stageflow 0.40.0 → 0.42.0

package/dist/config/loadEval.js

@@ -0,0 +1,1085 @@
+import { readFile } from "node:fs/promises";
+import path from "node:path";
+import { parse as parseYaml } from "yaml";
+import { AGENT_ID_PATTERN, isReservedAgentId } from "../agents/agentId.js";
+import { canonicalJson, hash6, sha256Hex } from "../evals/canonical.js";
+import { loadFailure, loadSuccess } from "./loadOutcome.js";
+import { yamlParsePosition } from "./yamlParsePosition.js";
+export const EVAL_DEFAULT_REPEATS = 3;
+export const EVAL_DEFAULT_PASS_THRESHOLD = 0.8;
+export const EVAL_DEFAULT_MAX_COST_USD = 2;
+export const EVAL_DEFAULT_CONCURRENCY = 4;
+export const EVAL_AGENT_DEFAULT_CONCURRENCY = 2;
+export const EVAL_AGENT_DEFAULT_MAX_COST_USD = 1;
+export const EVAL_CONVERSATION_DEFAULTS = {
+    mode: "script",
+    maxTurns: 8,
+    turnTimeoutMs: 120_000,
+    tools: "read_only",
+    approvals: "record_only",
+    readOnlyTools: [],
+    allowRealTools: false,
+};
+export function resolveConversation(conversation) {
+    const d = EVAL_CONVERSATION_DEFAULTS;
+    return {
+        mode: conversation?.mode ?? d.mode,
+        maxTurns: conversation?.max_turns ?? d.maxTurns,
+        turnTimeoutMs: conversation?.turn_timeout_ms ?? d.turnTimeoutMs,
+        tools: conversation?.tools ?? d.tools,
+        approvals: conversation?.approvals ?? d.approvals,
+        readOnlyTools: [...(conversation?.read_only_tools ?? d.readOnlyTools)],
+        allowRealTools: conversation?.allow_real_tools ?? d.allowRealTools,
+    };
+}
+const ID_PATTERN = /^[a-z0-9][a-z0-9-]{0,62}$/;
+const GRADER_TYPES = [
+    "schema",
+    "field",
+    "reference",
+    "rubric",
+    "code",
+    "tool_calls",
+    "conversation_check",
+    "goal_reached",
+];
+const FIELD_OPS = [
+    "exists",
+    "not_empty",
+    "equals",
+    "not_equals",
+    "contains",
+    "matches",
+    "gte",
+    "lte",
+    "between",
+    "in",
+    "length_gte",
+    "length_lte",
+];
+const FIELD_PATH_PATTERN = /^[^.[\]\s]+(?:\.[^.[\]\s]+|\[(?:\d+|\*)\])*$/;
+const TOP_KEYS = [
+    "version",
+    "id",
+    "name",
+    "description",
+    "type",
+    "target",
+    "dataset",
+    "task",
+    "defaults",
+    "conversation",
+    "graders",
+    "budget",
+];
+const TARGET_KEYS = ["kind", "pipeline", "stage", "file"];
+const AGENT_TARGET_KEYS = ["kind", "agent"];
+const CONVERSATION_KEYS = [
+    "mode",
+    "max_turns",
+    "turn_timeout_ms",
+    "tools",
+    "approvals",
+    "read_only_tools",
+    "allow_real_tools",
+];
+const READ_ONLY_TOOL_PATTERN = /^[A-Za-z0-9_.*:-]{1,128}$/;
+const TASK_KEYS = ["goal", "context", "constraints"];
+const DEFAULTS_KEYS = [
+    "repeats",
+    "judge",
+    "pass_threshold",
+    "max_cost_usd",
+    "concurrency",
+    "model",
+    "user_model",
+];
+const BUDGET_KEYS = ["latency_ms", "cost_usd"];
+const GRADER_COMMON_KEYS = ["id", "type", "description", "weight", "must_pass"];
+const GRADER_TYPE_KEYS = {
+    schema: [],
+    field: ["checks"],
+    reference: ["mode", "path", "expected", "pass_at"],
+    rubric: ["rubric", "scale", "pass_at", "judge", "examples"],
+    code: ["code"],
+    tool_calls: ["must_call", "must_not_call", "must_ask_first", "max_calls", "args"],
+    conversation_check: ["max_turns", "max_agent_questions_per_turn", "final_reply"],
+    goal_reached: ["judge", "guidance"],
+};
+const TOOL_ARG_KEYS = ["tool", "path", "op", "value"];
+const FINAL_REPLY_KEYS = ["op", "value"];
+const CHECK_KEYS = ["path", "op", "value"];
+function isPlainObject(value) {
+    return typeof value === "object" && value !== null && !Array.isArray(value);
+}
+function evalIssue(code, message) {
+    return { code, message, category: "eval" };
+}
+export function evalFileStem(filePath) {
+    const base = path.basename(filePath);
+    const match = /^(.*)\.eval\.ya?ml$/.exec(base);
+    if (match)
+        return match[1] ?? base;
+    return base.slice(0, base.length - path.extname(base).length);
+}
+function isInside(root, candidate) {
+    const rel = path.relative(root, candidate);
+    return rel === "" || (!rel.startsWith("..") && !path.isAbsolute(rel));
+}
+/** Paths in an eval file resolve against the eval file's directory (P6). */
+export function resolveEvalPaths(file, evalAbsPath) {
+    const dir = path.dirname(evalAbsPath);
+    const stage = file.target.kind === "stage" ? file.target : undefined;
+    return {
+        ...(stage?.pipeline !== undefined ? { pipelinePath: path.resolve(dir, stage.pipeline) } : {}),
+        ...(stage?.file !== undefined ? { stageFile: path.resolve(dir, stage.file) } : {}),
+        datasetPath: path.resolve(dir, file.dataset),
+    };
+}
+class Collector {
+    issues = [];
+    add(code, message) {
+        this.issues.push(evalIssue(code, message));
+    }
+    unknownKeys(obj, allowed, where) {
+        for (const key of Object.keys(obj)) {
+            if (!allowed.includes(key)) {
+                this.add("eval.unknown_key", `Unknown key "${key}" in ${where}`);
+            }
+        }
+    }
+}
+function optionalString(c, obj, key, where, limits = {}) {
+    const value = obj[key];
+    if (value === undefined)
+        return undefined;
+    if (typeof value !== "string") {
+        c.add("eval.invalid_field", `${where} "${key}" must be a string`);
+        return undefined;
+    }
+    const min = limits.min ?? 0;
+    if (value.length < min || (limits.max !== undefined && value.length > limits.max)) {
+        const range = limits.max !== undefined ? `${min}-${limits.max} characters` : `at least ${min} character(s)`;
