Engineer Guides

Everything on this app hangs off one picture. Refrigerant goes round a loop and is only ever in four states:

• High pressure vapour — out of the compressor, hot. Also known as the discharge line.
• High pressure liquid — out of the condenser, warm. Also known as the liquid line.
• Low pressure liquid — after the expansion valve, cold and starting to boil. Also known as the expansion line.
• Low pressure vapour — out of the evaporator, cool. Also known as the suction line — the big insulated pipe your gauge connects to.

Only two things change the pressure: the compressor pushes it up, the expansion valve lets it back down (on many single splits and fridges that's a fixed capillary tube rather than a valve — same job, no moving parts). The two coils don't change pressure — they just move heat, and the change of state is what carries it.

Superheat and subcooling are both the same idea — how far past the change of state has it gone? — measured as a temperature. Superheat: at the end of the evaporator, how far the pipe has warmed past boiling. Subcooling: at the end of the condenser, how far the liquid has cooled past condensing. That's it. Both live at the marked spots on the map.

Cooling vs heating — why “high” and “low” move. On a reversible split the 4-way valve swaps which coil does which job: cooling, the indoor coil is the evaporator; heating, it becomes the condenser. And on most UK wall splits the expansion valve lives in the outdoor unit — so in cooling both interconnecting pipes run on the low side, and in heating both run on the high side (big pipe hot gas, small pipe liquid). Your single gauge port on the big pipe reads the low side in cooling and the high side in heating — which is exactly why the sheet asks which mode you're in before it reads anything.

Before a cover comes off or a meter goes near a terminal, the circuit has to be proven dead — and the tester proven to still work either side of that. Miss the re-prove and a tester that died mid-test would have shown you a live circuit as dead. This is general best practice; your scheme's or employer's own safe-isolation procedure and method statement is the authority, and the work must be within your competence.

1. Identify the exact circuit to work on and its point of isolation.
2. Choose the right tester — an approved two-pole voltage indicator to GS38. Check the leads, probes and fused tips for damage.
3. Prove the tester on a known live source or a proving unit — it must read live.
4. Isolate — switch off at the point of isolation: breaker off, fuse withdrawn, or main switch off.
5. Lock off — your own lock, and you keep the only key. Fit a caution / warning notice.
6. Check for other supplies — a second feed, a borrowed neutral, standby/UPS, PV or battery. Any of them can make a “dead” circuit live.
7. Test dead at the point of work — every combination: L–N, L–E, N–E (and phase-to-phase on three-phase). All must read zero.
8. Re-prove the tester back on the known live source — proving it didn't fail during the dead test.
9. Safe to work — and the lock stays on until you're finished.

The one that catches people: skipping step 8. Prove, test, prove — the test dead only means something if the tester is proven working after it.

A residual current device does not measure a fault. It measures a difference — what went out on the line against what came back on the neutral. Anything that leaks to earth anywhere on the circuits it protects widens that gap, and the device only sees the total.

The number that makes every reading make sense. An RCD is only guaranteed not to operate below half its rated residual current, and it must operate by the rating itself. So a 30 mA device is designed to hold below 15 mA and to trip by 30 — and anywhere between those two it may legitimately go. That band is not a defect. It is the specification.

Which is why an installation sitting at 15 mA is not “fine until something breaks”. It is already at the edge, and it will trip on whatever switches on next. That thing is the trigger. It is not necessarily the fault.

Some leakage is by design, and it is worth knowing which. Inverter drives pass a small current to earth through the EMC filter that stops them interfering with everything else in the house. A few milliamps that sits steady is the filter doing its job. A figure that climbs as the compressor runs, or that on its own approaches the device rating, is not.

Careful with the two ratings — they are both printed on the device. The residual rating is what trips on leakage, and on a domestic device it is almost always 30 mA. The load current rating is what it can carry — 63 A on a main switch, 32 A on an RCBO — and it has nothing to do with leakage at all. “A 30 amp RCD” is a phrase everyone says and nobody means; get the two mixed up when you are working out headroom and every sum after it is wrong.

Half the rating is the number to work to, and a clamp is all you need to use it. On a 30 mA device that line is 15 mA. Under it you have headroom; at it you are inside the band where that device is allowed to operate, and whatever switches on next will be blamed for it. That is the whole test, and it needs nothing but a leakage clamp and the rating printed on the front of the device.

Where an electrician and an MFT are already on site, a ramp test will tell you where that particular device actually operates, and it is worth asking for while somebody with one is standing there. On the job this guide comes from it let go at 22 mA — so 15 mA was not comfortably under a 30 mA device, it was seven milliamps of headroom. A sharper answer, but a refinement rather than a requirement: working to half the rating told the same story and reached the same conclusion. Worth knowing too that a ramp test drops the supply, so it is not something to spring on a customer with a freezer or a home office on that board.

The cheapest test on this page: clamp it BEFORE you fit anything. A house already sitting at 15 mA on a 30 mA device has no headroom left, and the unit you are about to install will be blamed for the first trip. A minute with the clamp at the tails, written down, turns that future argument into a record — and if they do ring back, you already know what the installation was carrying before you arrived.

The meter has to go low enough, and most do not. This is a leakage clamp — one that resolves to a fraction of a milliamp. A general purpose AC clamp built for measuring load current typically steps in 10 mA at best, and plenty only read to 0.1 A. Put one of those round a pair of tails carrying 15 mA and it will show you 0.00, which looks exactly like a clean installation and is not one. A reading of zero on a meter that cannot resolve the number is not evidence of anything. Check the resolution in the specification before you trust a low reading — some multimeters will do this test properly, many clamps will not.

