Biggest miss is that I failed to guess that the distributions were identical up to a constant person/resonance multiplier. Actually I did guess that initially, but decided it probably wasn’t true, which is why it’s my biggest miss. I started thinking it when Maria hit more x1s than x0s on Gamma while Janelle hit more x0s than x1s, leading me to think that there were more person-dependent factors than just one. IIRC the nail in the coffin of that theory was looking at Epsilon resonance. It was clear that Epsilon was non-random and inspecting the overlapping areas of the curve showed (e.g.)
Later, I lowered my confidence in Will’s Epsilon prediction because I knew our instruments have limited precision. I didn’t connect that with “maybe 0.77 vs 0.86 isn’t that far off given imprecision!”.
Also I think I was modeling the precision incorrectly, probably. I took “for example, since they say Earwax has an amplitude of 3.2 kCept, you can be 100% sure the true value is between 3.15 and 3.25 kCept” to mean that every value could be plus or minus 0.05, but I think now it actually meant that values were rounded to the nearest digit shown, so a listed value of 0.28 kCept was not between 0.23 and 0.33, but rather between 0.275 and 0.285?
Biggest hit, I think, was correctly determining Janelle’s actual chances: I said 25% win, 39% double; actual was 37% win, 30% double. Method was seeing graphs that were clearly 5 linear trends by power, estimating the zero, estimating their slopes, and noticing the multiplier.
Also I think I was modeling the precision incorrectly, probably. I took “for example, since they say Earwax has an amplitude of 3.2 kCept, you can be 100% sure the true value is between 3.15 and 3.25 kCept” to mean that every value could be plus or minus 0.05, but I think now it actually meant that values were rounded to the nearest digit shown, so a listed value of 0.28 kCept was not between 0.23 and 0.33, but rather between 0.275 and 0.285?
Yes, that’s exactly what happened. That ambiguity didn’t occur to me; I’ve now edited the original post to clarify so future players won’t have the same issue; mea culpa.
This was a fun one! Post-mortem:
Biggest miss is that I failed to guess that the distributions were identical up to a constant person/resonance multiplier. Actually I did guess that initially, but decided it probably wasn’t true, which is why it’s my biggest miss. I started thinking it when Maria hit more x1s than x0s on Gamma while Janelle hit more x0s than x1s, leading me to think that there were more person-dependent factors than just one. IIRC the nail in the coffin of that theory was looking at Epsilon resonance. It was clear that Epsilon was non-random and inspecting the overlapping areas of the curve showed (e.g.)
A|Janelle|Maria|multiplier
-----
0.28|0.17|0.21|0.86
1.28|0.20|0.26|0.77
Later, I lowered my confidence in Will’s Epsilon prediction because I knew our instruments have limited precision. I didn’t connect that with “maybe 0.77 vs 0.86 isn’t that far off given imprecision!”.
Also I think I was modeling the precision incorrectly, probably. I took “for example, since they say Earwax has an amplitude of 3.2 kCept, you can be 100% sure the true value is between 3.15 and 3.25 kCept” to mean that every value could be plus or minus 0.05, but I think now it actually meant that values were rounded to the nearest digit shown, so a listed value of 0.28 kCept was not between 0.23 and 0.33, but rather between 0.275 and 0.285?
Biggest hit, I think, was correctly determining Janelle’s actual chances: I said 25% win, 39% double; actual was 37% win, 30% double. Method was seeing graphs that were clearly 5 linear trends by power, estimating the zero, estimating their slopes, and noticing the multiplier.
Yes, that’s exactly what happened. That ambiguity didn’t occur to me; I’ve now edited the original post to clarify so future players won’t have the same issue; mea culpa.