@velum-labs/routekit-eval-core 1.0.0

package/LICENSE

@@ -0,0 +1,201 @@
+                                 Apache License
+                           Version 2.0, January 2004
+                        http://www.apache.org/licenses/
+
+   TERMS AND CONDITIONS FOR USE, REPRODUCTION, AND DISTRIBUTION
+
+   1. Definitions.
+
+      "License" shall mean the terms and conditions for use, reproduction,
+      and distribution as defined by Sections 1 through 9 of this document.
+
+      "Licensor" shall mean the copyright owner or entity authorized by
+      the copyright owner that is granting the License.
+
+      "Legal Entity" shall mean the union of the acting entity and all
+      other entities that control, are controlled by, or are under common
+      control with that entity. For the purposes of this definition,
+      "control" means (i) the power, direct or indirect, to cause the
+      direction or management of such entity, whether by contract or
+      otherwise, or (ii) ownership of fifty percent (50%) or more of the
+      outstanding shares, or (iii) beneficial ownership of such entity.
+
+      "You" (or "Your") shall mean an individual or Legal Entity
+      exercising permissions granted by this License.
+
+      "Source" form shall mean the preferred form for making modifications,
+      including but not limited to software source code, documentation
+      source, and configuration files.
+
+      "Object" form shall mean any form resulting from mechanical
+      transformation or translation of a Source form, including but
+      not limited to compiled object code, generated documentation,
+      and conversions to other media types.
+
+      "Work" shall mean the work of authorship, whether in Source or
+      Object form, made available under the License, as indicated by a
+      copyright notice that is included in or attached to the work
+      (an example is provided in the Appendix below).
+
+      "Derivative Works" shall mean any work, whether in Source or Object
+      form, that is based on (or derived from) the Work and for which the
+      editorial revisions, annotations, elaborations, or other modifications
+      represent, as a whole, an original work of authorship. For the purposes
+      of this License, Derivative Works shall not include works that remain
+      separable from, or merely link (or bind by name) to the interfaces of,
+      the Work and Derivative Works thereof.
+
+      "Contribution" shall mean any work of authorship, including
+      the original version of the Work and any modifications or additions
+      to that Work or Derivative Works thereof, that is intentionally
+      submitted to Licensor for inclusion in the Work by the copyright owner
+      or by an individual or Legal Entity authorized to submit on behalf of
+      the copyright owner. For the purposes of this definition, "submitted"
+      means any form of electronic, verbal, or written communication sent
+      to the Licensor or its representatives, including but not limited to
+      communication on electronic mailing lists, source code control systems,
+      and issue tracking systems that are managed by, or on behalf of, the
+      Licensor for the purpose of discussing and improving the Work, but
+      excluding communication that is conspicuously marked or otherwise
+      designated in writing by the copyright owner as "Not a Contribution."
+
+      "Contributor" shall mean Licensor and any individual or Legal Entity
+      on behalf of whom a Contribution has been received by Licensor and
+      subsequently incorporated within the Work.
+
+   2. Grant of Copyright License. Subject to the terms and conditions of
+      this License, each Contributor hereby grants to You a perpetual,
+      worldwide, non-exclusive, no-charge, royalty-free, irrevocable
+      copyright license to reproduce, prepare Derivative Works of,
+      publicly display, publicly perform, sublicense, and distribute the
+      Work and such Derivative Works in Source or Object form.
+
+   3. Grant of Patent License. Subject to the terms and conditions of
+      this License, each Contributor hereby grants to You a perpetual,
+      worldwide, non-exclusive, no-charge, royalty-free, irrevocable
+      (except as stated in this section) patent license to make, have made,
+      use, offer to sell, sell, import, and otherwise transfer the Work,
+      where such license applies only to those patent claims licensable
+      by such Contributor that are necessarily infringed by their
+      Contribution(s) alone or by combination of their Contribution(s)
+      with the Work to which such Contribution(s) was submitted. If You
+      institute patent litigation against any entity (including a
+      cross-claim or counterclaim in a lawsuit) alleging that the Work
+      or a Contribution incorporated within the Work constitutes direct
+      or contributory patent infringement, then any patent licenses
+      granted to You under this License for that Work shall terminate
+      as of the date such litigation is filed.
+
+   4. Redistribution. You may reproduce and distribute copies of the
+      Work or Derivative Works thereof in any medium, with or without
+      modifications, and in Source or Object form, provided that You
+      meet the following conditions:
+
+      (a) You must give any other recipients of the Work or
+          Derivative Works a copy of this License; and
+
+      (b) You must cause any modified files to carry prominent notices
+          stating that You changed the files; and
+
+      (c) You must retain, in the Source form of any Derivative Works
+          that You distribute, all copyright, patent, trademark, and
+          attribution notices from the Source form of the Work,