How to take the reading. Clamp an earth-leakage meter around the line and neutral together — the imbalance between them is the leakage, the same sum the RCD is doing. Clamping the earth conductor alone also works where you can get at it. Take it with the installation as you found it first: that is the background everything else has to add to.

Then take it away and watch. Isolate loads one at a time with the customer's agreement and note what each contributes — a kitchen at 7 mA, a hob at 4, an upstairs socket circuit at 0.3. The individual figures are rarely alarming. The total is what trips the device, and the total is what nobody measures.

The comparison is the evidence, not the reading. One number tells you what was flowing. Two — as found, then with the background reduced — tell you where it was coming from. Record both, because that pair is what turns “the compressor tripped it” into something you can put in a report and defend.

And what it does not settle. Reducing the background and finding the machine runs clean proves it was not the cause on the day. It does not certify the machine. Say what you measured and what changed, and leave it there — the remedy for a loaded installation is dividing circuits or tracking down the leakage, and that is an electrician's job.

This is the most convincing wrong answer in the trade. The compressor starts, the board goes off, and everything about it says compressor. The correlation is real. The conclusion usually is not.

What actually happened on the job this guide comes from. Customer reported nuisance tripping, and it did trip on compressor start, every time. The obvious read was an inverter or compressor earth fault — a strip-down, a recovery and an expensive part. Before any of that, a clamp meter on the installation as found read about 15 mA, on a 30 mA device, with the air conditioning switched off.

The rest was arithmetic. Kitchen about 7 mA, hob about 4, upstairs sockets 0.3. Take those away, run the machine, and the compressor started and ran without tripping anything. The compressor was never the fault. It was the last few milliamps onto a device already sitting in the band where it is allowed to operate.

Separate two questions that feel like one. What triggered the trip, and why was the device already close to operating? The compressor answers the first. It says nothing about the second. Condemn on the first alone and you replace a healthy compressor into an installation that will trip again on the kettle.

“I disconnected the compressor and it stopped tripping.” This feels like proof and it is not, which is worth sitting with because it is the step most engineers reach for first. The compressor is the biggest thing on that circuit, so it is the most likely thing to be last over the line. Take it out and the tripping stops — whether or not it was ever faulty. Removing the trigger looks exactly like removing the fault. The two only separate when you run the unit with the other circuits isolated: a genuinely faulty unit trips on its own, a healthy one on a loaded installation runs perfectly and only trips in company.

And the insulation test had already answered it. On this job the compressor windings were tested to earth and read open circuit on U, V and W, with a little something on one lead — which looked suspicious enough to order a compressor. Ohm says otherwise, and the sum takes ten seconds. A path to earth leaks V/R. To pull 30 mA off a 230 V supply, that path has to be about 7.7 kΩ — seven and a half thousand ohms, not megohms. Even a genuinely poor 1 MΩ leaks 0.23 mA, roughly a hundred and thirty times too little to operate the device. A compressor capable of tripping a 30 mA RCD does not read “a bit on one lead”. It reads like a short.

But do not read a clean insulation test as no leakage. An insulation tester applies DC. It finds resistive breakdown and is blind to capacitive leakage — and the capacitive kind, through the EMC filters in every inverter drive, charger and switched-mode supply in the building, is most of what a modern installation actually carries. That is why the compressor could be resistively sound and the house still sat at 15 mA. The insulation test clears a fault it can see; only a clamp on the running circuit sees the total.

The order that settles it. Measure the background before you suspect anything. If it is at or above half the device rating, reduce it and retest before touching the machine. Only when the background is low and the machine still trips the device is the equipment the suspect — and then measure what the machine itself contributes, isolated, energised, and running.

The exception that proves it is worth asking when. A trip the instant the supply is switched on, before anything runs, is not a cumulative-leakage story — that is a dead short to earth, and it is findable with the unit isolated. A trip only during defrost points at the base-tray heater sitting in standing water. Same symptom, three different faults, and the timing is what separates them.

One more worth checking before anyone condemns anything: whether the device suits the load at all. An inverter drive can produce a DC component that a Type AC device cannot see, and fitting the wrong type produces exactly this complaint. The protection guide covers reading the type off the front.

You are not designing the circuit. You are reading what is already on the wall and working out whether it matches what the manufacturer asked for — because on an inverter machine, the wrong device is not a detail. It is either a nuisance trip you get called back to, or a protective device that cannot see the fault it was fitted for.

The MCB curve is the letter before the number. A B16 and a C16 are both 16 A; what differs is how much inrush they tolerate before tripping instantly. A B curve trips magnetically at roughly 3–5 times its rating, a C at roughly 5–10 times, a D at roughly 10–20. That is the whole reason curve appears in a heat pump manual: a compressor drawing a heavy start current through a B curve can trip a perfectly healthy circuit, which is why several makers specify C. The overload side is the same for all three — the curve only changes the instantaneous end.

Read the device, not the drawing. The rating and the curve letter are printed on the front of the breaker. Note both. “16 A” on its own is half an answer and it is the half that does not explain a nuisance trip.

RCDs are where inverters change the answer, and it is worth understanding why. An RCD compares what goes out against what comes back. The older Type AC can only see a clean alternating imbalance. An inverter drive does not produce one — its rectifier and switching can put pulsating or even smooth DC into a fault current, and smooth DC does not just go unseen by a Type AC, it can saturate the core and blind the device to the AC it could otherwise detect. That is why manufacturers of inverter equipment specify a type above AC, and why BS 7671 requires the type to suit the load rather than the other way round.

The four types, in the order they see more: AC sees sinusoidal AC only. A adds pulsating DC — the practical minimum on modern equipment. F adds mixed frequencies and is aimed squarely at single-phase inverter drives. B adds smooth DC and is what three-phase drives, PV and EV charging generally need. Each one does everything the ones before it do.

