fix: raise reward margin to 1.35x H100 cost

docs/AUDIT_PREP.md

@@ -20,7 +20,7 @@ Mainnet must use freshly generated program keypairs (see `docs/launch/MAINNET_DE
 - Audit probability from `verification_percent`
 - Tolerance band 0.05 operating point and within-band damage budget
 - `slash_bounty_bps` reporter share on forfeit
-- Zero-fraud treasury burn projection (~8.15% of rewards at p=0.1)
+- Zero-fraud treasury burn projection (~8.15% of rewards at p=0.1 with 1.35x reward margin)
 
 ## Attachments for the auditor
 

docs/TOKENOMICS.md

@@ -5,13 +5,6 @@ Issue #6. Design consistent with the economic model already coded in
 engineering and product design document. It is not a legal opinion. Issue #5
 (counsel review) gates any public token offer.
 
-## Token
-
-The ticker is **$LEVI**. The name is set; the mainnet launch is deliberately
-sequenced after a credible public run demonstrates the bonded training economy
-working end to end, not before. A token with no proven protocol behind it has
-nothing to price.
-
 ## Design goals
 
 1. Pay for accepted learning work (Proof of Gradient), not for raw hash power.
@@ -39,7 +32,7 @@ From `genesis_parameters()` and `calibration_table()`:
 
 | Preset | Round cost (H100) | Round reward (1.35x) | Bond at p=0.1 | Expected catch rounds |
 |---|---|---|---|---|
-| 125M proof | $0.0120 | $0.0162 | $0.146 | 10 |
+| 125M proof | $0.0120 | $0.0162 | $0.15 | 10 |
 | 1B genesis | $0.240 | $0.324 | $2.91 | 10 |
 | 7B scale | $3.36 | $4.53 | $40.79 | 10 |
 
@@ -61,6 +54,9 @@ be sized for it.
 ## Supply sketch (pre-legal, adjustable)
 
 Working totals for modelling only. Counsel may require structural change.
+These percentages live only in this document. Round economics in
+`economy.py` and `docs/assets/tokenomics.json` are bond/reward/audit math, not
+supply allocation.
 
 | Allocation | Share | Vesting | Purpose |
 |---|---|---|---|
@@ -71,6 +67,11 @@ Working totals for modelling only. Counsel may require structural change.
 | Early contributors / community | 10% | TGE unlock + short vest | Genesis participants, bug bounties |
 | Liquidity / market making | 5% | At TGE under multisig policy | CEX/DEX depth if pursued |
 
+Total 100%. The +10% team allocation was taken entirely from the training
+rewards endowment (45% to 35%). Training rewards remain the largest single
+bucket (35% > 25%), so the thesis that the network that trains the model owns
+the primary share of emission still holds.
+
 Emission is not a fixed block subsidy. Each run configures epoch earning rates on
 the coordinator; the treasury tops up the run vault. Unused endowment stays
 unminted or locked, never inflated ad hoc.

docs/assets/tokenomics.json

@@ -161,5 +161,6 @@
     }
   ],
   "slash_bounty_fraction_default": 0.5,
-  "slash_bounty_bps_default": 5000
+  "slash_bounty_bps_default": 5000,
+  "reward_margin": 1.35
 }

sim/leviathan_sim/economy.py

@@ -57,9 +57,6 @@ def break_even_bond(audit_probability: float, reward_per_round: float) -> float:
     return reward_per_round * (1.0 - audit_probability) / audit_probability
 
 
-REWARD_COST_MULTIPLIER = 1.35
-
-
 @dataclass(frozen=True)
 class RunPreset:
     label: str
@@ -85,11 +82,15 @@ def h100_round_cost_usd(
     return seconds / 3600.0 * usd_per_gpu_hour
 
 
+REWARD_MARGIN = 1.35
+AUDIT_FEE_MULTIPLIER = 1.1
+
+
 def calibration_table(audit_probabilities: list[float]) -> list[dict]:
     rows = []
     for preset in PRESETS:
         cost = h100_round_cost_usd(preset)
-        reward = REWARD_COST_MULTIPLIER * cost
+        reward = REWARD_MARGIN * cost
         for p in audit_probabilities:
             rows.append(
                 {
@@ -107,15 +108,12 @@ def calibration_table(audit_probabilities: list[float]) -> list[dict]:
 def audit_burn_projection(
     audit_probabilities: list[float],
     n_workers: int = 100,
-    fee_multiplier: float = 1.1,
+    fee_multiplier: float = AUDIT_FEE_MULTIPLIER,
 ) -> list[dict]:
-    """Zero-fraud equilibrium: with nobody to slash, verifier income is pure
-    treasury burn. A replay audit redoes one contribution's compute, so the fee
-    floor is the per-contribution H100 cost times a verifier margin."""
     rows = []
     for preset in PRESETS:
         cost = h100_round_cost_usd(preset)
-        reward = REWARD_COST_MULTIPLIER * cost
+        reward = REWARD_MARGIN * cost
         for p in audit_probabilities:
             fee = fee_multiplier * cost
             burn = p * n_workers * fee
@@ -135,13 +133,9 @@ def audit_burn_projection(
 
 
 def genesis_parameters(audit_probability: float = 0.1, band: float = 0.05) -> dict:
-    """The published operating point for the Genesis Run testnet, derived from
-    the same (1-p)/p discipline as the calibration table. Values are per worker
-    per the 1B preset; testnet bonds use these numbers denominated in the
-    testnet collateral mint."""
     preset = next(p for p in PRESETS if p.params_billion == 1.0)
     cost = h100_round_cost_usd(preset)
-    reward = REWARD_COST_MULTIPLIER * cost
+    reward = REWARD_MARGIN * cost
     return {
         "preset": preset.label,
         "audit_probability": audit_probability,
@@ -150,5 +144,7 @@ def genesis_parameters(audit_probability: float = 0.1, band: float = 0.05) -> di
         "bond_usd": break_even_bond(audit_probability, reward),
         "bond_rounds_of_reward": (1.0 - audit_probability) / audit_probability,
         "expected_rounds_to_catch": 1.0 / audit_probability,
-        "audit_burn_share_of_rewards": audit_probability * 1.1 / REWARD_COST_MULTIPLIER,
+        "audit_burn_share_of_rewards": (
+            audit_probability * AUDIT_FEE_MULTIPLIER / REWARD_MARGIN
+        ),
     }

sim/tests/test_economy.py

@@ -1,5 +1,4 @@
 from leviathan_sim.economy import (
-    REWARD_COST_MULTIPLIER,
     EconomyConfig,
     StakeLedger,
     audit_burn_projection,
@@ -50,7 +49,7 @@ def test_calibration_bond_scales_with_reward():
 
 def test_zero_fraud_burn_share_is_preset_independent():
     rows = audit_burn_projection([0.1])
-    expected = 0.1 * 1.1 / REWARD_COST_MULTIPLIER
+    expected = 0.1 * 1.1 / 1.35
     for row in rows:
         assert abs(row["burn_share_of_rewards"] - expected) < 1e-9
         assert row["treasury_burn_per_round_usd"] > 0.0