+          excluding those notices that do not pertain to any part of
+          the Derivative Works; and
+
+      (d) If the Work includes a "NOTICE" text file as part of its
+          distribution, then any Derivative Works that You distribute must
+          include a readable copy of the attribution notices contained
+          within such NOTICE file, excluding those notices that do not
+          pertain to any part of the Derivative Works, in at least one
+          of the following places: within a NOTICE text file distributed
+          as part of the Derivative Works; within the Source form or
+          documentation, if provided along with the Derivative Works; or,
+          within a display generated by the Derivative Works, if and
+          wherever such third-party notices normally appear. The contents
+          of the NOTICE file are for informational purposes only and
+          do not modify the License. You may add Your own attribution
+          notices within Derivative Works that You distribute, alongside
+          or as an addendum to the NOTICE text from the Work, provided
+          that such additional attribution notices cannot be construed
+          as modifying the License.
+
+      You may add Your own copyright statement to Your modifications and
+      may provide additional or different license terms and conditions
+      for use, reproduction, or distribution of Your modifications, or
+      for any such Derivative Works as a whole, provided Your use,
+      reproduction, and distribution of the Work otherwise complies with
+      the conditions stated in this License.
+
+   5. Submission of Contributions. Unless You explicitly state otherwise,
+      any Contribution intentionally submitted for inclusion in the Work
+      by You to the Licensor shall be under the terms and conditions of
+      this License, without any additional terms or conditions.
+      Notwithstanding the above, nothing herein shall supersede or modify
+      the terms of any separate license agreement you may have executed
+      with Licensor regarding such Contributions.
+
+   6. Trademarks. This License does not grant permission to use the trade
+      names, trademarks, service marks, or product names of the Licensor,
+      except as required for reasonable and customary use in describing the
+      origin of the Work and reproducing the content of the NOTICE file.
+
+   7. Disclaimer of Warranty. Unless required by applicable law or
+      agreed to in writing, Licensor provides the Work (and each
+      Contributor provides its Contributions) on an "AS IS" BASIS,
+      WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or
+      implied, including, without limitation, any warranties or conditions
+      of TITLE, NON-INFRINGEMENT, MERCHANTABILITY, or FITNESS FOR A
+      PARTICULAR PURPOSE. You are solely responsible for determining the
+      appropriateness of using or redistributing the Work and assume any
+      risks associated with Your exercise of permissions under this License.
+
+   8. Limitation of Liability. In no event and under no legal theory,
+      whether in tort (including negligence), contract, or otherwise,
+      unless required by applicable law (such as deliberate and grossly
+      negligent acts) or agreed to in writing, shall any Contributor be
+      liable to You for damages, including any direct, indirect, special,
+      incidental, or consequential damages of any character arising as a
+      result of this License or out of the use or inability to use the
+      Work (including but not limited to damages for loss of goodwill,
+      work stoppage, computer failure or malfunction, or any and all
+      other commercial damages or losses), even if such Contributor
+      has been advised of the possibility of such damages.
+
+   9. Accepting Warranty or Additional Liability. While redistributing
+      the Work or Derivative Works thereof, You may choose to offer,
+      and charge a fee for, acceptance of support, warranty, indemnity,
+      or other liability obligations and/or rights consistent with this
+      License. However, in accepting such obligations, You may act only
+      on Your own behalf and on Your sole responsibility, not on behalf
+      of any other Contributor, and only if You agree to indemnify,
+      defend, and hold each Contributor harmless for any liability
+      incurred by, or claims asserted against, such Contributor by reason
+      of your accepting any such warranty or additional liability.
+
+   END OF TERMS AND CONDITIONS
+
+   APPENDIX: How to apply the Apache License to your work.
+
+      To apply the Apache License to your work, attach the following
+      boilerplate notice, with the fields enclosed by brackets "[]"
+      replaced with your own identifying information. (Don't include
+      the brackets!)  The text should be enclosed in the appropriate
+      comment syntax for the file format. We also recommend that a
+      file or class name and description of purpose be included on the
+      same "printed page" as the copyright notice for easier
+      identification within third-party archives.
+
+   Copyright 2026 Velum Labs
+
+   Licensed under the Apache License, Version 2.0 (the "License");
+   you may not use this file except in compliance with the License.
+   You may obtain a copy of the License at
+
+       http://www.apache.org/licenses/LICENSE-2.0
+
+   Unless required by applicable law or agreed to in writing, software
+   distributed under the License is distributed on an "AS IS" BASIS,
+   WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+   See the License for the specific language governing permissions and
+   limitations under the License.

