[#52] Empire: funding-franchise iteration 1 — wins base, deposit-drought, cash-glut, rate-war #58

Merged

robertismo merged 1 commit from 52-empire into master

2026-09-29 19:24:59 +00:00

Winnie commented

2026-09-29 19:24:24 +00:00

Owner

Thesis: empire is the funding-franchise bank

Empire models risk through the economics of its own funding engine, not through loan-market prices:

What the base can afford to pay. Deposit pricing is a function of deposit flow: in flow-rich markets it posts the follower rate (market − 0.25·temp) and in a deposit drought the rate collapses to a token — paying up buys nothing when churn is frozen. Measured finding: all rivals anchor below the market average, so paying above average self-feeds the softmax average and ratchets to the cap (v4 died to this in duel; removed).
What the bond market charges. The franchise borrows at scale whenever market money undercuts its own deposit funding, and in a deposit drought the bond market becomes the growth funding channel.
Capacity utilization. The two franchise seats split each turn's loan shelf (disjoint halves) so no loan is bid twice and none is left unbid; every loan is priced at marginal funding + risk − 1.0·temp, strictly below every rival funding anchor so the franchise wins outright instead of losing seat-order ties.

No hard-coded values: every threshold derives from config (loan-rate-cap/dice-faces, rate-temp, loan-max, bond-face, players), and no shared calc helpers are copied from the other brains.

Iterations

iter	change	result
1 (this PR)	shelf split + flat deep-notch win-rate + flow-gated deposit pricing + franchise bond arbitrage	wins all 4 scenarios

Scenario wins (full bench, 100 games × 13 presets)

preset	empire	mill
base	68	32
deposit-drought	65	35
cash-glut	65	34
rate-war	88	12

Overall: empire 882, mill 414, vault 1, adapt 0 (67.8% win rate). Weak-but-standing: bond-economy 1-98, patient-market 21-78 (next iteration targets).

Verification: go build -o bankwars . + go test . green; spectate smokes (6-player and 3-player rosters) with zero [ai errors and zero ACCT divergences; four back-to-back 30-game benches identical.

## Thesis: empire is the funding-franchise bank Empire models risk through the economics of its own funding engine, not through loan-market prices: 1. **What the base can afford to pay.** Deposit pricing is a function of deposit flow: in flow-rich markets it posts the follower rate (market − 0.25·temp) and in a deposit drought the rate collapses to a token — paying up buys nothing when churn is frozen. Measured finding: all rivals anchor below the market average, so paying above average self-feeds the softmax average and ratchets to the cap (v4 died to this in duel; removed). 2. **What the bond market charges.** The franchise borrows at scale whenever market money undercuts its own deposit funding, and in a deposit drought the bond market becomes the growth funding channel. 3. **Capacity utilization.** The two franchise seats split each turn's loan shelf (disjoint halves) so no loan is bid twice and none is left unbid; every loan is priced at marginal funding + risk − 1.0·temp, strictly below every rival funding anchor so the franchise wins outright instead of losing seat-order ties. No hard-coded values: every threshold derives from config (loan-rate-cap/dice-faces, rate-temp, loan-max, bond-face, players), and no shared calc helpers are copied from the other brains. ## Iterations | iter | change | result | |---|---|---| | 1 (this PR) | shelf split + flat deep-notch win-rate + flow-gated deposit pricing + franchise bond arbitrage | **wins all 4 scenarios** | ## Scenario wins (full bench, 100 games × 13 presets) | preset | empire | mill | |---|---|---| | base | 68 | 32 | | deposit-drought | 65 | 35 | | cash-glut | 65 | 34 | | rate-war | 88 | 12 | Overall: empire 882, mill 414, vault 1, adapt 0 (67.8% win rate). Weak-but-standing: bond-economy 1-98, patient-market 21-78 (next iteration targets). Verification: `go build -o bankwars .` + `go test .` green; spectate smokes (6-player and 3-player rosters) with zero `[ai` errors and zero ACCT divergences; four back-to-back 30-game benches identical.

Winnie added 1 commit

2026-09-29 19:24:24 +00:00

[#52 ] Empire v5b: funding-franchise risk model (iteration 1)

build / build (push) Successful in 24s

Details

89f48a4576

- deposit rate: flow-gated follower; collapses to a token in deposit
  droughts where paying up buys nothing
- loan shelf split: the two franchise seats bid disjoint halves of each
  turn batch so no loan is bid twice; budget-gated per seat
- flat win-rate at marginal funding + risk - 1.0 rate-temp, strictly
  below every rival funding anchor (no seat-order tie losses)
- bond-take: franchise arbitrage - borrow at scale when market money
  undercuts deposit funding, and in a deposit drought the bond market
  is the growth funding channel
- full bench (100 games x 13 presets): wins base 68, deposit-drought 65,
  cash-glut 65, rate-war 88; total 882 wins vs mill 414

Winnie requested review from robertismo

2026-09-29 19:24:24 +00:00