@spendgraph/workflows 0.3.0 → 0.3.2

package/README.md

@@ -96,7 +96,9 @@ and win where both are set.
 **`params` does not follow a model it was not written for.** A stage naming its
 own `model` gets its own `params` or none: Haiku takes `temperature`, Sonnet 5
 rejects it outright, and one shared request body fails on whichever it was not
-written for. A stage overriding only `params` still merges over `stage`.
+written for. A stage overriding only `params` still merges over `stage` — except
+`answer`, which escalates through tiers, so its `params` reach it only alongside
+a `model`.
 
 ## What a run cost
 

package/dist/noesis/execute/index.d.ts

@@ -2,4 +2,4 @@ export { DEFAULT_TOOLS, execute } from "./execute.js";
 export { order } from "./order.js";
 export type { ExecuteOptions, Execution, SubAnswer, Waves, Wiring } from "./types.js";
 export type { ToolSpend, ToolUsage } from "./spend.js";
-export { addToolSpend, NO_TOOL_SPEND, readToolUsage, toolSpendOf } from "./spend.js";
+export { addToolSpend, NO_TOOL_SPEND, readToolUsage, recorded, toolSpendOf } from "./spend.js";

package/dist/noesis/execute/index.js

@@ -1,3 +1,3 @@
 export { DEFAULT_TOOLS, execute } from "./execute.js";
 export { order } from "./order.js";
-export { addToolSpend, NO_TOOL_SPEND, readToolUsage, toolSpendOf } from "./spend.js";
+export { addToolSpend, NO_TOOL_SPEND, readToolUsage, recorded, toolSpendOf } from "./spend.js";

package/dist/noesis/execute/spend.d.ts

@@ -1,7 +1,8 @@
-import type { SubAnswer, Wiring } from "./types.js";
+import type { Execution, SubAnswer, Wiring } from "./types.js";
 import type { SubQuestionTool } from "../producers/decompose/index.js";
 /** What one tool reported it cost, keyed by the tool that reported it. */
 export interface ToolSpend {
+    /** Tool calls that ran, whether or not the tool priced itself. */
     calls: number;
     /** Summed where a tool priced itself. Undefined where none did. */
     costMicros?: number;
@@ -34,3 +35,11 @@ export declare function readToolUsage(result: unknown): Omit<ToolUsage, "calls">
 export declare function toolSpendOf(answers: SubAnswer[], wiring?: Partial<Record<SubQuestionTool, Wiring>>): ToolSpend;
 /** Two executions' tool spend, added. A retry does not refund the first attempt. */
 export declare function addToolSpend(a: ToolSpend, b: ToolSpend): ToolSpend;
+/**
+ * The execution as the ledger should keep it: everything but the billing.
+ *
+ * Stage 4's artifact is serialized into every later stage's prompt and into the
+ * answer's evidence, so what a tool cost would ride along as citable material
+ * and be paid for again in tokens at each stage that reads it.
+ */
+export declare function recorded(execution: Execution): Omit<Execution, "spend">;

package/dist/noesis/execute/spend.js

@@ -26,10 +26,11 @@ export function toolSpendOf(answers, wiring = {}) {
     let calls = 0;
     let costMicros;
     for (const answer of answers) {
+        if (answer.status === "answered")
+            calls++;
         const one = readToolUsage(answer.result);
         if (!one)
             continue;
-        calls++;
         const named = wiring[answer.tool]?.name ?? answer.tool;
         const prior = byTool[named] ?? {
             calls: 0,
@@ -77,3 +78,10 @@ export function addToolSpend(a, b) {
         : undefined;
     return { calls: a.calls + b.calls, byTool, ...(costMicros !== undefined ? { costMicros } : {}) };
 }
+export function recorded(execution) {
+    return {
+        answers: execution.answers,
+        unresolved: execution.unresolved,
+        toolMs: execution.toolMs,
+    };
+}

package/dist/noesis/index.d.ts

@@ -6,7 +6,8 @@ export { assemble, render as renderDraft } from "./draft/index.js";
 export type { Emit, EventBase, NoesisEvent, NoesisEventInput } from "./events.js";
 export { emitter } from "./events.js";
 export type { ExecuteOptions, Execution, SubAnswer, Waves, Wiring, } from "./execute/index.js";
-export { DEFAULT_TOOLS, execute, order } from "./execute/index.js";
+export type { ToolSpend, ToolUsage } from "./execute/index.js";
+export { addToolSpend, DEFAULT_TOOLS, execute, NO_TOOL_SPEND, order, readToolUsage, recorded, toolSpendOf, } from "./execute/index.js";
 export type { ClosureChecklist, ClosureChecklistInput, ClosureChecklistOptions, ClosureChecklistResult, ClosureGate, } from "./gates/closure-checklist/index.js";
 export { invoke as closureChecklist } from "./gates/closure-checklist/index.js";
 export type { CqotGate, CqotGateInput, CqotGateOptions, CqotGateResult, Gate, GateQuestion, GateVerdict, } from "./gates/cqot-gate/index.js";