package/README.md

@@ -0,0 +1,6 @@
+# @velum-labs/routekit-eval-core
+
+Effect execution and aggregation for RouteKit evaluations. Candidate and judge
+calls use a dedicated token, explicit model IDs, attribution metadata, and a
+policy-bypass header so evaluation cannot recursively invoke the auto-router.
+The online request path never imports this package.

package/dist/effect-api.d.ts

@@ -0,0 +1,2 @@
+export type { EvalEgressOptions } from "./egress.js";
+export { type RoutingModelAvailability, type RoutingScoreConstraints, type RoutingScoreResult, RoutingScoringError, type RoutingScoringErrorCode, type ScoreRoutingCandidatesInput, scoreRoutingCandidates } from "./routing-score.js";

package/dist/effect-api.js

@@ -0,0 +1 @@
+export { RoutingScoringError, scoreRoutingCandidates } from "./routing-score.js";

package/dist/egress.d.ts

@@ -0,0 +1,15 @@
+import { type EvalRole } from "@velum-labs/routekit-eval-contracts";
+import { Effect } from "effect";
+import { HttpClient } from "effect/unstable/http";
+export type EvalEgressOptions = {
+    gatewayUrl: string;
+    token: string;
+};
+export declare function completeEvalChat(input: {
+    egress: EvalEgressOptions;
+    model: string;
+    role: EvalRole;
+    runId: string;
+    caseId: string;
+    prompt: string;
+}): Effect.Effect<string, Error, HttpClient.HttpClient>;

package/dist/egress.js

@@ -0,0 +1,46 @@
+import { assertExplicitEvalModel, EVAL_ATTRIBUTION_HEADER, EVAL_POLICY_BYPASS_HEADER } from "@velum-labs/routekit-eval-contracts";
+import { RouteKitFailure, toRouteKitFailure } from "@velum-labs/routekit-runtime/effect";
+import { Effect } from "effect";
+import { HttpClient, HttpClientRequest } from "effect/unstable/http";
+function completionText(payload) {
+    if (typeof payload !== "object" || payload === null)
+        return "";
+    const choices = payload.choices;
+    if (!Array.isArray(choices) || choices[0] === undefined)
+        return "";
+    const message = choices[0].message;
+    return typeof message?.content === "string" ? message.content : "";
+}
+export function completeEvalChat(input) {
+    const url = `${input.egress.gatewayUrl.replace(/\/$/, "")}/v1/chat/completions`;
+    const request = HttpClientRequest.post(url).pipe(HttpClientRequest.setHeaders({
+        authorization: `Bearer ${input.egress.token}`,
+        "content-type": "application/json",
+        [EVAL_POLICY_BYPASS_HEADER]: "1",
+        [EVAL_ATTRIBUTION_HEADER]: JSON.stringify({
+            purpose: "eval",
+            role: input.role,
+            runId: input.runId,
+            caseId: input.caseId
+        })
+    }), HttpClientRequest.bodyText(JSON.stringify({
+        model: input.model,
+        messages: [{ role: "user", content: input.prompt }],
+        stream: false
+    })));
+    return Effect.gen(function* () {
+        yield* Effect.try({
+            try: () => assertExplicitEvalModel(input.model, input.role),
+            catch: (cause) => toRouteKitFailure(cause)
+        });
+        const client = yield* HttpClient.HttpClient;
+        const response = yield* client.execute(request);
+        if (response.status < 200 || response.status >= 300) {
+            return yield* new RouteKitFailure({
+                message: `${input.role} call failed (${response.status})`
+            });
+        }
+        const payload = yield* response.json;
+        return completionText(payload);
+    });
+}

package/dist/index.d.ts

@@ -0,0 +1,2 @@
+export type { EvalEgressOptions } from "./egress.js";
+export { type RoutingModelAvailability, type RoutingScoreConstraints, type RoutingScoreResult, RoutingScoringError, type RoutingScoringErrorCode, type ScoreRoutingCandidatesInput, scoreRoutingCandidates } from "./routing-score.js";

