Audit the sample unit, not the showroom. Five measurable tests — leak rate, irradiance stability, temperature accuracy, run-to-run repeatability and response time — fit into one site visit and produce a pass or fail you can write into the purchase order. A system quoted at ≤0.1 μmol/h (O₂) either holds that number under your own blank run or it does not.
A demonstration proves a system can run; only a test on the unit you will receive proves yours does. The difference between two units usually appears in the joints, the seals and the calibration, not on the front panel.
Schedule the visit around measurements, not presentations. Bring your own catalyst, gas cylinder, calibrated meter and a laptop that logs. Run the system as you will run it at home: a full evacuation, an overnight blank, a step change in setpoint.
Write the pass criteria down before the visit and put them in the purchase order. A criterion that lives only in your notebook is a preference; a criterion in the contract is a specification.
These five cover the failure modes that destroy data most often, and all can be run with equipment you can carry.
| Test | What you do | Pass criterion | What it protects |
|---|---|---|---|
| Leak rate | Evacuate, isolate, log the pressure rise; then a catalyst-free blank under light | Blank flat within analyzer noise; leak rate at or below the quoted value | The lowest production rate you can believe |
| Irradiance stability | Meter at the reactor plane, log for a real run's length | Drift inside your own repeatability band | Catalyst effect separated from lamp effect |
| Temperature accuracy | Independent probe in the reactor against the setpoint | Within the quoted tolerance, ±1 °C on stages specified to it | Any rate compared across temperatures |
| Repeatability | Three identical runs, same catalyst and sampling | Spread smaller than the effect you intend to publish | Whether a difference between samples is real |
| Response time | Step the setpoint; inject an aliquot and time the readout | Settles inside the interval you log at | Whether transients are visible at all |
The table is ordered by cost of failure, not by difficulty. A leak test takes longest and settles the most: every rate you report is bounded by it. Repeatability is the test most often failed, because it exposes the whole chain — lamp, seal, sampling and detector — at once.
Run the blank before the catalyst. A catalyst-free system under illumination, filled with the same medium and purged the same way, shows what the rig produces on its own and what it draws in from the room.
Two numbers matter. The first is the leak rate under isolation: evacuate, isolate, and watch pressure long enough for a slow leak to show. The second is the blank signal under light, which should sit inside the detector noise. A quoted ≤0.1 μmol/h (O₂) is a claim about the first; the second is what you measure.
Line volume matters too. A 65 mL line volume responds quickly and dilutes little; a large dead volume does the opposite, and no detector resolution recovers it. Ask for the volume, then confirm it with a known injection.
Measure where the reaction happens, not at the port. Irradiance at the exit aperture is a brochure number; irradiance at the reactor plane, through your window and at your working distance, is the number that governs your rate.
Log for at least the length of a typical run. Look at the shape of the curve, not the maximum: a source that reaches its value and holds it is usable, while one that climbs for the whole run has not reached thermal equilibrium. With a 300 W xenon source delivering 50 W of total optical power, that is usually a marginal cooling path rather than a weak lamp.
Verify the spectrum as well: a swapped filter or a substituted window passes a power check and fails a chemistry check three months later.
Repeatability is the most useful number you can take home and the one least often demonstrated. Three runs, same catalyst, same light, same sampling: the spread across them is your real resolution, and no claimed detection limit beats it.
Response time decides whether you can see anything transient. Step the reactor temperature and time how long the readout takes to settle inside its band; inject a known aliquot into the loop and time the analyzer's rise. If logging runs at 1 min⁻¹, a transient shorter than that interval is invisible however good the detector is, so log interval and response time must be chosen together.
Then compare your expected rate with the system's resolution. If the difference you hope to publish is smaller than the spread you measured, the instrument is not your limiting factor.
Plan a full day on site, with the leak test started on the first evening so the pressure rise is logged overnight. The optical and repeatability work need a warm source and several hours of steady operation; compressing it into an afternoon produces numbers that describe a cold instrument.
Your own, under your conditions. The purpose is not to reproduce a published rate but to establish that the system produces a number you trust on the material you will actually study. Ask for the supplier's reference run as a separate data point, never as the acceptance criterion.
Establish whether it is a setup problem or a design problem before rejecting anything. A leak at a fitting is a setup problem and a short fix; a blank that rises under light every time is a design problem. Either way, record the failure and re-test after the fix, because a criterion you did not re-test is not a criterion.