package/dist/noesis/index.js

@@ -2,7 +2,7 @@ export { stageCited, unattributed } from "./attribution.js";
 export { asked } from "./context.js";
 export { assemble, render as renderDraft } from "./draft/index.js";
 export { emitter } from "./events.js";
-export { DEFAULT_TOOLS, execute, order } from "./execute/index.js";
+export { addToolSpend, DEFAULT_TOOLS, execute, NO_TOOL_SPEND, order, readToolUsage, recorded, toolSpendOf, } from "./execute/index.js";
 export { invoke as closureChecklist } from "./gates/closure-checklist/index.js";
 export { invoke as cqotGate } from "./gates/cqot-gate/index.js";
 export { invoke as premortem } from "./gates/premortem/index.js";

package/dist/noesis/run/ledger.d.ts

@@ -51,8 +51,15 @@ export declare const STAGE_KEYS: {
     readonly "6C": "answer";
     readonly R: "root_cause";
 };
-/** A stage by the key a config names it under. */
-export type StageKey = (typeof STAGE_KEYS)[StageId];
+/** Stages that dispatch or pair rather than call a model. */
+type CodeOnly = "execution" | "verification" | "discriminating_test";
+/**
+ * A stage by the key a config names it under.
+ *
+ * The code-only stages are excluded: naming a model for the tool dispatcher or
+ * the blind-check pairing would typecheck and do nothing.
+ */
+export type StageKey = Exclude<(typeof STAGE_KEYS)[StageId], CodeOnly>;
 /** What one stage left behind, and what it cost. */
 export interface Entry extends Usage {
     stage: StageId;
@@ -156,3 +163,4 @@ export interface Ledger {
  * failing once each are the same amount of a run going nowhere.
  */
 export declare function newLedger(maxRetries?: number, emit?: Emit): Ledger;
+export {};

package/dist/noesis/run/options.d.ts

@@ -7,5 +7,18 @@ import { type StageId } from "./ledger.js";
  * rather than inherited when the stage names a model of its own.
  */
 export declare function stageOptionsFor(opts: NoesisOptions, id: StageId): StageOptions;
-/** The answer's tiers, defaulting to whatever `stages.answer` named. */
+/**
+ * Triage's options: `stages.triage` under `opts.triage`.
+ *
+ * `stage` is deliberately not a base here — it is the model for the stages
+ * between triage and the answer, and folding it in would send triage a body
+ * written for a different model.
+ */
+export declare function triageOptions(opts: NoesisOptions): StageOptions;
+/**
+ * The answer's tiers, defaulting to whatever `stages.answer` named.
+ *
+ * `params` reaches the answer only alongside a `model`, because the answer
+ * escalates through tiers and a tier carries its own params.
+ */
 export declare function answerOptions(opts: NoesisOptions): AnswerOptions;

package/dist/noesis/run/options.js

@@ -9,6 +9,14 @@ export function stageOptionsFor(opts, id) {
     const rest = Object.fromEntries(Object.entries({ ...base, ...per }).filter(([field]) => field !== "params"));
     return params && Object.keys(params).length > 0 ? { ...rest, params } : rest;
 }
+export function triageOptions(opts) {
+    const per = opts.stages?.[STAGE_KEYS["0"]] ?? {};
+    const own = opts.triage ?? {};
+    const ownModel = own.model !== undefined && own.model !== per.model;
+    const params = ownModel ? own.params : { ...per.params, ...own.params };
+    const rest = Object.fromEntries(Object.entries({ ...per, ...own }).filter(([field]) => field !== "params"));
+    return params && Object.keys(params).length > 0 ? { ...rest, params } : rest;
+}
 export function answerOptions(opts) {
     const base = opts.answer ?? {};
     const sixC = opts.stages?.[STAGE_KEYS["6C"]];