package/dist/index.js

@@ -0,0 +1 @@
+export { RoutingScoringError, scoreRoutingCandidates } from "./routing-score.js";

package/dist/routing-score.d.ts

@@ -0,0 +1,41 @@
+import type { PublishedRoutingActivation, RequestDecomposition, RequestRoutingRequirements, RoutingCandidateDecision, RoutingEndpoint, RoutingObjectivePolicy } from "@velum-labs/routekit-eval-contracts";
+export type RoutingModelAvailability = Readonly<{
+    model: string;
+    served: boolean;
+    endpoints: readonly RoutingEndpoint[];
+    supportsTools: boolean;
+    supportsVision: boolean;
+    maxInputTokens?: number;
+    maxOutputTokens?: number;
+}>;
+export type RoutingScoreConstraints = Readonly<{
+    /** Per-dimension conservative quality floors, applied before aggregation. */
+    minimumDimensionQuality?: Readonly<Record<string, number>>;
+    /** Maximum weighted failure rate for a candidate. */
+    maximumFailureRate?: number;
+}>;
+export type ScoreRoutingCandidatesInput = Readonly<{
+    snapshot: PublishedRoutingActivation;
+    decomposition: RequestDecomposition;
+    requirements: RequestRoutingRequirements;
+    objective: RoutingObjectivePolicy;
+    availableModels: readonly RoutingModelAvailability[];
+    constraints?: RoutingScoreConstraints;
+}>;
+export type RoutingScoreResult = Readonly<{
+    candidates: readonly RoutingCandidateDecision[];
+    selectedModel: string;
+    fallbackModels: readonly string[];
+}>;
+export type RoutingScoringErrorCode = "invalid_input" | "no_eligible_models";
+export declare class RoutingScoringError extends Error {
+    readonly code: RoutingScoringErrorCode;
+    readonly candidates: readonly RoutingCandidateDecision[];
+    constructor(code: RoutingScoringErrorCode, message: string, candidates?: readonly RoutingCandidateDecision[]);
+}
+/**
+ * Deterministically rank candidate models from a semantic decomposition and a
+ * published model-by-dimension evidence matrix. This function performs no I/O and
+ * never treats missing or partially unpriced cost as zero.
+ */
+export declare function scoreRoutingCandidates(input: ScoreRoutingCandidatesInput): RoutingScoreResult;