The type is a symbol, not a word. Look on the front of the RCD or RCBO next to the rating: a sine wave alone is Type AC; a sine wave with a pulsed half-wave beneath it is Type A; F and B carry further marks for mixed frequency and smooth DC. If all you can find is a sine wave, you are looking at a Type AC — and on a heat pump circuit that is worth raising, whatever else the board looks like.

What our own model data says, and it is not uniform. Grant recommend a Type B or Type F at 30 mA high-speed and are explicit that an RCD is not their requirement in the first place. Vaillant require Type B, short-time delayed and suitable for inverters. Bosch scope it by part of the machine — the 5800i control unit must be a minimum Type A and states that Type AC must not be used, and that instruction is theirs about the control unit specifically, not the heater. NIBE publish 30 mA and a separate device, and no type letter at all. Check the model reference for the unit in front of you rather than assuming a house rule.

And the line we do not cross. Reading a device, comparing it to the manual and reporting a mismatch is our job. Deciding what the circuit should be — sizing, discrimination, earthing arrangement, Zs — belongs to a competent electrician working to BS 7671. If what is fitted does not match what the maker specifies, say so plainly, put it in the report, and let the electrician design.

Every cable has its specification printed along the sheath. It tells you what the cable is rated for, what it is made of and what it was built to do — and none of it tells you whether it is big enough, which is a separate question with a separate answer.

The harmonised code, read left to right. Take H07RN-F. H means it is made to the harmonised European standard. 07 is the voltage rating: 07 is 450/750 V, 05 is 300/500 V, 03 is 300/300 V. R is the insulation — R for rubber, V for PVC, S for silicone. N is the sheath, here polychloroprene, the tough oil- and weather-resistant outer you want on a flexible run outdoors. The -F at the end is the conductor: fine stranded, so it is flexible.

So H05VV-F is a 300/500 V PVC-insulated, PVC-sheathed flexible — ordinary appliance flex. H07RN-F is the rubber-sheathed 450/750 V version that survives being outside. That difference is why Vaillant name a minimum of 60245 IEC 57 / H05RN-F for their connecting cable rather than leaving it to taste — a PVC flex that hardens and cracks in two winters is not the same product.

The conductor letter matters more than it looks. -F is fine stranded and flexible. -R is stranded but rigid, -U is solid, both for fixed wiring. Put a solid-core cable where something moves or vibrates and it work-hardens and breaks, usually inside a terminal where nobody can see it.

The UK numbers you will meet on the same jobs. 6242Y is flat twin and earth — the grey stuff in the walls, fixed wiring only. 6243Y is the three-core-and-earth version. 6491X is a single insulated conductor for conduit and trunking. SWA is steel wire armoured, which is what usually takes the supply out to an outdoor unit — and its armour is a mechanical protection that may or may not be doing an earthing job, which is the electrician's call, not ours.

Where the size is printed, and what it does not mean. The cross-section is on the sheath in mm² — 1.5, 2.5, 4, 6, 10. It is the conductor area, not a current rating. What a cable can actually carry depends on how it is installed, whether it is bunched with others, the ambient temperature, the length and volt drop, and the disconnection time required. That is why the same 4 mm² reads very differently clipped to a wall than buried in insulation, and why cable sizing lives in BS 7671 Appendix 4 rather than on the sheath.

So what do we do with it? Record what is there — type, size, condition, how it is run — and compare it against what the manufacturer specifies. Grant put it plainly in their own manual: the power cable should be sized by a qualified electrician in accordance with current wiring regulations. Our job is to read the marking, write it down, and flag a mismatch. Sizing the cable is not our decision to make, and a report that says exactly what was found is worth more than one that guesses.

While there is still liquid, the temperature cannot rise. Once the last drop has boiled, it can. How far it climbs is the superheat.

Refrigerant boils its way through the evaporator. Superheat is how far the pipe temperature has climbed past the boiling (evaporating) temperature by the end of the coil — it tells you where the last of the liquid finished boiling.

High superheat = the liquid ran out early. That's either a coil being starved (charge, restriction) or a machine turned right down feeding a big coil — two different problems, same number. Zero / negative = liquid still boiling at the pipe — flooded suction, and a compressor at risk.

Total superheat needs suction pressure + a probe on the suction pipe (the outdoor unit's service valve is the no-covers-off spot). Evaporator superheat — the unit's own coil sensor (T2 and friends) against saturation — is the partial version you can read off the remote without tools. It points the same direction; it just can't be trusted as the final number.

The metering device sits between the liquid line and the evaporator. Its job is the same whatever the type: drop the high-pressure liquid to the low side and feed the coil at the right rate. But how it does that — whether it actively controls superheat or not — changes what your readings mean. Three types, and the first one is the one people forget.

Capillary tube — a fixed restriction: a long, thin, often-coiled length of small-bore copper, no moving parts, no sensing, no control. It's on most single splits, fridges, dehumidifiers and smaller kit, because it's cheap and reliable. The catch: being fixed, it does not hold a target superheat— superheat floats with the charge, the load and the ambient. So you cannot judge a cap- tube system by “is superheat in the 4–8 K window?” the way you can a TXV. It is also charge-critical: no receiver, no valve to buffer, so the charge has to be weighed in exactly to nameplate. On the job: there is nothing on the metering to test — no coil, no bulb. A blocked cap tube or filter drier starves the coil (high superheat, low suction) and reads just like undercharge, so you weigh the charge and look for the restriction — you never “check the EEV” on one of these, because there isn't one.

