Why datalogging matters
An ECU tune is a hypothesis. The datalog is the experiment that tells you whether the hypothesis is right. Without logs you have dyno graphs (which only tell you what happened on the dyno) and vibes (which tell you nothing). With logs, every change you make has a before-and-after you can actually point at.
The mistake most beginners make is logging too few channels at too low a sample rate, and then trying to draw conclusions from a single pull. Three of those problems are easy to fix. The fourth — knowing what to look at — is what separates someone who can read a log from someone who has one.
What to log
The minimum useful channel set on any modern ECU:
- RPM
- MAP / boost (kPa absolute or PSI gauge)
- MAF (g/s) if your ECU uses one
- Throttle position
- IAT (intake air temperature)
- ECT (engine coolant temperature)
- AFR / Lambda — boththe wideband sensor reading and the ECU's commanded target
- Knock — raw count, retard, or windowed integral (varies by platform)
- Ignition timing — actual commanded value, not just base table
- Injector duty cycle
On boosted cars, also log wastegate duty cycle and fuel rail pressure. On direct injection, fuel pressure under load is non-negotiable.
Sample rate
Aim for 30 Hz minimum for cruise and part-throttle work, and 50–100 Hzduring WOT pulls and any time you're hunting for knock. Anything under 10 Hz will miss transient events that matter — a 50 ms knock event is invisible at 5 Hz.
Higher sample rates fill disk fast. Capture in bursts (one pull, one cruise segment, one idle) rather than continuous logging.
What to capture
At minimum, in a single session:
- Three WOT pulls in 3rd or 4th gear (load, not gear, is what matters)
- Steady-state cruise at 40–60% throttle, multiple RPM bands
- Idle for 60+ seconds (look for AFR stability and IAC behaviour)
- One deceleration to check fuel cut and re-enrichment
One pull is data. Three pulls is evidence. If two are clean and one shows knock, you probably have a heat-soak problem rather than a tune problem.
What to look at first
- Knock activity during WOT. Any retard? Are windowed counts climbing? On most platforms, small isolated counts are noise; persistent climbs are real.
- Commanded vs actual AFR. Are you hitting your target across the load range? Drift more than ~0.5 AFR points usually points to a fuelling table or pressure problem, not a sensor issue.
- Boost vs target. Overshoot on tip-in? Falloff at redline? Both are common and both are fixable in the wastegate map.
The most common datalogging mistakes
- Sample rate too low to catch knock or transient AFR
- Logging only commanded values, not actual sensor readings
- Drawing conclusions from a single pull
- Logging on the road without a passenger to operate the laptop
- Trusting the ECU's “knock” channel without understanding what it actually counts on your platform