package/dist/routing-score.js

@@ -0,0 +1,429 @@
+import { REQUEST_DECOMPOSITION_TOLERANCE } from "@velum-labs/routekit-eval-contracts";
+const VECTOR_TOLERANCE = REQUEST_DECOMPOSITION_TOLERANCE;
+export class RoutingScoringError extends Error {
+    code;
+    candidates;
+    constructor(code, message, candidates = []) {
+        super(message);
+        this.name = "RoutingScoringError";
+        this.code = code;
+        this.candidates = candidates;
+    }
+}
+function invalid(message) {
+    throw new RoutingScoringError("invalid_input", message);
+}
+function finiteUnit(value, label) {
+    if (!Number.isFinite(value) || value < 0 || value > 1) {
+        invalid(`${label} must be a finite number between zero and one`);
+    }
+}
+function unique(values, label) {
+    if (new Set(values).size !== values.length)
+        invalid(`${label} must not contain duplicates`);
+}
+function compareModelIds(left, right) {
+    return left < right ? -1 : left > right ? 1 : 0;
+}
+function compareNumberAscending(left, right) {
+    if (left === undefined)
+        return right === undefined ? 0 : 1;
+    if (right === undefined)
+        return -1;
+    return left - right;
+}
+function compareNumberDescending(left, right) {
+    if (left === undefined)
+        return right === undefined ? 0 : 1;
+    if (right === undefined)
+        return -1;
+    return right - left;
+}
+function activeWeights(input) {
+    if (input.snapshot.basisDigest !== input.decomposition.basisDigest) {
+        invalid("dimension definition digest does not match the routing snapshot");
+    }
+    const dimensionIds = input.snapshot.dimensions.map((dimension) => dimension.id);
+    unique(dimensionIds, "snapshot dimension IDs");
+    unique(input.snapshot.candidateModels, "snapshot candidate models");
+    if (dimensionIds.length === 0)
+        invalid("routing snapshot must contain at least one dimension");
+    if (input.snapshot.candidateModels.length === 0) {
+        invalid("routing snapshot must contain at least one candidate model");
+    }
+    const weights = new Map();
+    for (const entry of input.decomposition.weights) {
+        finiteUnit(entry.weight, `weight for dimension ${entry.dimensionId}`);
+        if (weights.has(entry.dimensionId))
+            invalid(`duplicate decomposition dimension: ${entry.dimensionId}`);
+        if (!dimensionIds.includes(entry.dimensionId))
+            invalid(`unknown decomposition dimension: ${entry.dimensionId}`);
+        weights.set(entry.dimensionId, entry.weight);
+    }
+    for (const dimensionId of dimensionIds) {
+        if (!weights.has(dimensionId))
+            invalid(`missing decomposition dimension: ${dimensionId}`);
+    }
+    finiteUnit(input.decomposition.unknownWeight, "unknown weight");
+    const total = [...weights.values()].reduce((sum, weight) => sum + weight, 0);
+    if (Math.abs(total + input.decomposition.unknownWeight - 1) > VECTOR_TOLERANCE) {
+        invalid("dimension weights and unknown weight must sum to one");
+    }
+    if (total <= VECTOR_TOLERANCE)
+        invalid("routing requires at least one covered dimension");
+    return dimensionIds.flatMap((dimensionId) => {
+        const weight = weights.get(dimensionId) ?? 0;
+        return weight > VECTOR_TOLERANCE ? [{ dimensionId, weight: weight / total }] : [];
+    });
+}
+function validateInput(input) {
+    const active = activeWeights(input);
+    const availableIds = input.availableModels.map((model) => model.model);
+    unique(availableIds, "available model IDs");
+    const floors = input.constraints?.minimumDimensionQuality ?? {};
+    const knownDimensions = new Set(input.snapshot.dimensions.map((dimension) => dimension.id));
+    for (const [dimensionId, floor] of Object.entries(floors)) {
+        if (!knownDimensions.has(dimensionId))
+            invalid(`quality floor refers to unknown dimension: ${dimensionId}`);
+        finiteUnit(floor, `quality floor for dimension ${dimensionId}`);
+    }
+    const maximumFailureRate = input.constraints?.maximumFailureRate;
+    if (maximumFailureRate !== undefined) {
+        finiteUnit(maximumFailureRate, "maximum failure rate");
+    }
+    switch (input.objective.kind) {
+        case "highest-quality":
+            break;
+        case "lowest-cost":
+        case "lowest-latency":
+        case "pareto":
+            finiteUnit(input.objective.minimumQuality, "minimum quality");
+            break;
+        case "balanced": {
+            finiteUnit(input.objective.minimumQuality, "minimum quality");
+            const { quality, cost, latency } = input.objective.weights;
+            finiteUnit(quality, "balanced quality weight");
+            finiteUnit(cost, "balanced cost weight");
+            finiteUnit(latency, "balanced latency weight");
+            if (Math.abs(quality + cost + latency - 1) > VECTOR_TOLERANCE) {
+                invalid("balanced objective weights must sum to one");
+            }
+            break;
+        }
+        default: {
+            const unreachable = input.objective;
+            invalid(`unsupported routing objective: ${String(unreachable)}`);
+        }
+    }
+    return active;
+}
+function capabilityReasons(availability, requirements) {