TXV (thermostatic expansion valve) — mechanical. A sensing bulb clamped to the suction line, plus a diaphragm and spring, open and close the valve to hold a fairly fixed superheat (typically ~4–8 K). That's its whole purpose: on a healthy TXV system, superheat should sit in a tight range. You can't read its position — you infer what it's doing from the superheat.

EEV (electronic expansion valve) — a stepper-motor valve driven by the PCB from sensor inputs. Far finer control, a wider operating range, and the reason inverters can modulate — so it often runs lower and varying superheat by design (2–5 K). Crucially, its position is a number you can read on the service display (LEV pulse / EEV angle / "LA" on the Midea remote) — a free, no-covers reading of exactly what the metering is doing.

Why it matters on the job: don't condemn low superheat on an EEV/inverter system as flooding the way you might on a TXV — it can be normal control. On an EEV, read the valve position first: pegged fully open with high superheat points at a starved coil or a driver/valve fault; driven right down at minimum modulation confirms a low-load / oversized story (see the oversizing guide). On a TXV, a hunting or wrong superheat with the bulb loose, poorly insulated, or its charge lost is the classic mechanical failure. Identify which you've got before you trust the number: a TXVhas a bulb on the suction line (often with an equaliser tube); an EEV has a coil and wires back to the board; a capillary tube has neither — just a thin coiled tube off the liquid line — so a floating superheat there is expected, and the question is the charge, not the metering.

Both starve the evaporator. Both give you low suction and high superheat. On those numbers alone they're identical — and one of them gets "fixed" by adding gas the system doesn't need.

The discriminator is subcooling. An undercharged system has nothing spare in the condenser: subcooling collapses. A restricted system holds refrigerant back in the condenser: subcooling holds or climbs. Same starving evaporator, opposite liquid lines.

Low subcooling → shortage. Subcooling that holds — or climbs — while the coil starves → blockage (drier, TXV, kinked liquid line). Add gas to a restriction and you've made an overcharged blockage.

Suction creeping up, head falling away, capacity gone. The gauge picture for worn compressor valves and for a 4-way valve (US texts say "reversing valve") leaking hot gas across itself is the same — and one of these mistakes costs a compressor that was never faulty.

The discriminator isn't on the gauges. It's a temperature clamp — both clamps at the outdoor unit, either side of the 4-way valve: the suction line coming in from the coil, and the stub into the compressor. Those two points sit centimetres apart, so pipe run and ambient can't muddy the reading; more than about 3 K of rise between them means hot gas is crossing the valve. Prove it before anyone orders a compressor.

High superheat, low suction, poor capacity — every engineer reads that as low gas, and often it is. But the gauges are not measuring how much refrigerant is in the system. They are measuring pressure and temperature, and several different faults push those the same way: a restriction starving the coil, an oversized unit throttled back, low load across the evaporator, a sensor reading something the machine is not actually doing.

There is exactly one measurement that settles it, and it is not on your gauges. Recover the charge and weigh it against the nameplate. Mass is the only direct evidence of how much refrigerant was in there. Everything else is inference — good inference, worth acting on, but inference.

Why it matters beyond being right: if you top up a system that was never short, you have overcharged it and hidden the real fault behind a second one. And if it genuinely was low, the gas went somewhere — so it is a leak until proven otherwise. Find and fix the leak before anything goes back in, then weigh the charge in to nameplate rather than trimming to a gauge reading.

None of this makes the diagnostic wrong. It makes it honest: the readings tell you where to look and how hard to look, and the scales tell you what was true.

The compressor is the dearest part on the machine and the easiest one to blame. Faults escalate, symptoms stack up, and it starts to feel like everything points at the compressor. Before that call gets made, two checks are worth more than the rest put together — and both of them usually point somewhere else.

1. Know which test your meter just did

OL means opposite things on the two tests engineers run to earth, and getting them the wrong way round either condemns a good compressor or misses a dangerous fault.

Insulation resistance — leads from a compressor terminal pin to earth or bare copper, instrument applying its test voltage. You are asking: is this winding leaking to the shell? A healthy compressor is a near-perfect insulator, so the display reads OL, ∞, or tens of megohms — and that is a PASS. The failure here is a low number. Below about a megohm is breakdown, whatever a maker’s flowchart tolerates; never clear a compressor sitting at a few hundred ohms to earth.

Earth continuity — leads from the earth pin at the plug or isolator to exposed metal, low-ohms range. You are asking: is the earth path intact? You want near zero ohms, and here OL is a genuine FAIL — there is no earth path, and that is a safety fault to fix before the machine runs again.

Same two letters, opposite verdicts. Work out which test you did by where the leads were and what the instrument was doing, then read the result against that.

2. Ask what it does in the other mode

This one is nearly free and it is the strongest evidence you will get. A compressor that has failed electrically does not run at all. A shorted winding, an open winding, insulation broken down to the shell — none of those let it start and pull down in one mode and not the other. So if the machine cools properly but shuts down in heating, the motor and its windings have just proved themselves.

What that pattern is actually pointing at is whatever changes when the mode changes. The 4-way valve has to shift, and can stick or leak across itself. The outdoor coil becomes the evaporator, so it frosts and depends on a defrost cycle that has to start and has to clear. The indoor coil becomes the condenser, so indoor airflow — a blocked filter, a dirty blower — now limits head pressure, and it will trip on high pressure in heating while cooling perfectly happily. Add colder ambient and higher condensing temperatures and you have several protections that only ever see their limit in heating.

Still prove the compressor electrically, safely isolated — it costs ten minutes and settles the argument. But read the mode asymmetry for what it is: the machine telling you the compressor is fine and the fault lives in what the reversal brings with it.

