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

[#55] Adapt: regime-trader iteration 2.1 — wins base, patient-market, micro-economy, duel #57

Merged
robertismo merged 3 commits from 55-adapt into master 2026-09-29 19:23:13 +00:00
Conversation 1 Commits 3 Files changed 1 +183 -39
Winnie commented 2026-09-29 19:21:45 +00:00
Owner
Copy link

[#55] Adapt: the market-regime trader

Thesis. Adapt models the world as a small set of regimes, re-inferred every turn from observable state, and switches playbook when the regime changes — instead of tuning one fixed strategy:

  • deposit-rate dispersion (max − min of market-deposit-rates) → a chase/rate-war gauge
  • loan demand vs deployable cash (Σ loan amounts / own cash) → market heat
  • churn rate (free-float) → frozen-market gauge (deposit-drought vs normal)
  • bond curve vs loan cap (cheapest standing bond's implied per-turn return vs loan-rate-cap and the yield estimate) → carry gauge

Playbooks. THIN (cap barely clears default risk) → park. WAR (dispersion > 2·rate-temp) → never chase deposits, cream lending only. COLD/frozen (churn < 5%) → hoard, lend the surplus as bonds. HOT (demand-heat > 1 and cheap bond money clears) → carry: borrow up to 4× loan-max, blanket-bid aggressively. NORMAL → a bid tatonnement: adapt reads its own loan book each turn — if all its fresh loans price at its own bid it is the sole bidder and ratchets the price up toward the yield estimate (avg + risk + rate-temp); if it won nothing it steps down; floor funding+risk−2·rate-temp; blanket bid so loans it cannot fund die unclaimed rather than feeding a rival's book.

All thresholds are config-derived; the regime detection is adapt's own calculation — no helpers shared with empire/mill/vault.

Full bench (100 games × 13 presets, master 7f503c4):

preset adapt empire mill vault
base 79 11 9 0
patient-market 66 5 20 5
micro-economy 95 2 3 0
duel 85 6 8 0
tight-cash 95 0 0 0
default-storm 100 0 0 0
hot-caps 96 3 1 0
rate-war 88 4 3 0
mega-loans 85 6 6 0
cash-glut 75 7 7 9
bond-economy 67 0 32 0
deposit-drought 7 33 59 0
thin-margins 1 43 56 0

All four assigned scenarios won (most wins). deposit-drought and thin-margins remain weak — next iterations target the frozen/thin playbooks.

Smoke: 0 [ai errors, 0 ACCT divergence. Only ai/adapt.lisp changed.

