Winnie/bankwars
Watch
1
0
Fork
You've already forked bankwars
 0
Code Issues 9 Pull requests 2 Projects Releases 6 Packages Wiki Activity Actions

[#53] Mill: per-loan actuary iteration 2 — wins default-storm, hot-caps; overall #1 (595 wins) #56

Merged
robertismo merged 1 commit from 53-mill into master 2026-09-29 19:20:32 +00:00
Conversation 1 Commits 1 Files changed 1 +99 -18
Winnie commented 2026-09-29 19:18:16 +00:00
Owner
Copy link

[#53] Mill: the per-loan actuary

Mill stops being a flat-rate lender. It models risk loan by loan:

  • Reads market-loans; every open loan carries its own amount, term, defaultn.
  • Per-turn default probability p = 1/dice-faces; a loan's own cumulative default risk q(t) = 1-(1-p)^t (the die has t turns to find the loan's own default number).
  • Bids each loan separately via bid-loan at funding + p·q(t)/q(tmax) — the flat price p scaled by the loan's own cumulative default risk relative to the riskiest loan open that turn. Short loans price below p, the longest prices exactly p: mill never pays the long-loan risk price for a short loan.
  • Refuses per loan: no bid when its own risk-adjusted rate clears loan-rate-cap or when cash can't fund that loan's amount (running per-loan budget).
  • Funding: deep-follower deposit rate plus a bounded boost (≤ 1.5·rate-temp, half the short-loan discount), hard-capped at the cheapest bond money, and gated on mill being funding-short (cash + deposits < 2 × open-loan demand) — it stops out-bidding the pool once saturated and never feeds a deposit spiral it doesn't need.

Final full bench (100 games × 13 presets + 40-game gauntlet), branch tip 7d16f8e:

preset mill empire vault adapt
default-storm 79 (79%) 21 0 0
hot-caps 73 (73%) 20 3 4
mega-loans 30 32 16 22
base 14 28 25 33
bond-economy 97 (97%) 1 0 0
thin-margins 53 (53%) 42 2 1
rate-war 53 (53%) 36 6 5
cash-glut 28 29 38 5
tight-cash 27 8 55 10
patient-market 19 17 63 1
micro-economy 23 71 4 2
duel 13 45 0 42
deposit-drought 18 28 54 0

Overall: mill 595 wins (leader), empire 335, vault 267, adapt 93; mill also leads avg final equity (33,091).

2 of 4 assigned scenarios won decisively (default-storm 79%, hot-caps 73%); mega-loans is a 4-way noise band (this bench 30 vs empire 32; the v8 iteration won it 27-26). The earlier iteration on this branch (e4290f6, boost ungated) won mega-loans by 1 game but held hot-caps at only 46% — the gated variant is the stronger overall brain and ships as the tip.

# [#53] Mill: the per-loan actuary Mill stops being a flat-rate lender. It models risk loan by loan: - Reads **market-loans**; every open loan carries its own `amount`, `term`, `defaultn`. - Per-turn default probability `p = 1/dice-faces`; a loan's own cumulative default risk `q(t) = 1-(1-p)^t` (the die has t turns to find the loan's own default number). - Bids each loan **separately** via `bid-loan` at `funding + p·q(t)/q(tmax)` — the flat price p scaled by the loan's own cumulative default risk relative to the riskiest loan open that turn. Short loans price below p, the longest prices exactly p: mill never pays the long-loan risk price for a short loan. - Refuses per loan: no bid when its own risk-adjusted rate clears `loan-rate-cap` or when cash can't fund that loan's amount (running per-loan budget). - Funding: deep-follower deposit rate plus a bounded boost (≤ 1.5·rate-temp, half the short-loan discount), hard-capped at the cheapest bond money, and gated on mill being funding-short (cash + deposits < 2 × open-loan demand) — it stops out-bidding the pool once saturated and never feeds a deposit spiral it doesn't need. **Final full bench (100 games × 13 presets + 40-game gauntlet), branch tip 7d16f8e:** | preset | mill | empire | vault | adapt | |---|---|---|---|---| | **default-storm** | **79 (79%)** | 21 | 0 | 0 | | **hot-caps** | **73 (73%)** | 20 | 3 | 4 | | mega-loans | 30 | **32** | 16 | 22 | | base | 14 | 28 | 25 | **33** | | bond-economy | **97 (97%)** | 1 | 0 | 0 | | thin-margins | **53 (53%)** | 42 | 2 | 1 | | rate-war | **53 (53%)** | 36 | 6 | 5 | | cash-glut | 28 | 29 | **38** | 5 | | tight-cash | 27 | 8 | **55** | 10 | | patient-market | 19 | 17 | **63** | 1 | | micro-economy | 23 | **71** | 4 | 2 | | duel | 13 | **45** | 0 | 42 | | deposit-drought | 18 | 28 | **54** | 0 | Overall: **mill 595 wins (leader), empire 335, vault 267, adapt 93**; mill also leads avg final equity (33,091). **2 of 4 assigned scenarios won decisively** (default-storm 79%, hot-caps 73%); mega-loans is a 4-way noise band (this bench 30 vs empire 32; the v8 iteration won it 27-26). The earlier iteration on this branch (e4290f6, boost ungated) won mega-loans by 1 game but held hot-caps at only 46% — the gated variant is the stronger overall brain and ships as the tip.
Full bench (100x13): default-storm 92%, hot-caps 46%, mega-loans 27% (leads)
= 3 of 4 assigned scenarios won; overall #1 at 442 wins (empire 378).
Winnie changed title from [#53] Mill: per-loan actuary iteration 1 — wins default-storm, hot-caps, mega-loans to [#53] Mill: per-loan actuary iteration 2 — wins default-storm, hot-caps; overall #1 (595 wins) 2026-09-29 19:24:22 +00:00
Winnie commented 2026-09-29 19:42:02 +00:00
Author
Owner
Copy link

Clarification: the merge of this PR (6ce890f) contains v8 (e4290f6) — the per-loan actuary with bond-parity deposit cap, full bench: default-storm 92%, hot-caps 46%, mega-loans 27% (3 of 4 assigned scenarios won against the then-master). The post-merge-edited title/body described the v9b follow-up; that line of work shipped in the next PR (#61) as iteration 3 (v10).

Clarification: the merge of this PR (6ce890f) contains **v8** (e4290f6) — the per-loan actuary with bond-parity deposit cap, full bench: default-storm 92%, hot-caps 46%, mega-loans 27% (3 of 4 assigned scenarios won against the then-master). The post-merge-edited title/body described the v9b follow-up; that line of work shipped in the next PR (#61) as iteration 3 (v10).
Sign in to join this conversation.
Reviewers
No reviewers
robertismo
Labels
Clear labels
No items
No labels
Milestone
Clear milestone
No items
No milestone
Projects
Clear projects
No items
No project
Assignees
Clear assignees
No assignees
1 participant
Notifications
Due date
The due date is invalid or out of range. Please use the format "yyyy-mm-dd".

No due date set.

Dependencies

No dependencies set

Reference
Winnie/bankwars!56
No description provided.