sdods heal report -p storefront --write-historyThe docs say it biases future heals towards strategies that worked. Before I put it in the nightly I would like to know how much bias, and whether running it is reversible.
Accepted answer
Small, and yes.
The flag writes heal-history.json beside the artifacts directory, holding one entry per heal description with its success and failure counts and the strategy and selector that last worked. On the next run, a candidate whose strategy and selector match a remembered winner gets a bonus of successes over total, scaled by 0.1 and capped there, added after the multipliers.
So the most it can ever do is lift a candidate by a tenth. In practice that matters in exactly one place: it lets a candidate that keeps landing just short of minScore clear it, and it breaks ties between two otherwise equal candidates. It cannot promote a CSS candidate above the default threshold, and it cannot make a candidate that matches nothing win, because a zero count is a zero score before the bonus is added.
Reversible: delete the file. A missing or corrupt one is ignored rather than fatal, and the counts rebuild the next time you run with the flag.
Have a different question? Ask it.