+    if (availability === undefined || !availability.served)
+        return ["model_not_served"];
+    const reasons = [];
+    if (!availability.endpoints.includes(requirements.endpoint)) {
+        reasons.push(`endpoint_not_supported:${requirements.endpoint}`);
+    }
+    if (requirements.requiresTools && !availability.supportsTools) {
+        reasons.push("tools_not_supported");
+    }
+    if (requirements.requiresVision && !availability.supportsVision) {
+        reasons.push("vision_not_supported");
+    }
+    // Numeric limits are optional basis assertions. A known insufficient
+    // limit excludes the model; absent metadata must not be misrepresented as
+    // proof that an otherwise served model is incapable.
+    if (requirements.inputTokens !== undefined &&
+        availability.maxInputTokens !== undefined &&
+        availability.maxInputTokens < requirements.inputTokens) {
+        reasons.push("input_token_limit_insufficient");
+    }
+    if (requirements.maxOutputTokens !== undefined &&
+        availability.maxOutputTokens !== undefined &&
+        availability.maxOutputTokens < requirements.maxOutputTokens) {
+        reasons.push("output_token_limit_insufficient");
+    }
+    return reasons;
+}
+function calculateCandidates(input, active) {
+    const availability = new Map(input.availableModels.map((model) => [model.model, model]));
+    const floors = input.constraints?.minimumDimensionQuality ?? {};
+    return input.snapshot.candidateModels.map((model) => {
+        const reasons = capabilityReasons(availability.get(model), input.requirements);
+        let quality = 0;
+        let failureRate = 0;
+        let duration = 0;
+        let cost = 0;
+        let completeCoreEvidence = true;
+        let completeDuration = true;
+        let completeCost = true;
+        for (const { dimensionId, weight } of active) {
+            const cells = input.snapshot.evidence.filter((cell) => cell.model === model && cell.dimensionId === dimensionId);
+            if (cells.length === 0) {
+                reasons.push(`missing_evidence:${dimensionId}`);
+                completeCoreEvidence = false;
+                completeDuration = false;
+                completeCost = false;
+                continue;
+            }
+            if (cells.length > 1) {
+                reasons.push(`duplicate_evidence:${dimensionId}`);
+                completeCoreEvidence = false;
+                completeDuration = false;
+                completeCost = false;
+                continue;
+            }
+            const cell = cells[0];
+            if (cell === undefined)
+                continue;
+            quality += cell.quality.lowerConfidenceBound * weight;
+            failureRate += cell.failureRate * weight;
+            const floor = floors[dimensionId];
+            if (floor !== undefined && cell.quality.lowerConfidenceBound < floor) {
+                reasons.push(`quality_below_dimension_floor:${dimensionId}`);
+            }
+            if (cell.p95DurationMs === undefined)
+                completeDuration = false;
+            else
+                duration += cell.p95DurationMs * weight;
+            if (cell.averageCostUsd === undefined || cell.unpricedCalls > 0)
+                completeCost = false;
+            else
+                cost += cell.averageCostUsd * weight;
+        }
+        const candidate = {
+            model,
+            reasons,
+            costStatus: completeCoreEvidence && completeCost ? "known" : "unavailable",
+            ...(completeCoreEvidence ? { quality, failureRate } : {}),
+            ...(completeCoreEvidence && completeDuration ? { p95DurationMs: duration } : {}),
+            ...(completeCoreEvidence && completeCost ? { averageCostUsd: cost } : {})
+        };
+        if (candidate.failureRate !== undefined &&
+            input.constraints?.maximumFailureRate !== undefined &&
+            candidate.failureRate > input.constraints.maximumFailureRate) {
+            reasons.push("failure_rate_above_maximum");
+        }
+        return candidate;
+    });
+}
+function applyObjectiveEligibility(candidates, objective) {
+    for (const candidate of candidates) {
+        const quality = candidate.quality;
+        if (quality === undefined)
+            continue;
+        switch (objective.kind) {
+            case "highest-quality":
+                break;
+            case "lowest-cost":
+                if (quality < objective.minimumQuality)
+                    candidate.reasons.push("quality_below_minimum");
+                if (candidate.averageCostUsd === undefined)
+                    candidate.reasons.push("cost_unavailable");
+                break;
+            case "lowest-latency":
+                if (quality < objective.minimumQuality)
+                    candidate.reasons.push("quality_below_minimum");
+                if (candidate.p95DurationMs === undefined)
+                    candidate.reasons.push("latency_unavailable");
+                break;
+            case "balanced":
+                if (quality < objective.minimumQuality)
+                    candidate.reasons.push("quality_below_minimum");
+                if (objective.weights.cost > 0 && candidate.averageCostUsd === undefined) {