An inverter unit serving a room that's already satisfied winds the compressor and outdoor fan down to their minimum. A trickle of refrigerant now feeds a full-size coil with the indoor fan still shifting maximum air — so it boils off in the first passes, the coil sits near room temperature, and superheat reads high. Exactly the picture a low charge paints — but nothing is broken, and no fault code logs, because the machine is doing what it's told.

When it can't go low enough it overshoots and cuts out, over and over — short cycling. The customer's version: "after a couple of hours it's just blowing air."

The tell is off the gauges: compressor frequency at minimum, fans held low, no fault stored, and a unit big against the room's load. Field-proven check: dropping the indoor fan (where the maker allows it) pulls the coil temperature down and the compressor speeds back up — the machine re-matched to the load. That's a diagnostic observation, not the repair; the answer to an oversized system is sizing and control.

Some of the hardest call-outs aren't faults at all — the machine is doing exactly what its installation forces it to do. Three that catch everyone:

1. The boxed-in condenser. Fences, decks, tight corners: the outdoor unit swallows its own hot discharge air, entering-air temperature creeps up, head pressure climbs, capacity falls — reads like an overcharge or a dying condenser. The test costs nothing: measure the air temperature actually entering the coil against true ambient in the shade. Entering air warmer than ambient = recirculation. The fix is clearance, not gas.

2. The long pipe run nobody trim-charged. Every unit ships pre-charged for a stated base pipe length — the nameplate gives the base and the grams per extra metre. Skip the trim charge on a long run and the system is genuinely undercharged from day one: low suction, high superheat, collapsed subcooling. Before condemning it as a leak, compare the actual pipe run against the base length — the Charge Calculator holds the official figures per model.

3. Crossed pipes on a multi-split. Room A calls: its fan runs but nothing cools. Meanwhile room B — switched off — has an ice-cold coil. Turn both on and everything "works", which is why it hides for months. The refrigerant lines (or interconnect wiring) were swapped between ports at install. Touch test: with one unit calling, feel the coil of the units that are OFF. A freezing coil on a unit that isn't running is the giveaway.

A system that's off and equalised should sit at the saturation pressure for the ambient temperature — nothing more, nothing less. Ten seconds against a P-T chart before you've cracked a port.

Standing pressure well above saturation for the ambient means something in there isn't refrigerant — classically nitrogen never pulled out after pressure testing, sitting in the condenser strangling it. A low-pressure fault code with a high standing pressure is not a gas shortage, whatever the label says.

One glance, one honest number — it's why the diagnostic shows the expected standing pressure right next to the ambient slider. Never diagnose from standing pressure alone; use it to catch the story that doesn't add up.

R290 is propane — a natural refrigerant with a tiny GWP (about 3, against R32's 675), which is why the F-Gas phase-down is putting it into more and more heat pumps, and increasingly splits. The refrigeration cycle is the same as any other; what changes is the safety, and it changes a lot.

The one thing to hold onto: R290 is A3 — highly flammable, not A2L “mildly flammable” like R32. A leak that finds an ignition source is a fire or an explosion, not a smoulder. Two facts shape that hazard on site:

• It's denser than air — a leak sinks and pools at ground level, which is exactly where you're kneeling at an outdoor monobloc.
• It's odourless — no stink additive, so you won't smell it. A leak collects invisibly and silently around your feet.

The protective zone. The manufacturer defines a keep-clear zone around the unit, and inside it there must be no ignition sources: plug sockets, light switches, lamps and other switches, aerosols and sprays, other combustible gases, naked flames — and no smoking. Your own kit counts too: a phone charger, a torch that switches at the head, a heat gun.

Leak-check only with A3-rated equipment. A standard sniffer isn't necessarily rated for a flammable refrigerant — use a combustible-gas / A3-rated detector, and search low down, where the gas gathers.

The charge is small and tightly limited — monoblocs typically carry roughly 0.6–1.3 kg depending on the chassis, because the flammability rules cap how much can be used. Most R290 machines are factory-sealed monoblocs you rarely open at all: the refrigerant work has largely been designed out. If you do open the circuit, recover into a cylinder rated for flammable gas, ventilate, keep every ignition source out of the zone, and only proceed if you're trained in the specific risks of R290.

This is the shape of the hazard, not a substitute for the manufacturer's own instructions. The protective-zone size, the charge, and the handling steps for the specific unit are set by its documentation — that always governs.

A condensate pump has two jobs to wire: the mains that runs the pump, and the alarm / overflow pair that protects the ceiling if the water can't get away. The second one is the one people leave out — and it's the one that matters. Isolate first (see the safe-isolation guide) and follow the pump manufacturer's own diagram — terminal names differ between makes.

How it goes together:

1. Sensor in the tray. The float/sensor sits in the indoor unit's drain so it catches water before the tray does. It tells the pump when to run.
2. Discharge — lift, then fall. Small-bore up to above the pump, then a continuous fall to the drain. Keep the lift within the pump's rated head, and fit the loop / non-return so water can't run back and re-trigger it.
3. Mains — L, N, E. From a fused spur or the indoor unit's supply as the maker specifies. Earth it.
4. The safety pair — this is the important one. The pump's alarm/overflow contact (a volt-free pair, usually normally closed) wires back into the indoor unit so that on high water it cuts the cooling demand — the unit stops making condensate before the tray overflows. Land it on the terminal the AC manufacturer names for an external interlock / float switch.

Never leave the safety pair out or bridge it “to make the fault go away.” A pump can fail; the interlock is what stops a failed pump becoming a flooded ceiling. If the unit throws a drainage/float code, that's this circuit — test it, don't defeat it.

Most residential splits only give you a low-side service port — and half the advice out there quietly assumes you have both. "Check your subcooling" — with what port, exactly?

