Ran an annual flow test on a horizontal split case pump outside Macon a few years back. Rated 1000 gpm at 100 psi, and the churn test the week before had looked fine, like it always did. Then we opened the flow lines and it couldn't hold rated pressure much past 750 gpm. A decade of weekly no flow tests, all passed, and not one of them ever came close to catching it. That's the thing nobody explains well about pump testing. The weekly test and the annual test are not the same test done more often. They prove two completely different things.
What the weekly (or monthly) churn test actually proves
Churn just means no flow. Pump starts, runs at shutoff head for a set number of minutes, you read pressure and discharge, you shut it down. NFPA 25 wants this done regularly, usually weekly, though check your AHJ and your pump type, some jurisdictions allow monthly for certain electric driven setups with the right supervisory signals. I'm not going to quote you a universal interval because I've seen it enforced differently county to county.
What churn actually confirms: the pump starts, the driver runs, the controller does what it's supposed to, and the pump holds a reasonable shutoff pressure. That's it. It does not tell you whether the pump can actually deliver rated flow under load. A pump can churn beautifully for years while its impeller wears down, its bearings degrade, or its casing rings erode, and you'd never know from the weekly test alone. Churn catches dead batteries, failed starts, and stuck valves. It does not catch capacity loss.
What the annual flow test is actually checking
Once a year (with the right flow device, a header, sight glass, whatever your setup uses) you run the pump against three points: churn, rated flow (100%), and overload, usually 150% of rated. You plot pressure at each point and compare it against the pump's original acceptance curve, the one from the factory test or the initial commissioning test.
That comparison is the real point of the annual test. A pump curve degrades slowly, and a single year's reading in isolation tells you almost nothing. What tells you something is this year's curve laid against last year's, and against the baseline. If the current curve sags more than about 5% below the acceptance curve at a given flow point, that's generally treated as meaningful degradation and worth investigating, not just noting and moving on. Some techs call this the 95% rule, since you're really asking whether the pump is still producing at least 95% of what it produced when it was new.
Here's an anonymized set from a real annual test, comparing against the acceptance curve on file:
| Flow point | Acceptance (baseline) | This year | % of baseline |
|---|---|---|---|
| Churn (0 gpm) | 118 psi | 114 psi | 96.6% |
| Rated (1000 gpm) | 100 psi | 93 psi | 93.0% |
| 150% overload (1500 gpm) | 65 psi | 58 psi | 89.2% |
Churn held fine, which is exactly why the weekly test never flagged it. Rated and overload both dropped under 95%, and overload dropped hard. That pattern, fine at churn and progressively worse as flow increases, usually points at wear inside the pump itself, worn wear rings or an impeller losing clearance, not an electrical or controller issue. If it were flat across all three points instead, I'd be looking upstream at the supply side first.
What actually fails, in order of how often I see it
Packing. Should drip slowly during operation, roughly a drop or two per second is the old rule of thumb, not a stream and not bone dry. Bone dry packing means it's either mechanical seal (different rules apply) or the gland's been cranked too tight, which will cook the shaft sleeve over time. A steady stream means it's worn out and needs repacking. This is the single most common finding on a pump room visit, and it's also the cheapest one to fix if you catch it early instead of after it scores the shaft.
Relief valve. On pumps that have one, it should stay shut at normal pressures and open smoothly when system pressure exceeds the setpoint, then reseat cleanly. Ones I've seen fail either weep constantly (usually just needs adjustment or the seat's damaged) or, worse, don't open at all and let pressure climb past what the system's rated for during a churn test. Test it. Don't assume it's fine because it was fine three years ago.
Controller battery. This one's sneaky because the weekly churn test often uses the same start signal every time, and a battery with just enough juice left to crank the starter once can pass ten weeks running and still be marginal. What actually exposes a weak battery is the annual test's manual start, sometimes done cold after the pump's been sitting, or a second start attempt right after the first. If your controller logs cranking amps or start voltage, look at the trend over a few tests instead of just pass or fail on the day. A battery that's trending down for six months and then fails on test day isn't a surprise, it's just the first time anyone looked at the numbers together.
The paperwork the AHJ actually wants
Most AHJs want the annual flow test report with the pump curve plotted against the acceptance curve, not just three numbers in a table. They want to see the trend, not a snapshot. Bring the last two or three years if you can, because a single year in isolation reads fine even when the pump's clearly declining. If you found degradation past that 5% mark, expect the AHJ to want a note on what the corrective plan is, not just an FYI. And if the pump's going to be out of service for any of this, that's an impairment, tag it and notify per your impairment procedure before you start pulling it apart.
Related reading if you're deep in the sprinkler side already: the main drain test is the other annual number that tells you about supply health, and if a pump test turns into a multi day repair, the impairment procedure covers what notification actually looks like.
Anyway. Full disclosure, I built FireDeck, and the part of this that actually got old for me wasn't the test itself, it was pulling last year's curve out of a filing cabinet to compare against this year's by hand every single time.