Nice, just looked at dory. Cool to see different takes on the same problem. The write gate and the manual curation approach converge on the same insight: the model shouldn't necessarily decide what's permanently important.