+        c.add("eval.invalid_field", `${where} "${key}" must be ${range}`);
+        return undefined;
+    }
+    return value;
+}
+function optionalNumber(c, obj, key, where, rule, ok, code = "eval.invalid_field") {
+    const value = obj[key];
+    if (value === undefined)
+        return undefined;
+    if (typeof value !== "number" || !Number.isFinite(value) || !ok(value)) {
+        c.add(code, `${where} "${key}" must be ${rule}`);
+        return undefined;
+    }
+    return value;
+}
+function isIntegerIn(n, min, max) {
+    return Number.isInteger(n) && n >= min && n <= max;
+}
+function parseTarget(c, raw) {
+    if (raw === undefined) {
+        c.add("eval.invalid_field", `Missing required key "target"`);
+        return undefined;
+    }
+    if (!isPlainObject(raw)) {
+        c.add("eval.invalid_field", `"target" must be a mapping`);
+        return undefined;
+    }
+    if (raw.kind === undefined) {
+        c.add("eval.invalid_field", `target "kind" is required ("stage" or "agent")`);
+        return undefined;
+    }
+    if (raw.kind === "agent")
+        return parseAgentTarget(c, raw);
+    if (raw.kind !== "stage") {
+        c.add("eval.target_kind_unsupported", `target kind ${JSON.stringify(raw.kind)} is not supported (only "stage" and "agent")`);
+        return undefined;
+    }
+    if (raw.agent !== undefined) {
+        c.add("eval.target_invalid", `target "agent" is only valid with kind: agent`);
+        return undefined;
+    }
+    c.unknownKeys(raw, TARGET_KEYS, "target");
+    const pipeline = raw.pipeline;
+    const stage = raw.stage;
+    const file = raw.file;
+    for (const [key, value] of [
+        ["pipeline", pipeline],
+        ["stage", stage],
+        ["file", file],
+    ]) {
+        if (value !== undefined && (typeof value !== "string" || value.trim() === "")) {
+            c.add("eval.invalid_field", `target "${key}" must be a non-empty string`);
+            return undefined;
+        }
+    }
+    const pipelineForm = pipeline !== undefined || stage !== undefined;
+    const fileForm = file !== undefined;
+    if (pipelineForm === fileForm || (pipelineForm && (pipeline === undefined || stage === undefined))) {
+        c.add("eval.target_invalid", "target must be exactly one of: { kind: stage, pipeline, stage } or { kind: stage, file }");
+        return undefined;
+    }
+    return {
+        kind: "stage",
+        ...(pipeline !== undefined ? { pipeline: pipeline } : {}),
+        ...(stage !== undefined ? { stage: stage } : {}),
+        ...(file !== undefined ? { file: file } : {}),
+    };
+}
+function parseAgentTarget(c, raw) {
+    if (raw.revision !== undefined) {
+        c.add("eval.target_invalid", `target "revision" is recorded by the run, not authored`);
+        return undefined;
+    }
+    for (const key of ["pipeline", "stage", "file"]) {
+        if (raw[key] !== undefined) {
+            c.add("eval.target_invalid", `target "${key}" is only valid with kind: stage`);
+            return undefined;
+        }
+    }
+    c.unknownKeys(raw, AGENT_TARGET_KEYS, "target");
+    const agent = raw.agent;
+    if (typeof agent !== "string" || !AGENT_ID_PATTERN.test(agent)) {
+        c.add("eval.target_invalid", `target "agent" must be an agent id matching ${AGENT_ID_PATTERN.source}`);
+        return undefined;
+    }
+    if (isReservedAgentId(agent)) {
+        c.add("eval.target_invalid", `target agent id "${agent}" is reserved`);
+        return undefined;
+    }
+    return { kind: "agent", agent };
+}
+function parseTask(c, raw) {
+    if (raw === undefined)
+        return undefined;
+    if (!isPlainObject(raw)) {
+        c.add("eval.invalid_field", `"task" must be a mapping`);
+        return undefined;
+    }
+    c.unknownKeys(raw, TASK_KEYS, "task");
+    const out = {};
+    for (const key of TASK_KEYS) {
+        const value = optionalString(c, raw, key, "task");
+        if (value !== undefined)
+            out[key] = value;
+    }
+    return out;
+}
+function parseDefaults(c, raw, kind) {
+    if (raw === undefined)
+        return undefined;
+    if (!isPlainObject(raw)) {
+        c.add("eval.invalid_field", `"defaults" must be a mapping`);
+        return undefined;
+    }
+    c.unknownKeys(raw, DEFAULTS_KEYS, "defaults");
+    const out = {};
+    const repeats = optionalNumber(c, raw, "repeats", "defaults", "an integer from 1 to 10", (n) => isIntegerIn(n, 1, 10));
+    if (repeats !== undefined)
+        out.repeats = repeats;
+    const judge = optionalString(c, raw, "judge", "defaults", { min: 1 });
+    if (judge !== undefined)
+        out.judge = judge;
+    const passThreshold = optionalNumber(c, raw, "pass_threshold", "defaults", "a number greater than 0 and at most 1", (n) => n > 0 && n <= 1);
+    if (passThreshold !== undefined)
+        out.pass_threshold = passThreshold;
+    const maxCost = optionalNumber(c, raw, "max_cost_usd", "defaults", "a number greater than 0", (n) => n > 0);
+    if (maxCost !== undefined)
+        out.max_cost_usd = maxCost;
+    const concurrency = optionalNumber(c, raw, "concurrency", "defaults", "an integer from 1 to 8", (n) => isIntegerIn(n, 1, 8));
+    if (concurrency !== undefined)
+        out.concurrency = concurrency;
+    const model = optionalString(c, raw, "model", "defaults", { min: 1 });
+    if (model !== undefined)
+        out.model = model;
+    if (raw.user_model !== undefined) {
+        if (kind === "stage") {
