negative year inputs for Bmark_years in forecast file not working

[iantaylor-NOAA]

Piera Carpi (IPHC) got in touch after noticing that the Bmark_years reported in Forecast-report.sso all showed 2020 when she expected 2017 and 2020 based on forecast.ss inputs of -3 and 0.

I don't follow the logic of how the negative values are converted, but it looks like this line added 8 months ago to deal with uninitialized variables (https://vlab.ncep.noaa.gov/redmine/issues/80131) may have messed things up somehow.

I replicated the error with a version of the Pacific Hake model and it showed a small impact on the forecast spawning biomass, which makes me think that our automatic testing system could have caught the bug. @k-doering-NOAA, out of curiosity, did we have the automatic testing of the ss_examples models in place at that time (if I'm right that this is when the bug was introduced), and if so, is there a way to check for whether this change was visible but too small for us to flag?

[iantaylor-NOAA]

I should add (so that @kellijohnson-NOAA doesn't worry about it) that -999 inputs (as used by the hake model for Bmark_years) continue to be accurately translated into start year, and negative values for the Fcast_years input are also accurately translated. So the draft Pacific Hake assessment getting submitted next week should not be impacted by this issue.

[k-doering-NOAA]

Good thinking about automated testing, Ian! I just looked back at some old emails to remind myself of when we had automated testing set up, and it looks like that was just around the time when we started to make the routine more formal (on June 15 2020 there is a commit about saving the "fail file for the compare job" which indicates to me we were actually testing if values failed by then). My guess is that the issue with uninitialized parameters was the main concern. I remember some values in the models changing, but I guess we thought it was due to the uninit parameters being fixed, rather than a new bug introduced....

It is totally possible that we need to recalibrate our tolerance levels, though. I'm happy to dig into our testing set more to try to learn why this was missed.

[iantaylor-NOAA]

Thanks for looking @k-doering-NOAA. When the issue gets fixed I think we will see in reverse the same small change in forecast spawning biomass so will learn if it gets caught by the test or not.

[Rick-Methot-NOAA]

It all seems correct in the output I am looking at. And it is correct in forecast_report.ss_new

[Rick-Methot-NOAA]

-999 works as advertised; -3 does not. Will fix next

[k-doering-NOAA]

no changes were found on jenkins after this fix - I plan to investigate why tomorrow!