Here's what the low side plus your eyes actually gives you:

• Suction pressure + a clamp on the suction line = real superheat
• The unit's own service display = coil temps, compressor frequency, current
• Air on / air off = what the customer's actually getting
• Compressor speed = struggling, or just throttled back

That's enough to split a starved coil from an oversized system from a failing compressor — without ever touching a high-side port the unit doesn't have. The diagnostic's guidance runs in exactly that order.

Most "faults" in an install's first year were built in on day one. Commissioning is how you not only avoid them — it's how you prove you did. This is the general sequence; your unit's own installation manual is the authority on figures and steps.

1. Pressure test with nitrogen — never refrigerant, never air. Get it up high enough to find small leaks (Daikin recommends up to 30 bar — but never above the max working pressure on the nameplate). Proper leak-test solution on every joint, not washing-up liquid: soap can crack flare nuts, hold salt that freezes in the pipe, and its ammonia corrodes the brass-to-copper joint.
2. Release the nitrogen, then pull a deep vacuum. The vacuum isn't just removing air — it's boiling off moisture. Don't open the stop valves until the vacuum is finished. Hold it and watch: creeping back up means moisture still boiling off, or a leak the pressure test missed.
3. Trim charge by pipe length — weighed, not guessed. The factory charge covers a base length (often 5–7 m). Beyond that it's grams per metre from the data plate or manual. The Charge Calculator has the base lengths and rates per model.
4. Record the charge. What's in the machine after your addition, on the label and in your F-Gas records — the Refrigerant Register here does this.
5. Prove it works, in numbers. A proper run up to temperature, then air on / air off across the indoor unit, and superheat if the ports allow. Write the numbers down — they're the machine's birth certificate. The next engineer (maybe you) compares against them instead of guessing.

The classic shortcuts that come back: purging with refrigerant instead of nitrogen · washing-up liquid on flares · valves opened before the vacuum finished · a long run never trim-charged (reads as "undercharge" two summers later) · a 30-second test · nothing written down.

On a monobloc the refrigerant circuit arrived sealed and tested. What decides whether this install performs is the water side and the controls — and that's where first-winter "faults" are usually born. General sequence; the manufacturer's own commissioning sheet (and MCS paperwork where it applies) is the authority.

1. Before power: wiring, earth and fuse sizes checked against the manual, isolation valves open, the auto air-vent cap cracked open, supply voltage matching the nameplate, and a look inside for transit damage.
2. Clean, treated water. Flush the system, then dose inhibitor — and record the brand and product, because the commissioning form asks and the warranty cares. Antifreeze protection where the design needs it.
3. Vent it, then prove the flow. Air and low flow are behind half the early nuisance faults. Record the flow rate in l/min at commissioning — it's the number every future visit gets compared against.
4. Warm up gently on underfloor. A cold screed heated too fast can crack — bring the water temperature up gradually on first heat.
5. Balance the emitters. With flow temperature near its set point, check the drop across each emitter starting furthest from the pump, and trim lockshields a little at a time — against the design ΔT in your manual (Grant's Aerona guidance is about 4 K; many designs sit at 5 K).
6. Set the controls to the design, not the defaults. Weather compensation at the design flow temperature, hot water schedule set, legionella protection confirmed — then show the customer how it works.
7. Record everything. Flow rate, temperatures, settings, inhibitor — the completed commissioning checklist protects the guarantee and is the baseline for every service after it.

The gaps that surface later: no flush or inhibitor · balancing skipped (one room roasting, one cold) · weather compensation left on the factory curve · flow rate never recorded · a screed floor blasted on day one.

When a unit cools, the coil inside gets wet — that's normal, it's pulling water out of the air. Switch the unit off and you've got a wet, dark, warm box with a layer of dust on everything. That's all mould needs. It isn't a fault and it isn't bad gas — it's a wet coil left to sit.

How you know it's started: a musty smell when the unit first kicks in, black or green spots on the louvres, or water sitting in or under the unit. Don't bother with air freshener or wiping the grille — the mould is on the coil, not the plastic.

Stopping it — do these in order, the first one matters most:

1. Dry the coil after cooling. Most units have a setting that runs the fan on after cooling stops, exactly for this — Gree calls it X-Fan, others call it self-clean or mould-proof mode. Find it, turn it on, leave it on. No setting? Run fan-only for 20 minutes after cooling.
2. Keep the filters clean. The dust on the filter is what mould lives on. Little and often through the summer beats one big clean.
3. Make sure the water gets away. Check the drain tray and pipe run clear. Water standing in the tray is where it starts.
4. Use dry mode in muggy weather. A unit that blasts the room cold and shuts off leaves the air damp — the room feels cool but the moisture is still there, and it lands on the cold coil.
5. Don't box the unit in. Curtains, pelmets or a wardrobe tight to the unit stop the air moving, and still air is where spores settle.

One warning: never put household bleach near the coil — it eats the aluminium fins. A coil that's already mouldy needs a proper coil clean with the right product, and that's a job for an engineer.

The scales are the whole point of recovery. What comes out, weighed, is the only direct evidence of what was in there — so set up to measure it, not just to empty the system.

1. Bottle on the scales first. Recovery cylinder on the scales, tared or the starting weight written down. If you tare after connecting, the hoses lie to you.
2. Check the bottle has room. A recovery cylinder is never filled past 80% — liquid needs space to expand. If the nameplate charge is bigger than the room left in the bottle, you need another bottle before you start, not halfway through.
3. Short hoses, purged. Every hose holds gas. Keep runs short, purge the air out, and the number on the scales stays honest.
4. Liquid first, then vapour. Pulling liquid moves most of the charge in a fraction of the time. Finish on vapour and let the machine pull the system right down until it cuts out.
5. Wait, and watch the pressure. When the machine stops, give it a few minutes. Refrigerant dissolved in the oil keeps boiling off and the pressure creeps back up — run again until it stays down.
6. Write the weight down at the machine. Recovered kg goes on the job before you pack up — the register fills itself from it. Compare against the nameplate plus any trim charge for the pipe run: that difference is your leak, in grams.