+            c.add("eval.conversation_not_applicable", `defaults "user_model" is only valid with an agent target`);
+        }
+        else {
+            const userModel = optionalString(c, raw, "user_model", "defaults", { min: 1 });
+            if (userModel !== undefined)
+                out.user_model = userModel;
+        }
+    }
+    return out;
+}
+function parseConversation(c, raw, kind) {
+    if (raw === undefined)
+        return undefined;
+    if (kind === "stage") {
+        c.add("eval.conversation_not_applicable", `"conversation" is only valid with an agent target`);
+        return undefined;
+    }
+    const bad = (key, rule) => c.add("eval.conversation_invalid", `conversation "${key}" must be ${rule}`);
+    if (!isPlainObject(raw)) {
+        c.add("eval.conversation_invalid", `"conversation" must be a mapping`);
+        return undefined;
+    }
+    c.unknownKeys(raw, CONVERSATION_KEYS, "conversation");
+    const out = {};
+    const oneOf = (key, allowed) => {
+        const value = raw[key];
+        if (value === undefined)
+            return undefined;
+        if (typeof value === "string" && allowed.includes(value))
+            return value;
+        bad(key, `one of: ${allowed.join(", ")}`);
+        return undefined;
+    };
+    const mode = oneOf("mode", ["script", "simulated"]);
+    if (mode !== undefined)
+        out.mode = mode;
+    const maxTurns = optionalNumber(c, raw, "max_turns", "conversation", "an integer from 1 to 30", (n) => isIntegerIn(n, 1, 30), "eval.conversation_invalid");
+    if (maxTurns !== undefined)
+        out.max_turns = maxTurns;
+    const timeout = optionalNumber(c, raw, "turn_timeout_ms", "conversation", "an integer from 5000 to 600000", (n) => isIntegerIn(n, 5000, 600_000), "eval.conversation_invalid");
+    if (timeout !== undefined)
+        out.turn_timeout_ms = timeout;
+    const tools = oneOf("tools", ["real", "read_only", "mocked"]);
+    if (tools !== undefined)
+        out.tools = tools;
+    const approvals = oneOf("approvals", ["deny", "auto_approve_read", "record_only"]);
+    if (approvals !== undefined)
+        out.approvals = approvals;
+    if (raw.read_only_tools !== undefined) {
+        const list = raw.read_only_tools;
+        if (Array.isArray(list) &&
+            list.length <= 50 &&
+            list.every((item) => typeof item === "string" && READ_ONLY_TOOL_PATTERN.test(item))) {
+            out.read_only_tools = list;
+        }
+        else {
+            bad("read_only_tools", "an array of at most 50 tool grant ids");
+        }
+    }
+    if (raw.allow_real_tools !== undefined) {
+        if (typeof raw.allow_real_tools !== "boolean") {
+            bad("allow_real_tools", "a boolean");
+        }
+        else if (raw.allow_real_tools && (tools ?? "read_only") !== "real") {
+            c.add("eval.conversation_invalid", `conversation "allow_real_tools" is only valid with tools: real`);
+        }
+        else {
+            out.allow_real_tools = raw.allow_real_tools;
+        }
+    }
+    return out;
+}
+function parseBudget(c, raw) {
+    if (raw === undefined)
+        return undefined;
+    if (!isPlainObject(raw)) {
+        c.add("eval.invalid_field", `"budget" must be a mapping`);
+        return undefined;
+    }
+    c.unknownKeys(raw, BUDGET_KEYS, "budget");
+    const out = {};
+    const latency = optionalNumber(c, raw, "latency_ms", "budget", "a positive number", (n) => n > 0);
+    if (latency !== undefined)
+        out.latency_ms = latency;
+    const cost = optionalNumber(c, raw, "cost_usd", "budget", "a positive number", (n) => n > 0);
+    if (cost !== undefined)
+        out.cost_usd = cost;
+    return out;
+}
+function checkValueProblem(op, value, present) {
+    if (op === "exists" || op === "not_empty") {
+        return present ? `op "${op}" takes no value` : undefined;
+    }
+    if (!present)
+        return `op "${op}" requires a value`;
+    switch (op) {
+        case "equals":
+        case "not_equals":
+        case "contains":
+            return undefined;
+        case "matches": {
+            if (typeof value !== "string")
+                return `op "matches" value must be a regular expression string`;
+            try {
+                new RegExp(value);
+            }
+            catch {
+                return `op "matches" value is not a valid regular expression`;
+            }
+            return undefined;
+        }
+        case "gte":
+        case "lte":
+            return typeof value === "number" && Number.isFinite(value)
+                ? undefined
+                : `op "${op}" value must be a number`;
+        case "between":
+            return Array.isArray(value) &&
+                value.length === 2 &&
+                value.every((v) => typeof v === "number" && Number.isFinite(v)) &&
+                value[0] <= value[1]
+                ? undefined
+                : `op "between" value must be [min, max] with min <= max`;
+        case "in":
+            return Array.isArray(value) ? undefined : `op "in" value must be an array`;
+        case "length_gte":
+        case "length_lte":
+            return typeof value === "number" && Number.isInteger(value) && value >= 0
+                ? undefined
+                : `op "${op}" value must be a non-negative integer`;
+    }
+}
+function parseChecks(c, raw, where) {
+    if (!Array.isArray(raw) || raw.length === 0) {
+        c.add("eval.grader_invalid", `${where} "checks" must be a non-empty array`);
+        return undefined;
+    }