+                    candidate.reasons.push("cost_unavailable");
+                }
+                if (objective.weights.latency > 0 && candidate.p95DurationMs === undefined) {
+                    candidate.reasons.push("latency_unavailable");
+                }
+                break;
+            case "pareto":
+                if (quality < objective.minimumQuality)
+                    candidate.reasons.push("quality_below_minimum");
+                if (candidate.averageCostUsd === undefined)
+                    candidate.reasons.push("cost_unavailable");
+                if (candidate.p95DurationMs === undefined)
+                    candidate.reasons.push("latency_unavailable");
+                break;
+            default: {
+                const unreachable = objective;
+                invalid(`unsupported routing objective: ${String(unreachable)}`);
+            }
+        }
+    }
+}
+function normalizeHigher(value, minimum, maximum) {
+    return maximum === minimum ? 1 : (value - minimum) / (maximum - minimum);
+}
+function normalizeLower(value, minimum, maximum) {
+    return maximum === minimum ? 1 : (maximum - value) / (maximum - minimum);
+}
+function assignBalancedUtility(candidates, objective) {
+    const eligible = candidates.filter((candidate) => candidate.reasons.length === 0);
+    const qualities = eligible.flatMap((candidate) => candidate.quality === undefined ? [] : [candidate.quality]);
+    const costs = eligible.flatMap((candidate) => candidate.averageCostUsd === undefined ? [] : [candidate.averageCostUsd]);
+    const latencies = eligible.flatMap((candidate) => candidate.p95DurationMs === undefined ? [] : [candidate.p95DurationMs]);
+    if (qualities.length === 0)
+        return;
+    const qualityMin = Math.min(...qualities);
+    const qualityMax = Math.max(...qualities);
+    const costMin = costs.length === 0 ? 0 : Math.min(...costs);
+    const costMax = costs.length === 0 ? 0 : Math.max(...costs);
+    const latencyMin = latencies.length === 0 ? 0 : Math.min(...latencies);
+    const latencyMax = latencies.length === 0 ? 0 : Math.max(...latencies);
+    for (const candidate of eligible) {
+        const quality = candidate.quality;
+        if (quality === undefined)
+            continue;
+        const qualityUtility = normalizeHigher(quality, qualityMin, qualityMax);
+        const costUtility = objective.weights.cost === 0
+            ? 0
+            : normalizeLower(candidate.averageCostUsd ?? 0, costMin, costMax);
+        const latencyUtility = objective.weights.latency === 0
+            ? 0
+            : normalizeLower(candidate.p95DurationMs ?? 0, latencyMin, latencyMax);
+        candidate.utility =
+            objective.weights.quality * qualityUtility +
+                objective.weights.cost * costUtility +
+                objective.weights.latency * latencyUtility;
+    }
+}
+function standardComparator(objective) {
+    return (left, right) => {
+        let comparison = 0;
+        switch (objective.kind) {
+            case "highest-quality":
+                comparison =
+                    compareNumberDescending(left.quality, right.quality) ||
+                        compareNumberAscending(left.failureRate, right.failureRate) ||
+                        compareNumberAscending(left.p95DurationMs, right.p95DurationMs) ||
+                        compareNumberAscending(left.averageCostUsd, right.averageCostUsd);
+                break;
+            case "lowest-cost":
+                comparison =
+                    compareNumberAscending(left.averageCostUsd, right.averageCostUsd) ||
+                        compareNumberDescending(left.quality, right.quality) ||
+                        compareNumberAscending(left.failureRate, right.failureRate) ||
+                        compareNumberAscending(left.p95DurationMs, right.p95DurationMs);
+                break;
+            case "lowest-latency":
+                comparison =
+                    compareNumberAscending(left.p95DurationMs, right.p95DurationMs) ||
+                        compareNumberDescending(left.quality, right.quality) ||
+                        compareNumberAscending(left.failureRate, right.failureRate) ||
+                        compareNumberAscending(left.averageCostUsd, right.averageCostUsd);
+                break;
+            case "balanced":
+                comparison =
+                    compareNumberDescending(left.utility, right.utility) ||
+                        compareNumberDescending(left.quality, right.quality) ||
+                        compareNumberAscending(left.averageCostUsd, right.averageCostUsd) ||
+                        compareNumberAscending(left.p95DurationMs, right.p95DurationMs) ||
+                        compareNumberAscending(left.failureRate, right.failureRate);
+                break;
+            default: {
+                const unreachable = objective;
+                invalid(`unsupported routing objective: ${String(unreachable)}`);
+            }
+        }
+        return comparison || compareModelIds(left.model, right.model);
+    };
+}
+function dominates(left, right) {
+    const leftQuality = left.quality;
+    const rightQuality = right.quality;
+    const leftCost = left.averageCostUsd;
