoice.

Confirmation. Same grid, same plane and height, full output, full complement: 39.6 fc maintained, horizontal at 30 in above finished floor, against the 40 fc design. Sixty luminaires stayed on the ceiling.

What flips the reading

Two conditions change what the same numbers mean. If the outages are clustered, the step-3 correction is invalid and the finding moves from level to uniformity even at an identical count, because the average survives and the minimum does not. And if the space runs any daylight-responsive control, the found-state and full-output readings must be taken with the daylight contribution recorded separately, by reading once with the electric circuit opened at its panel, or every number in the visit belongs to the weather.

Checking your own figures

• The dimming quotient is derived, not assumed. 24 / 36 = 0.6667 against a 0.65 setpoint, a difference of 0.0167 on 0.65, which is 2.6 percent.
• The outage correction states its condition. 55 / 60 = 0.9167; 36 / 0.9167 = 39.27, reported as 39.3 fc, and the text names dispersal as the condition under which that division is allowed.
• Both sides of the design comparison are maintained. 39.3 fc measured against 40 fc design is 39.3 / 40 = 0.9825, reported as 98 percent, with the 0.75 light loss factor applied to neither.
• The three contributions sum to the gap, each measured from the previous step. 36 - 24 = 12.0; 39.3 - 36 = 3.3; 40 - 39.3 = 0.7; total 16.0, against 40 - 24 = 16.0. Neither of the first two is measured against 39.3, which would double-count the other.
• The dirt probe quotient. 34 / 37 = 0.919, so 1 - 0.919 = 8.1 percent, reported as about 8 percent.
• The vertical figure uses a same-visit pair of raw readings. 16 / 36 = 0.444, both taken on one visit with one meter in one outage state. The outage-corrected 39.3 fc is not used in any ratio, because it is a derived figure rather than a reading.
• Six steps promised, six steps given, and every illuminance in the visit carries plane, height and basis.

References

• 29 CFR 1910.333(b)(2) with NFPA 70E-2021, 120.5, in the edition your employer's electrical safety program adopts, for the luminaire replacements; 29 CFR 1910 Subpart D and 29 CFR 1926 Subpart M for the work at height; 29 CFR 1910.134 where cleaning disturbs settled process dust identified on the safety data sheet
• 29 CFR 1926.56(a) and its Table D-3 for enforceable construction illumination floors, which carry no plane and no maintained basis; an IES recommended illuminance, in the edition your specification names, binds only through that specification and is a design target rather than a floor
• See related: How to Take a Light Level Reading Somebody Else Can Repeat; What a Light Loss Factor Is Accounting For; Why Uniformity Matters More Than Average Level; What a Lighting Control System Is Being Asked To Do; What Glare Is, and the Two Kinds That Matter