Thank you all
Three weeks ago I posted here asking for testers. Since then you've analysed 186 games and left 45 section ratings, 21 of them with written detailed comments. Some of those comments were long and very precise, and a few were blunt ("it scored me 11 points, what more do you want"). All of them were useful. Thank you.
Here is what you told me, and what I did with it.
Fixed and live now - "It never gives more than 8, even for near-perfect play." True — every report found some "miss" and capped the score. Small quibbles (under 2 VP) are now listed as possible improvements but no longer pull the score down, and a clean game can get a 9 or 10.
- 3-player games: the verdict claimed a 46-point win when the real margin over the nearest opponent was 15. The margin is now calculated vs the nearest opponent, not a random one.
- Re-analyze sometimes analysed your opponent instead of you. Fixed. If an older report still shows your opponent's game, reply here and I'll sort it out.
- "It criticised a play based on eggs I didn't have yet." The coach was looking at the end-of-game board. It now sees your board as it was before each play.
- "It said I had White-faced Ibis, but I didn't." Birds that were only in the tray are no longer credited to you.
- Reports failing with an error. About 1 in 6 failed some days. Retries and better error handling are in, and there have been no failures since 2 October.
- Smaller ones: conceded games no longer show 0–0 (now the score up to that point, without bonus cards), Photographer is recognised, and internal codes like "eggs:-1" no longer leak into the text.
What I'm working on next - A round-by-round report. Several of you asked for it: final score first, then patterns, starting hand, one section per round with a green/yellow/red rating and the reason, then what to focus on next game. The scoreboard (which BGA already shows you) gets smaller.
- Less tunnel vision on eggs. The coach leans too hard on egg-laying and grassland engines and doesn't recognise other plans, like big-point birds fed from the forest. It also needs to weigh how many turns are left before calling something a "miss".
- A smarter starting hand. It should consider the tray, the round goals and food-giving birds, and know which bonus cards are easy (Oologist) and which are hard.
- Bird power accuracy. A few powers are tagged with the wrong timing (Carolina Wren is a one-off, not an engine). I'm going through the card data.
- Deleting a report, and showing bird powers or pictures in the report (I have to find a smart way because I don't want to go against any trademark, maybe only the siloutte of the bird is enough).
To be honest: one tester said it could take a 200 player to 350, but not a 400 player to 500. I think that's fair today, and deeper analysis for experienced players is the longer goal.
If you have played since your last report, try it again:
https://boardgame.coach The thumbs on each section are still the best way to tell me what's wrong. 3–4 player games and losses are especially useful.
Thanks again, Tedoron
This is why the data science engine is the key component- it is what evaluates moves and gives you the feedback on how to improve.
