Exhaust Lambda and mass flow Lambda disagree

Going through some old dyno data from a tuning session a customer did and I see a disagreement that I can't understand.

Calculating the air mass flow from ambient temp, pressure, and inlet restrictor choke pressure. Calculating fuel mass flow from fuel pressure delta, injector dead time (as a function of battery voltage and fuel pressure delta), and engine rpm.

Once on power and the turbo gets up to speed we get good signal on the air mass flow and can start comparing the wide band lambda value with the air/fuel mass flow lambda value.

Correlated astonishingly well.

There's a small blip at 5500rpm, when the exhaust lambda reads leaner than the mass flow calculated lambda. Maybe poor combustion, retarded ignition timing, uneven afr per cylinder etc. Causing some of the injected fuel to go past the exhaust lambda as droplets and not being measured, resulting in a lean reading. Makes sense.

But above about 7000rpm the exhaust lambda starts reading RICHER than the calculated mass flow lambda ratio. How is this even possible? There would need to be extra fuel being combusted from somewhere, or would need to be loosing air mass. Engine has no blow off valve. And I would find it very unlikely that there would be a air leak at the lower boost levels at higher rpm, but not at the higher boosts at lower rpm.

Could the lambda sensor be overheating and reading rich as a result?

Has anyone seem something similar? Or have ideas as to what could cause this?

Lamba sensors are pressure sensitive -- is your sensor located upstream (where there could be additional back pressure as the RPM rises), or downstream of the turbocharger?

Will you lose air mass as the air is heated by the turbo charger, or does the pressure build-up account for it all? Some of that mass may get converted to energy and so you need to compensate with the fuel mass you supply, or you will get an indicated rich mixture.

What does the manifold air temp look like above 7000 RPM?

Any chance there is another trim changing the amount of fuel being delivered?

Lambda sensor is post turbo.

I'm not sure you can convert mass to energy without nuclear reactions. I don't think our engine is operating at the pressures and temperatures needed for that.

Intake manifold air temp (actually at throttle body) looks ok. But should not matter regardless. The air mass should not change.

Not sure if there's something funky going on with the motec ecu reporting. I'm calculating the fuel mass from logged injector duty cycle % (primary and secondary at same % with same injectors).

After thinking about this some more a few options come to mind.

Could the chassis dyno room air be getting contaminated with exhaust gas and causing this much of a reading deviation?

Or could the lambda sensor heating rate cause a false rich reading on the end half of the pull?

I'm not very experienced with dyno use so not sure if either of these are possible.

Is this engine running an inlet restrictor?

Yes. Data is from a engine running a FIA 45mm inlet air restrictor.

This allows for calculating the air massflow from restrictor choke pressure (when at high enough flows). Interesting point though, could the choke pressure reading be going off due to the sensor ducting freezing shut? Seems a little unlikely though, since it seems to respond immediately to lifting off the throttle at the end of the run.