Exit plan

[tbenthompson]

finishing up with confirmasaurus, ~1.5 days.

• implement the grouping technique.
  • re-run on [-1.0, 0]^4 with 100k sims. should take about 13 hours.
• get the tests passing, fix up validation
• write lots of comments and docstrings throughout the adagrid and clickhouse code.

writing, ~2-7 days

• write up the WD41 results
  • figures and what we want to show for the WD41 problem.
    • calibrated test vs WD design.
    • power gains, differences in rejection sets
      • cost/practicality/did we do better...
• write a general design doc/white paper on the implementation of the adagrid system. include a section on important potential upgrades

maybe not now:

• cleanup/touchup on imprint, see Coordination issue for making Imprint presentable #257

[tbenthompson]

todo list from mid april:

• if some tiles is not refined or deepened, that batch of tiles needs to be reinserted into groups as an "complete" group.
• get everything working with duckdb/calibration
• then do clickhouse/calibration
• then launch a big job and see how the scaling looks. try out 64 workers.
• twb_mean_lams vs twb_min_lams in the deepen_likely_to_work formula.
• verify is broken
  • check done count vs group countl
• continue!!

how to handle tiles that aren't refined or deepened?