package/dist/noesis/run/produce.js

@@ -8,6 +8,7 @@ import { NoToolsError } from "./errors.js";
 import { plan } from "./plan.js";
 import { stageOptionsFor } from "./options.js";
 import { NO_TOOL_SPEND } from "../execute/spend.js";
+import { recorded } from "../execute/index.js";
 import { recordToolSpend } from "./state.js";
 export const soFar = (state, question) => JSON.stringify({ question, ...digest(state.ledger.artifacts()) }, null, 2);
 export async function frame(client, llm, question, opts, state) {
@@ -37,7 +38,7 @@ export async function work(client, llm, question, opts, state, split) {
     const execution = await execute(opts.tools, split.sub_questions, { emit: state.emit });
     recordToolSpend(state, execution.spend);
     state.ledger.record("4", {
-        data: execution,
+        data: recorded(execution),
         rolloutIds: [],
         model: "",
         inputTokens: 0,

package/dist/noesis/run/stages/classify.js

@@ -1,9 +1,9 @@
 import { routeFor } from "../../route.js";
 import { invoke as triage } from "../../router/triage/index.js";
 import { nextTask } from "../state.js";
-import { stageOptionsFor } from "../options.js";
+import { triageOptions } from "../options.js";
 export const classifyStage = (client, llm, question, opts, state) => async () => {
-    const stage = await triage(client, llm, { ...opts.triage, currentTask: state.currentTask, question }, { ...stageOptionsFor(opts, "0"), ...opts.triage });
+    const stage = await triage(client, llm, { ...opts.triage, currentTask: state.currentTask, question }, triageOptions(opts));
     state.estimates = state.ledger.record("0", { ...stage, data: stage.estimates });
     state.routing = routeFor(stage.estimates);
     state.currentTask = nextTask(stage.estimates, state.currentTask);

package/dist/noesis/run/stages/direct.js

@@ -1,6 +1,7 @@
 import { answer } from "../../producers/toulmin/index.js";
+import { answerOptions } from "../options.js";
 export const directStage = (client, llm, question, opts, state) => async () => {
-    const answered = await answer(client, llm, { question, ...state.ledger.artifacts() }, opts.answer ?? {});
+    const answered = await answer(client, llm, { question, ...state.ledger.artifacts() }, answerOptions(opts));
     state.answered = answered;
     state.text = state.ledger.record("6C", { ...answered, data: answered.answer }).claim;
     return state.text;

package/dist/noesis/run/stages/retrieve.js

@@ -1,6 +1,9 @@
 import { execute } from "../../execute/index.js";
 import { answer } from "../../producers/toulmin/index.js";
 import { NoToolsError } from "../errors.js";
+import { answerOptions } from "../options.js";
+import { recorded } from "../../execute/index.js";
+import { recordToolSpend } from "../state.js";
 const DEFAULT_TOOL = "deeprecall";
 const lookup = (question, tool) => ({
     id: "q",
@@ -15,15 +18,16 @@ export const retrieveStage = (client, llm, question, opts, state) => async () =>
     if (!opts.tools)
         throw new NoToolsError();
     const execution = await execute(opts.tools, [lookup(question, opts.retrieve?.tool ?? DEFAULT_TOOL)], { emit: state.emit });
+    recordToolSpend(state, execution.spend);
     state.ledger.record("4", {
-        data: execution,
+        data: recorded(execution),
         rolloutIds: [],
         model: "",
         inputTokens: 0,
         outputTokens: 0,
         latencyMs: execution.toolMs,
     });
-    const answered = await answer(client, llm, { question, ...state.ledger.artifacts() }, opts.answer ?? {});
+    const answered = await answer(client, llm, { question, ...state.ledger.artifacts() }, answerOptions(opts));
     state.answered = answered;
     state.text = state.ledger.record("6C", { ...answered, data: answered.answer }).claim;
     return state.text;

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@spendgraph/workflows",
-  "version": "0.3.0",
+  "version": "0.3.2",
   "description": "Ready-made workflows assembled from the spendgraph packages.",
   "license": "MIT",
   "repository": {
@@ -33,10 +33,10 @@
     "README.md"
   ],
   "dependencies": {
-    "@spendgraph/harness": "^0.3.0",
-    "@spendgraph/llms": "^0.3.0",
-    "@spendgraph/prompt": "^0.3.0",
-    "@spendgraph/tools": "^0.3.0"
+    "@spendgraph/harness": "^0.3.2",
+    "@spendgraph/llms": "^0.3.2",
+    "@spendgraph/prompt": "^0.3.2",
+    "@spendgraph/tools": "^0.3.2"
   },
   "scripts": {
     "build": "tsc -p tsconfig.json --emitDeclarationOnly && tsc -p tsconfig.json --declaration false --removeComments",
@@ -47,7 +47,7 @@
   },
   "devDependencies": {
     "@locusgraph/client": "^0.8.1",
-    "@spendgraph/evals": "^0.3.0",
+    "@spendgraph/evals": "^0.3.2",
     "typescript": "^5"
   },
   "engines": {