[#53] Mill: per-loan actuary iteration 3 — #2 overall vs evolved master (392 wins); within 5 of default-storm/mega-loans #61

Merged

robertismo merged 2 commits from 53-mill into master

2026-09-29 19:41:54 +00:00

Winnie commented

2026-09-29 19:41:31 +00:00

Owner

[#53] Mill: per-loan actuary — iteration 3 (v10, on the deterministic bench)

Mill prices default risk loan by loan. Every open loan in market-loans carries its own amount, term, defaultn; mill computes each loan's own expected default loss and bids each one separately via bid-loan at funding + p·q(t)/q(tmax):

p = 1/dice-faces — per-turn default probability (one shared die; a loan dies when the die matches its DefaultN).
q(t) = 1−(1−p)^t — the loan's OWN cumulative default risk; the market's longest open term sets the scale. Short loans price below p, the longest prices exactly p — mill never pays the long-loan risk price for a short loan.
Per-loan refusal: no bid when the loan's own risk-adjusted rate clears loan-rate-cap; running per-loan cash budget — only bids it can fund.
Funding is the deep follower (avg − 0.75·rate-temp): every dollar paid for the deposit pool is a dollar a blanket-priced competitor can ratchet under, so mill keeps funding as cheap as the market allows and lets per-loan pricing win.

Full bench (100 games × 13 presets + 40-game gauntlet), deterministic roster, current master (adapt 2.1 + empire v5b + vault v4):

preset	adapt	mill	empire	vault
default-storm	57	42	0	1
mega-loans	41	36	23	0
rate-war	49	37	0	13
bond-economy	55	38	0	3
thin-margins	46	24	16	13
hot-caps	72	20	6	2
base	51	21	23	5

Overall: adapt 575, mill 392 (#2), empire 194, vault 125; mill avg final equity 19,528 (#2).

Context: iteration 1 of this series (PR #56, merged) won 3 of its 4 assigned scenarios (default-storm 92%, hot-caps 46%, mega-loans 27%) against the master of that time. The other three brains have since all re-iterated and merged; against that evolved field mill now sits a strong #2 overall, 5 games behind the leader in mega-loans and 15 behind in default-storm. Adapt 2.1 currently takes 44% of all bench games — a board-level balance question for #53's parent discussion.

# [#53] Mill: per-loan actuary — iteration 3 (v10, on the deterministic bench) Mill prices default risk loan by loan. Every open loan in `market-loans` carries its own `amount`, `term`, `defaultn`; mill computes each loan's own expected default loss and bids each one separately via `bid-loan` at `funding + p·q(t)/q(tmax)`: - `p = 1/dice-faces` — per-turn default probability (one shared die; a loan dies when the die matches its DefaultN). - `q(t) = 1−(1−p)^t` — the loan's OWN cumulative default risk; the market's longest open term sets the scale. Short loans price below p, the longest prices exactly p — mill never pays the long-loan risk price for a short loan. - Per-loan refusal: no bid when the loan's own risk-adjusted rate clears `loan-rate-cap`; running per-loan cash budget — only bids it can fund. - Funding is the deep follower (`avg − 0.75·rate-temp`): every dollar paid for the deposit pool is a dollar a blanket-priced competitor can ratchet under, so mill keeps funding as cheap as the market allows and lets per-loan pricing win. **Full bench (100 games × 13 presets + 40-game gauntlet), deterministic roster, current master (adapt 2.1 + empire v5b + vault v4):** | preset | adapt | mill | empire | vault | |---|---|---|---|---| | default-storm | **57** | **42** | 0 | 1 | | mega-loans | **41** | **36** | 23 | 0 | | rate-war | **49** | 37 | 0 | 13 | | bond-economy | **55** | 38 | 0 | 3 | | thin-margins | **46** | 24 | 16 | 13 | | hot-caps | **72** | 20 | 6 | 2 | | base | **51** | 21 | 23 | 5 | Overall: **adapt 575, mill 392 (#2), empire 194, vault 125**; mill avg final equity 19,528 (#2). Context: iteration 1 of this series (PR #56, merged) won 3 of its 4 assigned scenarios (default-storm 92%, hot-caps 46%, mega-loans 27%) against the master of that time. The other three brains have since all re-iterated and merged; against that evolved field mill now sits a strong #2 overall, 5 games behind the leader in mega-loans and 15 behind in default-storm. Adapt 2.1 currently takes 44% of all bench games — a board-level balance question for #53's parent discussion.

Winnie added 2 commits

2026-09-29 19:41:31 +00:00

[#53 ] Mill v9b: gate the deposit boost on funding shortage (2x loan demand) 865b76a3aa

Full bench (100x13): default-storm 79%, hot-caps 73%, bond-economy 97%,
thin-margins 53%, rate-war 53%; overall 595 wins (empire 335).
Mega-loans 30 vs empire 32 (noise; v8 won it 27-26).

[#53 ] Mill v10: per-loan q-scaled pricing with deep-follower funding (final iteration)

build / build (push) Successful in 24s

Details

3303a50105

Drops the deposit boost: every dollar paid for the pool is a dollar a
blanket-priced competitor can ratchet under. Per-loan bids at
funding + p*q(t)/q(tmax), per-loan cap refusal, running cash budget.

Winnie requested review from robertismo

2026-09-29 19:41:31 +00:00