josh/AIHostingTycoon
1
0
Fork 0
Code Issues 1 Pull Requests Actions Packages Projects Releases Wiki Activity

Remove benchmark evaluation system, use training capabilities directly

Browse Source 
Model quality for market segments and product lines now derives from deployed
model capabilities (coding, reasoning, agents, etc.) instead of requiring a
separate manual benchmark evaluation step. This eliminates an unbounded
benchmarkResults[] array that was scanned 5x per tick and removes ~480 lines
of dead-weight UI, types, and engine code.

Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
This commit is contained in:
Josh Wright
2026-04-26 19:28:59 -04:00
parent db034687d6
commit bbb69a315c
10 changed files with 57 additions and 535 deletions
Download Patch File Download Diff File Expand all files Collapse all files
packages/game-engine/src/systems/market/index.ts
+24 -24
View File
@@ -1,7 +1,6 @@
import type { GameState, MarketState, BenchmarkResult } from '@ai-tycoon/shared';
import type { GameState, MarketState, ModelCapabilities } from '@ai-tycoon/shared';
import { CONSUMER_TOKENS_PER_SUBSCRIBER, API_TOKENS_PER_DEVELOPER_PER_TICK, BATCH_API_DEMAND_PER_DEV, makeInitialServingMetrics } from '@ai-tycoon/shared';
import type { TrafficPriority, TierServingMetrics } from '@ai-tycoon/shared';
import { BENCHMARKS } from '../../data/benchmarks';
import { computeSeasonal } from './seasonalSystem';
import { updateObsolescence } from './obsolescenceSystem';
import { buildPlayerProfile, buildCompetitorProfile, computeMarketShares, updateTAMGrowth } from './tamSystem';
@@ -21,31 +20,30 @@ export interface MarketTickResult {
totalTokenDemand: number;
}
const SEGMENT_CAPABILITY_WEIGHTS: Record<string, Partial<Record<keyof ModelCapabilities, number>>> = {
consumer: { creative: 0.35, knowledge: 0.25, reasoning: 0.15, multimodal: 0.15, coding: 0.05, agents: 0.05 },
enterprise: { reasoning: 0.25, coding: 0.20, agents: 0.20, knowledge: 0.15, math: 0.10, multimodal: 0.10 },
developer: { coding: 0.35, reasoning: 0.20, agents: 0.20, math: 0.15, knowledge: 0.10 },
research: { reasoning: 0.30, math: 0.30, knowledge: 0.20, coding: 0.10, agents: 0.10 },
};
function getSegmentQuality(
segment: 'consumer' | 'enterprise' | 'developer' | 'research',
benchmarkResults: BenchmarkResult[],
capabilities: ModelCapabilities,
fallbackScore: number,
): number {
if (benchmarkResults.length === 0) return fallbackScore / 100;
const bestByBenchmark = new Map<string, number>();
for (const r of benchmarkResults) {
const prev = bestByBenchmark.get(r.benchmarkId) ?? 0;
if (r.score > prev) bestByBenchmark.set(r.benchmarkId, r.score);
}
const weights = SEGMENT_CAPABILITY_WEIGHTS[segment];
if (!weights) return fallbackScore / 100;
let weightedSum = 0;
let totalWeight = 0;
for (const bench of BENCHMARKS) {
const score = bestByBenchmark.get(bench.id);
if (score == null) continue;
const weight = bench.marketRelevance[segment];
weightedSum += (score / 100) * weight;
totalWeight += weight;
for (const [cap, weight] of Object.entries(weights)) {
const score = capabilities[cap as keyof ModelCapabilities] ?? 0;
if (score > 0) {
weightedSum += (score / 100) * weight;
totalWeight += weight;
}
}
if (totalWeight === 0) return fallbackScore / 100;
return weightedSum / totalWeight;
return totalWeight > 0 ? weightedSum / totalWeight : fallbackScore / 100;
}
export function processMarketV2(
@@ -54,9 +52,11 @@ export function processMarketV2(
effectiveInferenceFlops?: number,
researchBonuses?: ResearchBonuses,
): MarketTickResult {
const consumerQuality = getSegmentQuality('consumer', state.models.benchmarkResults, state.models.bestDeployedModelScore);
const enterpriseQuality = getSegmentQuality('enterprise', state.models.benchmarkResults, state.models.bestDeployedModelScore);
const modelQuality = state.models.benchmarkResults.length > 0
const caps = state.models.bestDeployedCapabilities;
const hasDeployed = state.models.bestDeployedModelScore > 0;
const consumerQuality = getSegmentQuality('consumer', caps, state.models.bestDeployedModelScore);
const enterpriseQuality = getSegmentQuality('enterprise', caps, state.models.bestDeployedModelScore);
const modelQuality = hasDeployed
? (consumerQuality + enterpriseQuality) / 2
: state.models.bestDeployedModelScore / 100;
@@ -115,7 +115,7 @@ export function processMarketV2(
const productResult = processProductLines(
state.market.codeAssistant,
state.market.agentsPlatform,
state.models.benchmarkResults,
caps,
playerDevCustomers,
playerEntCustomers,
seasonal.multipliers.consumer,