| 1 | import type { WireUsage, TurnUsage } from "./types"; |
| 2 | |
| 3 | // Failed-attempt estimates remain billable but cannot replace calibrated |
| 4 | // context occupancy. Context* counters describe only the latest attempt. |
| 5 | export function measuredContextPromptTokens(usage?: WireUsage): number | undefined { |
| 6 | if (!usage || usage.estimated) return undefined; |
| 7 | return (usage.contextPromptTokens ?? 0) > 0 ? usage.contextPromptTokens : usage.promptTokens ?? 0; |
| 8 | } |
| 9 | |
| 10 | export function usageTotalTokens(usage?: WireUsage): number { |
| 11 | if (!usage) return 0; |
| 12 | if (usage.totalTokens > 0) return usage.totalTokens; |
| 13 | const promptTokens = usage.promptTokens || usage.cacheHitTokens + usage.cacheMissTokens; |
| 14 | return Math.max(0, promptTokens + usage.completionTokens); |
| 15 | } |
| 16 | |
| 17 | export function mergeChatTurnUsage(current: TurnUsage | undefined, usage: WireUsage | undefined): TurnUsage | undefined { |
| 18 | if (!usage) return current; |
| 19 | const route = usage.costQuote?.modelRef?.trim(); |
| 20 | const routes = current?.routes ? [...current.routes] : []; |
| 21 | if (route && !routes.includes(route)) routes.push(route); |
| 22 | const hasCacheBuckets = usage.cacheHitTokens > 0 || usage.cacheMissTokens > 0; |
| 23 | return { |
| 24 | uncachedInputTokens: (current?.uncachedInputTokens ?? 0) + (hasCacheBuckets ? usage.cacheMissTokens : usage.promptTokens), |
| 25 | outputTokens: (current?.outputTokens ?? 0) + usage.completionTokens, |
| 26 | totalTokens: (current?.totalTokens ?? 0) + usageTotalTokens(usage), |
| 27 | cacheReadTokens: (current?.cacheReadTokens ?? 0) + usage.cacheHitTokens, |
| 28 | reasoningTokens: (current?.reasoningTokens ?? 0) + (usage.reasoningTokens ?? 0), |
| 29 | routes: routes.length ? routes : undefined, |
| 30 | }; |
| 31 | } |
| 32 |