+    const checks = [];
+    let ok = true;
+    raw.forEach((item, index) => {
+        const at = `${where} checks[${index}]`;
+        if (!isPlainObject(item)) {
+            c.add("eval.grader_invalid", `${at} must be a mapping`);
+            ok = false;
+            return;
+        }
+        c.unknownKeys(item, CHECK_KEYS, at);
+        if (typeof item.path !== "string" || !FIELD_PATH_PATTERN.test(item.path)) {
+            c.add("eval.grader_invalid", `${at} "path" must be dot-separated keys with optional [n] or [*] (for example payload.findings[*].line)`);
+            ok = false;
+            return;
+        }
+        if (typeof item.op !== "string" || !FIELD_OPS.includes(item.op)) {
+            c.add("eval.grader_invalid", `${at} "op" must be one of: ${FIELD_OPS.join(", ")}`);
+            ok = false;
+            return;
+        }
+        const op = item.op;
+        const problem = checkValueProblem(op, item.value, "value" in item);
+        if (problem) {
+            c.add("eval.grader_invalid", `${at} ${problem}`);
+            ok = false;
+            return;
+        }
+        checks.push({ path: item.path, op, ...("value" in item ? { value: item.value } : {}) });
+    });
+    return ok ? checks : undefined;
+}
+function parseExamples(c, raw, scale, where) {
+    if (!Array.isArray(raw) || raw.length > 6) {
+        c.add("eval.grader_invalid", `${where} "examples" must be an array of at most 6 items`);
+        return undefined;
+    }
+    const out = [];
+    let ok = true;
+    raw.forEach((item, index) => {
+        const at = `${where} examples[${index}]`;
+        if (!isPlainObject(item)) {
+            c.add("eval.grader_invalid", `${at} must be a mapping`);
+            ok = false;
+            return;
+        }
+        c.unknownKeys(item, [scale === "pass_fail" ? "verdict" : "score", "output", "reason"], at);
+        if (typeof item.output !== "string") {
+            c.add("eval.grader_invalid", `${at} "output" must be a string`);
+            ok = false;
+            return;
+        }
+        if (item.reason !== undefined && typeof item.reason !== "string") {
+            c.add("eval.grader_invalid", `${at} "reason" must be a string`);
+            ok = false;
+            return;
+        }
+        const reason = item.reason;
+        if (scale === "pass_fail") {
+            if (item.verdict !== "pass" && item.verdict !== "fail") {
+                c.add("eval.grader_invalid", `${at} "verdict" must be "pass" or "fail"`);
+                ok = false;
+                return;
+            }
+            out.push({ verdict: item.verdict, output: item.output, ...(reason !== undefined ? { reason } : {}) });
+            return;
+        }
+        if (typeof item.score !== "number" || !isIntegerIn(item.score, 1, 5)) {
+            c.add("eval.grader_invalid", `${at} "score" must be an integer from 1 to 5`);
+            ok = false;
+            return;
+        }
+        out.push({ score: item.score, output: item.output, ...(reason !== undefined ? { reason } : {}) });
+    });
+    return ok ? out : undefined;
+}
+function optionalToolList(c, raw, key, where) {
+    const value = raw[key];
+    if (value === undefined)
+        return undefined;
+    if (!Array.isArray(value) ||
+        value.length > 50 ||
+        !value.every((item) => typeof item === "string" && item.length >= 1 && item.length <= 128)) {
+        c.add("eval.grader_invalid", `${where} "${key}" must be an array of at most 50 tool names`);
+        return undefined;
+    }
+    return value;
+}
+function parseToolArgs(c, raw, where) {
+    if (!Array.isArray(raw) || raw.length > 20) {
+        c.add("eval.grader_invalid", `${where} "args" must be an array of at most 20 checks`);
+        return undefined;
+    }
+    const out = [];
+    let ok = true;
+    raw.forEach((item, index) => {
+        const at = `${where} args[${index}]`;
+        if (!isPlainObject(item)) {
+            c.add("eval.grader_invalid", `${at} must be a mapping`);
+            ok = false;
+            return;
+        }
+        c.unknownKeys(item, TOOL_ARG_KEYS, at);
+        if (typeof item.tool !== "string" || item.tool.length < 1 || item.tool.length > 128) {
+            c.add("eval.grader_invalid", `${at} "tool" must be a tool name`);
+            ok = false;
+            return;
+        }
+        if (typeof item.path !== "string" || !FIELD_PATH_PATTERN.test(item.path)) {
+            c.add("eval.grader_invalid", `${at} "path" must be dot-separated keys with optional [n] or [*]`);
+            ok = false;
+            return;
+        }
+        if (typeof item.op !== "string" || !FIELD_OPS.includes(item.op)) {
+            c.add("eval.grader_invalid", `${at} "op" must be one of: ${FIELD_OPS.join(", ")}`);
+            ok = false;
+            return;
+        }
+        const op = item.op;
+        const problem = checkValueProblem(op, item.value, "value" in item);
+        if (problem) {
+            c.add("eval.grader_invalid", `${at} ${problem}`);
+            ok = false;
+            return;
+        }
+        out.push({ tool: item.tool, path: item.path, op, ...("value" in item ? { value: item.value } : {}) });
+    });
+    return ok ? out : undefined;
+}
+function parseFinalReply(c, raw, where) {
+    if (!Array.isArray(raw) || raw.length === 0 || raw.length > 10) {
+        c.add("eval.grader_invalid", `${where} "final_reply" must be an array of 1-10 checks`);
+        return undefined;
+    }
+    const out = [];
+    let ok = true;