Short of nameplate means the gas went somewhere. It is a leak until proven otherwise — which is the next guide.

One job, five stages, in an order that never changes. Skip a stage and the callback books itself.

1. Recover and weigh. Everything out, onto the scales (the guide above). The weight tells you how bad the leak was, and the recovered gas is evidence, not waste.
2. Make the repair. Braze with nitrogen running through the pipe (see the brazing guide) — a repair that fills the circuit with oxide scale has traded a leak for a blockage.
3. Prove it holds — with nitrogen, not refrigerant. Strength test, then drop to the tightness pressure and leave it. Write down the pressure AND the ambient at the start and the end: pressure follows temperature, and the app splits the weather from a real leak so a cold morning doesn't condemn your own braze.
4. Vacuum — and watch what it does after you stop. Pull down to 500 microns (0.67 mBar, 0.5 torr), isolate the pump, and watch. Creeping up and levelling off is moisture still boiling — keep pulling. Climbing steadily and not stopping is a leak — back to stage 3. The standing vacuum is the test; the pump running proves nothing.
5. Recharge by weight. Nameplate plus the trim for the pipe run (the charge calculator knows your model), weighed in on the scales. Not topped up to a gauge reading — weighed. Then run it and let the numbers prove the job: superheat and subcooling where the model expects them is your evidence the repair worked.

Log the recovered and added weights on the job as you go — the F-Gas register, the CO₂e figure and the paperwork all write themselves from those two numbers.

Heat copper with air inside it and the inside wall grows black oxide scale. That scale doesn't stay put: the refrigerant washes it round the circuit until it finds the narrowest point — the drier, the expansion valve, the capillary — and blocks it. You braze a leak today and build a restriction for next month.

The fix costs pennies: a whisper of nitrogen flowing through the pipe while you braze. Not pressure — flow. Crack the regulator until you can just feel it at the open end, keep it running the whole time the pipe is hot, and the inside stays as clean as the day it was drawn.

The check is the colour: cut a brazed joint that had nitrogen through it and it's bright inside. No nitrogen, and it's black. Every manufacturer's installation manual asks for this, and every drier cut open after a “mystery restriction” tells the same story.

Pump-down uses the system's own compressor to push the whole charge into the outdoor unit — condenser and receiver — so you can open the low side, swap an indoor unit or repair pipework without recovering a gram. Quicker than recovery, no bottle, no scales. But it only works when the machine can actually do it.

When you can: the outdoor unit has service valves that isolate it from the pipework, the high side has room to hold the full charge, and the compressor runs. When you can't: the compressor is dead, the leak is on the high side (you'd be storing the gas next to the hole), or the unit is being scrapped — a scrap unit gives its gas to a bottle, never keeps it.

1. Gauge on the suction side, unit running in cooling — many inverter units have a forced-cooling mode for exactly this; the manual names it.
2. Front-seat the liquid line valve. The compressor keeps running, feeding gas into the condenser with nothing coming back out.
3. Watch the suction pressure fall. As it approaches zero, front-seat the suction valve and stop the unit. Just above zero, not below — pull the compressor into vacuum and it can arc internally, and any weep on the low side now sucks air and moisture into your clean charge.
4. Watch it for a minute. Suction pressure climbing back up means the liquid valve isn't holding — the charge is leaking back. If it holds, the low side is yours to open.

Going back: reverse it — open the valves, purge the disturbed joints, and the same charge goes back to work. Nothing recovered, nothing added, nothing to log but the leak check on the joints you opened.

Two places the charge can live while you work: the machine's own high side (pump-down) or a recovery bottle. The choice isn't preference — it's three questions.

1. Can the machine hold it? Pump-down parks the charge in the condenser and receiver. On a split with a modest charge, fine. On a big system, the receiver has to be sized to take the FULL charge with expansion room — if you can't show that from the data, it goes in a bottle. A receiver filled solid with liquid has no give at all: warm it a few degrees and something lets go.

2. Where's the work? Pump-down only frees the LOW side. High side repair, compressor swap, drier change on the liquid line — the charge would be sitting in the part you're cutting. Bottle.

3. Is the gas staying? Suspected contamination (burnout, moisture, air) or a unit being scrapped — bottle, always. Pump-down keeps the gas; recovery takes it away for judging or disposal.

Bottle rules, non-negotiable: a proper recovery cylinder — never a virgin bottle, the valve and rating are different. Never past 80% full: liquid needs expansion room. Never mix refrigerants in one bottle — a mixed bottle can't be reclaimed, only destroyed. Label it with what's in it, and weigh what went in — that number is your register entry and your evidence.

The rule comes first: if the gas got out, nothing goes back in until you know where. Regas a leaking system and you've sold the customer the same failure twice — and on a big system, that's a serious quantity of refrigerant with somewhere to go.

Start cheap: if there's still refrigerant in the system, an electronic sniffer round every joint, valve core, flare and braze — slowly, from the top down (most refrigerants are heavier than air, so a leak drifts downward). Oil stains are a signed confession: refrigerant leaks carry a trace of oil out with them. Bubble spray confirms the exact spot the sniffer suspects.

When the system is empty — or the sniffer finds nothing: tracer gas. Pressurise with a nitrogen–hydrogen trace mix (the 5% hydrogen “forming gas” sold for exactly this) and go round with a hydrogen detector. Hydrogen is the smallest molecule there is — it gets out of holes nothing else will, and the detector picks it up in tiny concentrations. Plain OFN alone holds pressure but can't be sniffed; the trace mix does both jobs.

