In the interest of making a concrete proposal, I suggest that we forced photometer DIAObjects for 30 days after their last DIASource detection. I think that's long enough that most transients will have faded completely into the noise. So every time a new DIASource detection arrives the counter resets.

If 30 days is sufficient, then I believe this just requires updating DMS-REQ-0317 with a caveat on which DIAObjects get photometered and a parameter set to 30 days. The sizing model already assumes this duration, and we can't find anything that says 12 months that would require changing.

Some of the notes attached to bottom of the DPDD include:

• SAC: do we need more than 30 days of precovery fluxes for new DIASources? ZI: yes, 30 days is only about two data points in a given band and we don't know colors for transients

We should double check that we are satisfied with the 30 day precovery window along with deciding on the post-discovery window length.

My view is that a 30 day precovery window is a decent balance between the technical and scientific concerns.

Precovery measurements tell us a) if there is flux below threshold and b) when the last non-detection was (flux indistinguishable from noise). Most classes of supernovae and other explosive transients will rise to peak within 30 days, so a 30-day window would be sufficient to capture all the information LSST has if it were discovered at peak brightness. For the tiny minority of sources that exhibit longer initial activity below threshold, users can use the precovery service.

Zeljko's point about only two data points per band is a limitation of the cadence rather than the precovery window itself. Extending the window further back doesn't help much, either, because the color may be changing on ~week timescales as the transient evolves.

Stephen Smartt provided some useful comments on how long to force photometer sources after their last detection here.