+    raw.forEach((item, index) => {
+        const at = `${where} final_reply[${index}]`;
+        if (!isPlainObject(item)) {
+            c.add("eval.grader_invalid", `${at} must be a mapping`);
+            ok = false;
+            return;
+        }
+        c.unknownKeys(item, FINAL_REPLY_KEYS, at);
+        if (typeof item.op !== "string" || !FIELD_OPS.includes(item.op)) {
+            c.add("eval.grader_invalid", `${at} "op" must be one of: ${FIELD_OPS.join(", ")}`);
+            ok = false;
+            return;
+        }
+        const op = item.op;
+        const problem = checkValueProblem(op, item.value, "value" in item);
+        if (problem) {
+            c.add("eval.grader_invalid", `${at} ${problem}`);
+            ok = false;
+            return;
+        }
+        out.push({ op, ...("value" in item ? { value: item.value } : {}) });
+    });
+    return ok ? out : undefined;
+}
+function parseGrader(c, raw, index, defaultJudge, kind) {
+    if (!isPlainObject(raw)) {
+        c.add("eval.grader_invalid", `graders[${index}] must be a mapping`);
+        return undefined;
+    }
+    const label = typeof raw.id === "string" ? `grader "${raw.id}"` : `graders[${index}]`;
+    const before = c.issues.length;
+    if (typeof raw.id !== "string" || !ID_PATTERN.test(raw.id)) {
+        c.add("eval.invalid_field", `graders[${index}] "id" must match ${ID_PATTERN.source}`);
+    }
+    const type = raw.type;
+    if (type === undefined) {
+        c.add("eval.grader_invalid", `${label} "type" is required`);
+        return undefined;
+    }
+    if (type === "command") {
+        c.add("eval.grader_type_unsupported", `${label} type "command" is not available in release 1 (supported: ${GRADER_TYPES.join(", ")})`);
+        return undefined;
+    }
+    if (typeof type !== "string" || !GRADER_TYPES.includes(type)) {
+        c.add("eval.grader_type_unsupported", `${label} type ${JSON.stringify(type)} is not supported (supported: ${GRADER_TYPES.join(", ")})`);
+        return undefined;
+    }
+    const graderType = type;
+    const scaleOnGoal = graderType === "goal_reached" && raw.scale !== undefined;
+    c.unknownKeys(scaleOnGoal ? Object.fromEntries(Object.entries(raw).filter(([key]) => key !== "scale")) : raw, [...GRADER_COMMON_KEYS, ...GRADER_TYPE_KEYS[graderType]], label);
+    if (scaleOnGoal) {
+        c.add("eval.grader_invalid", `${label} "scale" is not accepted: goal_reached is always pass_fail`);
+    }
+    const description = optionalString(c, raw, "description", label, { max: 200 });
+    const weight = optionalNumber(c, raw, "weight", label, "a number greater than 0", (n) => n > 0);
+    let mustPass = false;
+    if (raw.must_pass !== undefined) {
+        if (typeof raw.must_pass === "boolean")
+            mustPass = raw.must_pass;
+        else
+            c.add("eval.invalid_field", `${label} "must_pass" must be a boolean`);
+    }
+    const common = {
+        id: String(raw.id),
+        ...(description !== undefined ? { description } : {}),
+        ...(weight !== undefined ? { weight } : {}),
+        ...(raw.must_pass !== undefined && typeof raw.must_pass === "boolean"
+            ? { must_pass: raw.must_pass }
+            : {}),
+    };
+    const resolvedCommon = {
+        id: String(raw.id),
+        weight: weight ?? 1,
+        mustPass,
+        ...(description !== undefined ? { description } : {}),
+    };
+    let result;
+    switch (graderType) {
+        case "schema":
+            result = {
+                config: { ...common, type: "schema" },
+                resolved: { ...resolvedCommon, type: "schema" },
+            };
+            break;
+        case "field": {
+            const checks = parseChecks(c, raw.checks, label);
+            if (checks) {
+                result = {
+                    config: { ...common, type: "field", checks },
+                    resolved: { ...resolvedCommon, type: "field", checks },
+                };
+            }
+            break;
+        }
+        case "reference": {
+            const mode = raw.mode;
+            let ok = true;
+            if (mode !== "each_mentioned" && mode !== "equals") {
+                c.add("eval.grader_invalid", `${label} "mode" must be "each_mentioned" or "equals"`);
+                ok = false;
+            }
+            let pathValue = kind === "agent" ? "final_reply" : "payload";
+            if (raw.path !== undefined) {
+                if (typeof raw.path === "string" && FIELD_PATH_PATTERN.test(raw.path))
+                    pathValue = raw.path;
+                else {
+                    c.add("eval.grader_invalid", `${label} "path" must be a valid envelope path`);
+                    ok = false;
+                }
+            }
+            if (typeof raw.expected !== "string" || !FIELD_PATH_PATTERN.test(raw.expected)) {
+                c.add("eval.grader_invalid", `${label} "expected" must be a valid path into the case (for example expected.findings)`);
+                ok = false;
+            }
+            const passAt = optionalNumber(c, raw, "pass_at", label, "a number from 0 to 1", (n) => n >= 0 && n <= 1, "eval.grader_invalid");
+            if (raw.pass_at !== undefined && passAt === undefined)
+                ok = false;
+            if (ok) {
+                result = {
+                    config: {
+                        ...common,
+                        type: "reference",
+                        mode: mode,
+                        ...(raw.path !== undefined ? { path: pathValue } : {}),
