l2gen version consistency for re-processing full dataset for MODIS and VIIRS

[andrei_mitrich]

Dear Forum experts,

We are processing L1A data for MODIS, VIIRSN and VIIRSJ1 with SEADAS l2gen, to L1C, which effectively means L2 with all gains set to 1 for use in processing with the POLYMER atmos correction that has its own SVC gains. We have noticed that in NASA reprocessings (say, r2018, r2022) all available data are processed with the same, latest version of software, say, R2022.1, then for later processing, newer data inputs, the latest available version is used, say, 9.7-V2024.7, 9.11-V2025.5 etc). Our tests show that different versions of software could lead in L1C data differences about 1%, in some examples, with following processing to Level2 magnify the differences by ~10, i.e., differences are roughly within 10%, in some examples - even higher. Effectively, this means we are observing 10% and above errors. Then two questions arise:
Which way is best/correct/recommended:

1) applying always the same l2gen version, as at SVC processing, even if newer versions appear;
2) regardless version used for SVC (i.e., at re-processing), always use the newest available - exactly what NASA is doing in operational processing;

Suspecting the way (2) would produce less accurate results, why NASA follows this way of processing?

Thank you
Andrei