On a big system, divide and conquer. Don't hunt a whole plant in one go — valve it into sections, pressurise one section at a time, and watch which one loses pressure. Write the start pressure and ambient down for each section: pressure follows temperature, and the app splits the weather from the leak so you're not chasing a cold evening round a plant room. Every joint in the guilty section gets the detector treatment until the leak has a name.

Found it? Fix it, then prove the fix: the full repair cycle — strength test, tightness against ambient, deep vacuum, standing vacuum — before a gram of refrigerant goes back. The repair-cycle guide walks it stage by stage.

A scrapped unit still full of gas is a leak with a delivery date — whoever cuts it up releases the lot. Decommissioning is recovery with paperwork, and the paperwork is what proves the gas didn't just vanish.

1. Recover everything to a bottle. Never pump-down — the gas can't stay in a machine that's leaving. Liquid first, vapour to finish, wait for the pressure to stay down (the oil keeps giving gas back), and run again until it does.
2. Weigh it and log it. Recovered kg on the job — the register and the recovery note write themselves from that number. What you recovered versus the nameplate is the unit's last diagnosis: short means it was leaking to the end.
3. The oil counts too. Compressor oil holds dissolved refrigerant and goes as hazardous waste, not down a drain — it travels with the unit or to the waste oil route, named on the paperwork.
4. Cap and label. Cap the open pipe ends, and mark the unit as degassed with the date — the scrap yard's question answered before it's asked.
5. The gas goes back up the chain. Recovered refrigerant goes to your wholesaler or reclaimer, not onto a shelf forever. Clean single-gas bottles can be reclaimed; mixed or contaminated bottles are destruction only — one more reason never to mix.

The app prints the recovery note from the job record — quantities, dates, the unit's identity — so the trail from wall to scrap yard is one document, not a memory.

The hardest part of a dead unit isn't the diagnosis — it's the conversation. You're standing in someone's kitchen telling them the thing they hoped was a £100 fix is actually a replacement. Here's the shape of that conversation, in plain words.

The classic case: an older single split, tripping the electrics, and when you get the gauges on there's nothing in it. That's two separate verdicts at once — the gas has escaped (a leak you'd have to find and fix), and something electrical is breaking down, usually the compressor going to earth after running hot on no gas. Either fault alone might be worth chasing on a newer machine. Both together, on an old one, almost never are.

Say what you found, not what they did wrong:

“I've found two problems. The gas has leaked out, and the compressor is breaking down electrically — that's what's tripping your electrics. I can fix either one, but honestly, fixing both on a unit this age costs most of what a new one does — and you'd still have an old unit.”

Then give them the comparison, not a lecture:

“The repair is a new compressor, finding and fixing the leak, a full test and a regas — two visits, and no guarantee the next part isn't far behind. The same money is most of the way to a new unit with a warranty on the whole thing and cheaper running. I'd replace it — but it's your money and your call.”

Why this works: you named both faults (so it doesn't sound like an upsell), you offered the repair (so they know you're not dodging work), and you gave a recommendation with the reason attached (so they can decide like an adult). Most customers say “what would you do?” — and you've already answered it.

If they want the repair anyway: price it honestly, warn about the next part in writing, and put “recommend replacement” on the report. If they want the new unit: quote it on the spot from the job — the sizing, the quote and the deposit are three taps in the app while you're still at the kitchen table.

Every kilogram of gas you put in or take out has to be written down, and the record has to survive five years. Not five years if you remember — five years, produced on request. Most engineers know the number and far fewer know what the record actually has to say, which is where audits go wrong.

Two people need it, not one. The operator — whoever controls the equipment, usually your customer — has to hold the record. You, or the firm you work for, have to keep your own copy of the work you did and give theirs to them. An engineer who logs it beautifully in their own system and never sends it on has done half the job.

What has to be on it, per system:

• Which refrigerant, and how much is in the system
• How much you added — and whether it was new, reclaimed or recycled
• How much you recovered
• Who did the work: the company and the engineer, with certificate numbers
• The dates and the results of leak checks — a pass is a record too
• If it was scrapped: what you recovered and where it went

The detail people leave off is the source of the gas and the certificate numbers. A quantity with no named engineer behind it proves the gas moved but not that anyone qualified moved it.

The leak-check clock runs off CO₂e, not kilograms. Charge × GWP ÷ 1000 gives tonnes of CO₂ equivalent, and the thresholds sit at 5 t, 50 t and 500 t — 12-monthly, 6-monthly, 3-monthly. Fixed automatic leak detection doubles those intervals and is a requirement in its own right at the top end. Hermetically sealed equipment under 10 t is exempt if it is labelled as hermetically sealed — the label is the exemption, not your opinion of the unit.

And the one that catches people out: a repaired leak has to be re-checked within a month. That check is its own record.

What an audit actually feels like. Nobody asks to see your filing system. They pick a system, or a date, and ask what happened to it — and expect a record that names the gas, the amount, the engineer and the certificate, with the leak checks either side of it. If you can produce that in a minute you are fine. The firms that struggle are the ones holding a pile of job sheets that were never added up.

Where LatterHQ sits in this. Log the gas on the job — on the job sheet, in the commissioning steps, or at the wrap-up — and the register writes itself, tCO₂e included, with your certificate on it. The share button produces a usage report: totals per gas first, the individual movements underneath. That is the thing to send when someone asks what you have used.

Thresholds and duties here follow the GB F-Gas Regulation. Northern Ireland follows the EU regime, and the rules do get amended — if you are near a threshold or an edge case, check the current regulation or ask your certification body rather than taking a guide's word for it.