+                        expected: raw.expected,
+                        ...(passAt !== undefined ? { pass_at: passAt } : {}),
+                    },
+                    resolved: {
+                        ...resolvedCommon,
+                        type: "reference",
+                        mode: mode,
+                        path: pathValue,
+                        expected: raw.expected,
+                        passAt: passAt ?? 1,
+                    },
+                };
+            }
+            break;
+        }
+        case "rubric": {
+            let ok = true;
+            if (typeof raw.rubric !== "string" || raw.rubric.length < 1 || raw.rubric.length > 4000) {
+                c.add("eval.grader_invalid", `${label} "rubric" must be a string of 1-4000 characters`);
+                ok = false;
+            }
+            let scale = "pass_fail";
+            if (raw.scale !== undefined) {
+                if (raw.scale === "pass_fail" || raw.scale === "1_to_5")
+                    scale = raw.scale;
+                else {
+                    c.add("eval.grader_invalid", `${label} "scale" must be "pass_fail" or "1_to_5"`);
+                    ok = false;
+                }
+            }
+            let passAt;
+            if (raw.pass_at !== undefined) {
+                if (scale === "pass_fail") {
+                    c.add("eval.grader_invalid", `${label} "pass_at" is only allowed with scale "1_to_5"`);
+                    ok = false;
+                }
+                else if (typeof raw.pass_at === "number" && isIntegerIn(raw.pass_at, 1, 5)) {
+                    passAt = raw.pass_at;
+                }
+                else {
+                    c.add("eval.grader_invalid", `${label} "pass_at" must be an integer from 1 to 5`);
+                    ok = false;
+                }
+            }
+            let judge;
+            if (raw.judge !== undefined) {
+                if (typeof raw.judge === "string" && raw.judge.length > 0)
+                    judge = raw.judge;
+                else {
+                    c.add("eval.grader_invalid", `${label} "judge" must be a model id string`);
+                    ok = false;
+                }
+            }
+            let examples;
+            if (raw.examples !== undefined) {
+                examples = parseExamples(c, raw.examples, scale, label);
+                if (examples === undefined)
+                    ok = false;
+            }
+            if (ok) {
+                result = {
+                    config: {
+                        ...common,
+                        type: "rubric",
+                        rubric: raw.rubric,
+                        ...(raw.scale !== undefined ? { scale } : {}),
+                        ...(passAt !== undefined ? { pass_at: passAt } : {}),
+                        ...(judge !== undefined ? { judge } : {}),
+                        ...(examples !== undefined ? { examples } : {}),
+                    },
+                    resolved: {
+                        ...resolvedCommon,
+                        type: "rubric",
+                        rubric: raw.rubric,
+                        scale,
+                        passAt: scale === "1_to_5" ? (passAt ?? 4) : 1,
+                        judge: judge ?? defaultJudge ?? "",
+                        examples: examples ?? [],
+                    },
+                };
+            }
+            break;
+        }
+        case "code": {
+            if (typeof raw.code !== "string" || raw.code.trim() === "") {
+                c.add("eval.grader_invalid", `${label} "code" must be a non-empty string`);
+            }
+            else {
+                result = {
+                    config: { ...common, type: "code", code: raw.code },
+                    resolved: { ...resolvedCommon, type: "code", code: raw.code },
+                };
+            }
+            break;
+        }
+        case "tool_calls": {
+            const mustCall = optionalToolList(c, raw, "must_call", label);
+            const mustNotCall = optionalToolList(c, raw, "must_not_call", label);
+            const mustAskFirst = optionalToolList(c, raw, "must_ask_first", label);
+            const maxCalls = optionalNumber(c, raw, "max_calls", label, "an integer from 0 to 200", (n) => isIntegerIn(n, 0, 200), "eval.grader_invalid");
+            const args = raw.args === undefined ? [] : parseToolArgs(c, raw.args, label);
+            if (args !== undefined) {
+                result = {
+                    config: {
+                        ...common,
+                        type: "tool_calls",
+                        ...(mustCall !== undefined ? { must_call: mustCall } : {}),
+                        ...(mustNotCall !== undefined ? { must_not_call: mustNotCall } : {}),
+                        ...(mustAskFirst !== undefined ? { must_ask_first: mustAskFirst } : {}),
+                        ...(maxCalls !== undefined ? { max_calls: maxCalls } : {}),
+                        ...(raw.args !== undefined ? { args } : {}),
+                    },
+                    resolved: {
+                        ...resolvedCommon,
+                        type: "tool_calls",
+                        ...(mustCall !== undefined ? { mustCall } : {}),
+                        ...(mustNotCall !== undefined ? { mustNotCall } : {}),
+                        ...(mustAskFirst !== undefined ? { mustAskFirst } : {}),
+                        ...(maxCalls !== undefined ? { maxCalls } : {}),