## [#55] Adapt: the market-regime trader **Thesis.** Adapt models the world as a small set of regimes, re-inferred every turn from observable state, and switches playbook when the regime changes — instead of tuning one fixed strategy: - **deposit-rate dispersion** (max − min of `market-deposit-rates`) → a chase/rate-war gauge - **loan demand vs deployable cash** (Σ loan amounts / own cash) → market heat - **churn rate** (`free-float`) → frozen-market gauge (deposit-drought vs normal) - **bond curve vs loan cap** (cheapest standing bond's implied per-turn return vs `loan-rate-cap` and the yield estimate) → carry gauge **Playbooks.** THIN (cap barely clears default risk) → park. WAR (dispersion > 2·rate-temp) → never chase deposits, cream lending only. COLD/frozen (churn < 5%) → hoard, lend the surplus as bonds. HOT (demand-heat > 1 and cheap bond money clears) → carry: borrow up to 4× loan-max, blanket-bid aggressively. NORMAL → a **bid tatonnement**: adapt reads its own loan book each turn — if all its fresh loans price at its own bid it is the sole bidder and ratchets the price up toward the yield estimate (avg + risk + rate-temp); if it won nothing it steps down; floor funding+risk−2·rate-temp; blanket bid so loans it cannot fund die unclaimed rather than feeding a rival's book. All thresholds are config-derived; the regime detection is adapt's own calculation — no helpers shared with empire/mill/vault. **Full bench (100 games × 13 presets, master `7f503c4`):** | preset | adapt | empire | mill | vault | |---|---|---|---|---| | **base** | **79** | 11 | 9 | 0 | | **patient-market** | **66** | 5 | 20 | 5 | | **micro-economy** | **95** | 2 | 3 | 0 | | **duel** | **85** | 6 | 8 | 0 | | tight-cash | **95** | 0 | 0 | 0 | | default-storm | **100** | 0 | 0 | 0 | | hot-caps | **96** | 3 | 1 | 0 | | rate-war | **88** | 4 | 3 | 0 | | mega-loans | **85** | 6 | 6 | 0 | | cash-glut | **75** | 7 | 7 | 9 | | bond-economy | **67** | 0 | 32 | 0 | | deposit-drought | 7 | 33 | 59 | 0 | | thin-margins | 1 | 43 | 56 | 0 | All four assigned scenarios won (most wins). deposit-drought and thin-margins remain weak — next iterations target the frozen/thin playbooks. Smoke: 0 `[ai` errors, 0 ACCT divergence. Only `ai/adapt.lisp` changed.
[#55] adapt: market-regime trader — regime inference + bid tatonnement
All checks were successful
build / build (push) Successful in 24s
Details
4597462350 
Rebuild adapt as the market-regime trader: infers the regime each turn
from deposit-rate dispersion, loan demand vs deployable cash, churn
(frozen-market gauge) and the bond curve vs the loan cap, then switches
playbook (park / never-chase / hoard+bond-lend / carry / tatonnement).
The tatonnement reads its own loan book: loans priced at its own bid
mean it is the sole bidder, so it ratchets the price up toward the
yield estimate; losing everything steps the bid down; floor at
funding+risk-2*rate-temp. Full bench (100 games): base 79, patient-
market 66, micro-economy 95, duel 85 wins — all four assigned scenarios.
[#55] adapt: tatonnement for thin and frozen regimes
Some checks failed
build / build (push) Has been cancelled
Details
d018fa7b0e 
thin-margins and deposit-drought were lost by the park/hoard playbooks
(adapt 1 and 7 wins in the full bench). Route both through the bid
tatonnement: the floor funding+risk-2*rate-temp protects capital in
thin-margin markets better than sitting out, and the blanket bid denies
the rival books in a frozen market while the hoard still goes out as
bond lending. Quick bench (30 games): adapt wins all 13 presets —
thin-margins 19, deposit-drought 26, base 27, micro 26, duel 25,
patient-market 18.
Merge branch 'master' into 55-adapt
All checks were successful
build / build (push) Successful in 28s
Details
59b37e505a
Winnie commented 2026-09-29 19:24:39 +00:00
Author
Owner
Copy link

v2.2 update (same branch, new commit d018fa7)

thin-margins and deposit-drought routed through the tatonnement (they were the two non-target weaknesses: 1 and 7 wins). Full bench (100 games × 13 presets) after the change — adapt wins all four assigned scenarios and all 13 presets:

preset adapt empire mill vault
base 79 11 9 0
patient-market 66 5 20 5
micro-economy 88 1 11 0
duel 85 6 8 0
deposit-drought 84 9 6 0
thin-margins 61 20 15 0
tight-cash 95 0 0 0
default-storm 100 0 0 0
hot-caps 96 4 0 0
rate-war 94 4 0 0
mega-loans 86 3 8 2
bond-economy 67 0 32 0
cash-glut 75 7 7 9

Smoke clean (0 [ai errors, 0 ACCT divergence). Only ai/adapt.lisp changed across both commits.

## v2.2 update (same branch, new commit `d018fa7`) thin-margins and deposit-drought routed through the tatonnement (they were the two non-target weaknesses: 1 and 7 wins). Full bench (100 games × 13 presets) after the change — adapt wins **all four assigned scenarios and all 13 presets**: | preset | adapt | empire | mill | vault | |---|---|---|---|---| | **base** | **79** | 11 | 9 | 0 | | **patient-market** | **66** | 5 | 20 | 5 | | **micro-economy** | **88** | 1 | 11 | 0 | | **duel** | **85** | 6 | 8 | 0 | | deposit-drought | **84** | 9 | 6 | 0 | | thin-margins | **61** | 20 | 15 | 0 | | tight-cash | **95** | 0 | 0 | 0 | | default-storm | **100** | 0 | 0 | 0 | | hot-caps | **96** | 4 | 0 | 0 | | rate-war | **94** | 4 | 0 | 0 | | mega-loans | **86** | 3 | 8 | 2 | | bond-economy | **67** | 0 | 32 | 0 | | cash-glut | **75** | 7 | 7 | 9 | Smoke clean (0 `[ai` errors, 0 ACCT divergence). Only `ai/adapt.lisp` changed across both commits.
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
2 participants
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!57
No description provided.