@thehammer/danx-session-telemetry 0.1.47 → 0.1.48

package/dist/.source-hash

@@ -1 +1 @@
-ecc5da92c3cb5fcb4ef6668b9d9307de4f31ba3665c1d9437fdc1f45bd6f8b06
+0662dcb067d30a9edb3819c5ee6ce60e7bdc40785373cb1beda5420b57d70e28

package/dist/pricing.d.ts

@@ -73,6 +73,8 @@ export interface PricedTokens {
 export declare const SYNTHETIC_MODEL = "<synthetic>";
 /** The `usage.speed` value a fast-mode reply carries. */
 export declare const FAST_SPEED = "fast";
+/** DX-4867 - a prompt longer than this many tokens (input + cache reads + cache writes) is billed at the model's long-prompt card. */
+export declare const LONG_PROMPT_TOKENS = 100000;
 /** DX-4593 - whether a `speed:"fast"` request on `model` is billed at the STANDARD row (see `FAST_BILLS_AT_STANDARD_RATE`): it has no fast row of its own. */
 export declare function billsFastAtStandardRate(model: string): boolean;
 /** One row of the list-price table: a model at a speed and its five per-million-token prices. */
@@ -119,6 +121,16 @@ export declare function priceUsage(model: string, speed: string, tokens: PricedT
  * standard table), or `null` when the tables have none. The default `PriceLookup` of {@link priceLedgerCredit}.
  */
 export declare function listedPrice(model: string, speed: string): ModelPrice | null;
+/** The prompt tokens of one reply: everything it sent in, cached or not. */
+export declare function promptTokensOf(tokens: PricedTokens): number;
+/** DX-4867 - whether `model` has a second, long-prompt rate card (a model with one card never switches, whatever the prompt length). */
+export declare function hasLongPromptCard(model: string): boolean;
+/**
+ * DX-4867 - price ONE reply (not a bucket of replies): like {@link priceUsage}, but a reply whose prompt is longer than
+ * {@link LONG_PROMPT_TOKENS} on a model with a long-prompt card is charged that card. A model with one rate card prices
+ * exactly as `priceUsage` does.
+ */
+export declare function priceReplyUsage(model: string, speed: string, tokens: PricedTokens): PriceResult;
 /**
  * DX-4595 - dollars for `tokens` at `price` (USD per million tokens): each bucket with its own price. The ONE copy of the formula: `priceUsage`
  * charges the list price with it, and spend pacing charges an account's own (editable) prices with it.

package/dist/pricing.js

@@ -94,9 +94,23 @@ const PRICING = {
     "claude-sonnet-4-6": price(3, 15, 0.3),
     "claude-sonnet-4-5-20250929": price(3, 15, 0.3),
     "claude-sonnet-4-20250514": price(3, 15, 0.3),
+    // DX-4867 - Claude Haiku 5.5, the page's row for prompts up to 100K tokens ($0.10 in / $0.50 out, cache hits
+    // 0.1x = $0.01, 5m write 1.25x, 1h write 2x). It has a SECOND rate card above 100K prompt tokens
+    // (LONG_CONTEXT_PRICING below): this row is the card a usage BUCKET is priced with, because a bucket sums
+    // many replies and cannot say which of them crossed the line. `priceReplyUsage` picks the card per reply.
+    "claude-haiku-5-5": price(0.1, 0.5, 0.01),
     "claude-haiku-4-5": price(1, 5, 0.1),
     "claude-haiku-4-5-20251001": price(1, 5, 0.1),
 };
+/** DX-4867 - a prompt longer than this many tokens (input + cache reads + cache writes) is billed at the model's long-prompt card. */
+export const LONG_PROMPT_TOKENS = 100_000;
+/**
+ * DX-4867 - the rate card of a reply whose prompt is longer than {@link LONG_PROMPT_TOKENS}, for the models that have
+ * one (Claude Haiku 5.5: $0.50 in / $2.50 out, cache hits 0.1x). Source: the pricing page's Claude Haiku 5.5 row.
+ */
+const LONG_CONTEXT_PRICING = {
+    "claude-haiku-5-5": price(0.5, 2.5, 0.05),
+};
 /**
  * Fast-mode prices (the page's "Fast mode pricing" section): premium
  * input/output with the prompt-caching multipliers applied ON TOP (so a 5m
@@ -175,6 +189,27 @@ export function listedPrice(model, speed) {
     const table = useFastTable ? FAST_PRICING : PRICING;
     return Object.hasOwn(table, model) ? table[model] : null;
 }
+/** The prompt tokens of one reply: everything it sent in, cached or not. */
+export function promptTokensOf(tokens) {
+    return tokens.input + tokens.cacheRead + tokens.cacheWrite5m + tokens.cacheWrite1h;
+}
+/** DX-4867 - whether `model` has a second, long-prompt rate card (a model with one card never switches, whatever the prompt length). */
+export function hasLongPromptCard(model) {
+    return Object.hasOwn(LONG_CONTEXT_PRICING, model);
+}
+/**
+ * DX-4867 - price ONE reply (not a bucket of replies): like {@link priceUsage}, but a reply whose prompt is longer than
+ * {@link LONG_PROMPT_TOKENS} on a model with a long-prompt card is charged that card. A model with one rate card prices
+ * exactly as `priceUsage` does.
+ */
+export function priceReplyUsage(model, speed, tokens) {
+    const long = promptTokensOf(tokens) > LONG_PROMPT_TOKENS && Object.hasOwn(LONG_CONTEXT_PRICING, model);
+    if (!long)
+        return priceUsage(model, speed, tokens);
+    if (speed === FAST_SPEED)
+        return { kind: "unpriced", model, speed };
+    return { kind: "priced", costUsd: costOfTokens(tokens, LONG_CONTEXT_PRICING[model]) };
+}
 /**
  * DX-4595 - dollars for `tokens` at `price` (USD per million tokens): each bucket with its own price. The ONE copy of the formula: `priceUsage`
  * charges the list price with it, and spend pacing charges an account's own (editable) prices with it.

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@thehammer/danx-session-telemetry",
-  "version": "0.1.47",
+  "version": "0.1.48",
   "description": "Shared Claude Code transcript collector (DX-2962) — incremental JSONL tailing for the danxbot worker and the danx-dashboard-mcp MCP server's relay, so both run identical collector code.",
   "license": "MIT",
   "type": "module",