+                        args,
+                    },
+                };
+            }
+            break;
+        }
+        case "conversation_check": {
+            const maxTurns = optionalNumber(c, raw, "max_turns", label, "an integer from 1 to 100", (n) => isIntegerIn(n, 1, 100), "eval.grader_invalid");
+            const maxQuestions = optionalNumber(c, raw, "max_agent_questions_per_turn", label, "an integer from 0 to 10", (n) => isIntegerIn(n, 0, 10), "eval.grader_invalid");
+            const finalReply = raw.final_reply === undefined ? [] : parseFinalReply(c, raw.final_reply, label);
+            if (raw.max_turns === undefined && raw.max_agent_questions_per_turn === undefined && raw.final_reply === undefined) {
+                c.add("eval.grader_invalid", `${label} needs at least one of: max_turns, max_agent_questions_per_turn, final_reply`);
+            }
+            if (finalReply !== undefined) {
+                result = {
+                    config: {
+                        ...common,
+                        type: "conversation_check",
+                        ...(maxTurns !== undefined ? { max_turns: maxTurns } : {}),
+                        ...(maxQuestions !== undefined ? { max_agent_questions_per_turn: maxQuestions } : {}),
+                        ...(raw.final_reply !== undefined ? { final_reply: finalReply } : {}),
+                    },
+                    resolved: {
+                        ...resolvedCommon,
+                        type: "conversation_check",
+                        ...(maxTurns !== undefined ? { maxTurns } : {}),
+                        ...(maxQuestions !== undefined ? { maxAgentQuestionsPerTurn: maxQuestions } : {}),
+                        finalReply,
+                    },
+                };
+            }
+            break;
+        }
+        case "goal_reached": {
+            let ok = true;
+            let judge;
+            if (raw.judge !== undefined) {
+                if (typeof raw.judge === "string" && raw.judge.length > 0)
+                    judge = raw.judge;
+                else {
+                    c.add("eval.grader_invalid", `${label} "judge" must be a model id string`);
+                    ok = false;
+                }
+            }
+            let guidance;
+            if (raw.guidance !== undefined) {
+                if (typeof raw.guidance === "string" && raw.guidance.length <= 2000)
+                    guidance = raw.guidance;
+                else {
+                    c.add("eval.grader_invalid", `${label} "guidance" must be a string of at most 2000 characters`);
+                    ok = false;
+                }
+            }
+            if (ok) {
+                result = {
+                    config: {
+                        ...common,
+                        type: "goal_reached",
+                        ...(judge !== undefined ? { judge } : {}),
+                        ...(guidance !== undefined ? { guidance } : {}),
+                    },
+                    resolved: {
+                        ...resolvedCommon,
+                        type: "goal_reached",
+                        judge: judge ?? defaultJudge ?? "",
+                        ...(guidance !== undefined ? { guidance } : {}),
+                    },
+                };
+            }
+            break;
+        }
+    }
+    return c.issues.length > before ? undefined : result;
+}
+/** Grader hash object: the grader's file keys with defaults filled in; a rubric carries its effective judge. */
+function graderHashObject(grader) {
+    const base = {
+        id: grader.id,
+        type: grader.type,
+        description: grader.description,
+        weight: grader.weight,
+        must_pass: grader.mustPass,
+    };
+    switch (grader.type) {
+        case "schema":
+            return base;
+        case "field":
+            return { ...base, checks: grader.checks };
+        case "reference":
+            return {
+                ...base,
+                mode: grader.mode,
+                path: grader.path,
+                expected: grader.expected,
+                pass_at: grader.passAt,
+            };
+        case "rubric":
+            return {
+                ...base,
+                rubric: grader.rubric,
+                scale: grader.scale,
+                ...(grader.scale === "1_to_5" ? { pass_at: grader.passAt } : {}),
+                judge: grader.judge,
+                examples: grader.examples,
+            };
+        case "code":
+            return { ...base, code: grader.code };
+        case "tool_calls":
+            return {
+                ...base,
+                must_call: grader.mustCall,
+                must_not_call: grader.mustNotCall,
+                must_ask_first: grader.mustAskFirst,
+                max_calls: grader.maxCalls,
+                args: grader.args,
+            };
+        case "conversation_check":
+            return {
+                ...base,
+                max_turns: grader.maxTurns,
+                max_agent_questions_per_turn: grader.maxAgentQuestionsPerTurn,
+                final_reply: grader.finalReply,
+            };
+        case "goal_reached":
+            return { ...base, judge: grader.judge, guidance: grader.guidance };
+    }
+}
+export function graderHash(grader) {
+    return sha256Hex(canonicalJson(graderHashObject(grader)));
+}
+export function graderHashShort(grader) {
+    return hash6(grader.hash);
+}
+export function parseEvalOutcome(yamlText, filePath, projectRoot) {
