diff --git a/docs/defense/LIMITATIONS.md b/docs/defense/LIMITATIONS.md index 4147c9d..44c9fe9 100644 --- a/docs/defense/LIMITATIONS.md +++ b/docs/defense/LIMITATIONS.md @@ -19,10 +19,15 @@ The benchmark is a synthetic fixture: an analytic trajectory with the filter initialised at the truth plus an error drawn from its own declared prior ([ADR-0027](../adr/0027-how-the-synthetic-runs-are-initialised.md)). Two public recorded datasets (EuRoC and TUM VI) can be fetched and evaluated, with GNSS simulated from their ground truth, and one -study uses real camera frames. None of that data is vendored. - -**Consequence:** no number here is a field measurement. GNSS is never real, so -receiver latency, multipath, lever arm and real interference are not characterised. +study uses real camera frames. A third (UrbanNav) has a real GNSS receiver: one u-blox +F9P on two recordings, scored against a post-processed reference whose error is unknown +([UrbanNav](../urbannav.md)). None of that data is vendored. + +**Consequence:** no number here is a field measurement. GNSS is real only on the two +UrbanNav recordings, from one receiver, and simulated everywhere else. That is recorded +data, not field validation: one receiver, one reference of unknown error, and outages that +are either the receiver's own or imposed on its recorded fixes. Receiver latency, +multipath, lever arm and real interference are not characterised. Bias stability, temperature behaviour and vibration response are only as good as the public recordings.