+    let raw;
+    try {
+        raw = parseYaml(yamlText);
+    }
+    catch (err) {
+        return loadFailure([
+            {
+                ...evalIssue("eval.load_error", err instanceof Error ? err.message : String(err)),
+                ...yamlParsePosition(err),
+            },
+        ]);
+    }
+    if (!isPlainObject(raw)) {
+        return loadFailure([evalIssue("eval.not_object", "Eval file must be a YAML mapping")]);
+    }
+    const c = new Collector();
+    c.unknownKeys(raw, TOP_KEYS, "the eval file");
+    if (raw.version !== 1) {
+        c.add("eval.invalid_version", `"version" must be 1 (found ${JSON.stringify(raw.version) ?? "none"})`);
+    }
+    const absPath = path.resolve(filePath);
+    const stem = evalFileStem(absPath);
+    const id = raw.id;
+    if (typeof id !== "string" || !ID_PATTERN.test(id)) {
+        c.add("eval.invalid_id", `"id" must match ${ID_PATTERN.source}`);
+    }
+    else if (id !== stem) {
+        c.add("eval.id_filename_mismatch", `Eval id "${id}" does not match filename stem "${stem}"`);
+    }
+    const name = optionalString(c, raw, "name", "eval", { min: 1, max: 80 });
+    const description = optionalString(c, raw, "description", "eval", { max: 500 });
+    let type;
+    if (raw.type !== undefined) {
+        if (raw.type === "capability" || raw.type === "regression")
+            type = raw.type;
+        else
+            c.add("eval.invalid_field", `"type" must be "capability" or "regression"`);
+    }
+    const target = parseTarget(c, raw.target);
+    let dataset;
+    if (raw.dataset === undefined) {
+        c.add("eval.invalid_field", `Missing required key "dataset"`);
+    }
+    else if (typeof raw.dataset !== "string" || raw.dataset.trim() === "") {
+        c.add("eval.invalid_field", `"dataset" must be a non-empty path string`);
+    }
+    else {
+        dataset = raw.dataset;
+    }
+    const root = path.resolve(projectRoot);
+    const dir = path.dirname(absPath);
+    if (dataset !== undefined && !isInside(root, path.resolve(dir, dataset))) {
+        c.add("eval.dataset_unresolved", `Dataset path "${dataset}" resolves outside the project root`);
+    }
+    if (target?.kind === "stage") {
+        for (const [key, value] of [
+            ["pipeline", target.pipeline],
+            ["file", target.file],
+        ]) {
+            if (value !== undefined && !isInside(root, path.resolve(dir, value))) {
+                c.add("eval.target_unresolved", `Target ${key} "${value}" resolves outside the project root`);
+            }
+        }
+    }
+    let task;
+    if (target?.kind === "agent" && raw.task !== undefined) {
+        c.add("eval.task_not_applicable", `"task" is not accepted for an agent target (the scenario goal is input.goal)`);
+    }
+    else {
+        task = parseTask(c, raw.task);
+    }
+    const defaults = parseDefaults(c, raw.defaults, target?.kind);
+    const conversation = parseConversation(c, raw.conversation, target?.kind);
+    const budget = parseBudget(c, raw.budget);
+    const parsedGraders = [];
+    if (raw.graders === undefined) {
+        c.add("eval.invalid_field", `Missing required key "graders"`);
+    }
+    else if (!Array.isArray(raw.graders) || raw.graders.length < 1 || raw.graders.length > 20) {
+        c.add("eval.invalid_field", `"graders" must be an array of 1-20 graders`);
+    }
+    else {
+        const seen = new Set();
+        raw.graders.forEach((rawGrader, index) => {
+            const parsed = parseGrader(c, rawGrader, index, defaults?.judge, target?.kind);
+            if (isPlainObject(rawGrader) && typeof rawGrader.id === "string") {
+                if (seen.has(rawGrader.id)) {
+                    c.add("eval.grader_duplicate_id", `Duplicate grader id "${rawGrader.id}"`);
+                    return;
+                }
+                seen.add(rawGrader.id);
+            }
+            if (parsed)
+                parsedGraders.push(parsed);
+        });
+    }
+    if (c.issues.length > 0 || !target || dataset === undefined || typeof id !== "string") {
+        return loadFailure(c.issues);
+    }
+    const file = {
+        version: 1,
+        id,
+        ...(name !== undefined ? { name } : {}),
+        ...(description !== undefined ? { description } : {}),
+        ...(type !== undefined ? { type } : {}),
+        target,
+        dataset,
+        ...(task !== undefined ? { task } : {}),
+        ...(defaults !== undefined ? { defaults } : {}),
+        ...(conversation !== undefined ? { conversation } : {}),
+        graders: parsedGraders.map((g) => g.config),
+        ...(budget !== undefined ? { budget } : {}),
+    };
+    const graders = parsedGraders.map((g) => ({ ...g.resolved, hash: graderHash(g.resolved) }));
+    return loadSuccess({
+        path: absPath,
+        relPath: path.relative(root, absPath).split(path.sep).join("/"),
+        id,
+        fileHash: sha256Hex(Buffer.from(yamlText, "utf8")),
+        file,
+        graders,
+    });
+}
+export async function loadEvalOutcome(absPath, projectRoot) {
+    let bytes;
+    try {
+        bytes = await readFile(absPath);
+    }
+    catch (err) {
+        return loadFailure([
+            evalIssue("eval.load_error", err instanceof Error ? err.message : String(err)),
+        ]);
+    }
+    const outcome = parseEvalOutcome(bytes.toString("utf8"), absPath, projectRoot);
+    if (!outcome.ok)
+        return outcome;
+    return loadSuccess({ ...outcome.value, fileHash: sha256Hex(bytes) });
+}