+    const rightCost = right.averageCostUsd;
+    const leftLatency = left.p95DurationMs;
+    const rightLatency = right.p95DurationMs;
+    if (leftQuality === undefined ||
+        rightQuality === undefined ||
+        leftCost === undefined ||
+        rightCost === undefined ||
+        leftLatency === undefined ||
+        rightLatency === undefined) {
+        return false;
+    }
+    const noWorse = leftQuality >= rightQuality && leftCost <= rightCost && leftLatency <= rightLatency;
+    const strictlyBetter = leftQuality > rightQuality || leftCost < rightCost || leftLatency < rightLatency;
+    return noWorse && strictlyBetter;
+}
+function paretoPreferenceComparator(preference) {
+    return (left, right) => {
+        const preferred = preference === "quality"
+            ? compareNumberDescending(left.quality, right.quality)
+            : preference === "cost"
+                ? compareNumberAscending(left.averageCostUsd, right.averageCostUsd)
+                : compareNumberAscending(left.p95DurationMs, right.p95DurationMs);
+        return (preferred ||
+            compareNumberDescending(left.quality, right.quality) ||
+            compareNumberAscending(left.averageCostUsd, right.averageCostUsd) ||
+            compareNumberAscending(left.p95DurationMs, right.p95DurationMs) ||
+            compareNumberAscending(left.failureRate, right.failureRate) ||
+            compareModelIds(left.model, right.model));
+    };
+}
+function paretoOrder(candidates, preference) {
+    const remaining = [...candidates];
+    const ordered = [];
+    const comparator = paretoPreferenceComparator(preference);
+    while (remaining.length > 0) {
+        const frontier = remaining.filter((candidate) => !remaining.some((other) => other !== candidate && dominates(other, candidate)));
+        frontier.sort(comparator);
+        ordered.push(...frontier);
+        const frontierModels = new Set(frontier.map((candidate) => candidate.model));
+        for (let index = remaining.length - 1; index >= 0; index -= 1) {
+            const candidate = remaining[index];
+            if (candidate !== undefined && frontierModels.has(candidate.model))
+                remaining.splice(index, 1);
+        }
+    }
+    return ordered;
+}
+function publicDecision(candidate) {
+    return {
+        model: candidate.model,
+        eligible: candidate.reasons.length === 0,
+        exclusionReasons: candidate.reasons,
+        costStatus: candidate.costStatus,
+        ...(candidate.quality === undefined ? {} : { quality: candidate.quality }),
+        ...(candidate.failureRate === undefined ? {} : { failureRate: candidate.failureRate }),
+        ...(candidate.p95DurationMs === undefined ? {} : { p95DurationMs: candidate.p95DurationMs }),
+        ...(candidate.averageCostUsd === undefined ? {} : { averageCostUsd: candidate.averageCostUsd }),
+        ...(candidate.utility === undefined ? {} : { utility: candidate.utility }),
+        ...(candidate.rank === undefined ? {} : { rank: candidate.rank })
+    };
+}
+/**
+ * Deterministically rank candidate models from a semantic decomposition and a
+ * published model-by-dimension evidence matrix. This function performs no I/O and
+ * never treats missing or partially unpriced cost as zero.
+ */
+export function scoreRoutingCandidates(input) {
+    const active = validateInput(input);
+    const candidates = calculateCandidates(input, active);
+    applyObjectiveEligibility(candidates, input.objective);
+    if (input.objective.kind === "balanced") {
+        assignBalancedUtility(candidates, input.objective);
+    }
+    const eligible = candidates.filter((candidate) => candidate.reasons.length === 0);
+    const rejected = candidates
+        .filter((candidate) => candidate.reasons.length > 0)
+        .sort((left, right) => compareModelIds(left.model, right.model));
+    const ranked = input.objective.kind === "pareto"
+        ? paretoOrder(eligible, input.objective.preference)
+        : [...eligible].sort(standardComparator(input.objective));
+    ranked.forEach((candidate, index) => {
+        candidate.rank = index + 1;
+    });
+    const decisions = [...ranked, ...rejected].map(publicDecision);
+    const selected = ranked[0];
+    if (selected === undefined) {
+        throw new RoutingScoringError("no_eligible_models", "no candidate model satisfies the routing requirements and objective", decisions);
+    }
+    return {
+        candidates: decisions,
+        selectedModel: selected.model,
+        fallbackModels: ranked.slice(1).map((candidate) => candidate.model)
+    };
+}

package/dist/test/retired-eng814-imports.test.d.ts

@@ -0,0 +1 @@
+export {};

package/dist/test/retired-eng814-imports.test.js

@@ -0,0 +1,5 @@
+import assert from "node:assert/strict";
+import { test } from "node:test";
+test("retired ENG-814 imports remain compile failures", () => {
+    assert.ok(true);
+});

package/dist/test/routing-score.test.d.ts

@@ -0,0 +1 @@
+export {};

package/dist/test/routing-score.test.js

@@ -0,0 +1,211 @@
+import assert from "node:assert/strict";
+import { test } from "node:test";
+import { RoutingScoringError, scoreRoutingCandidates } from "../routing-score.js";
+const models = ["provider/alpha", "provider/beta", "provider/gamma"];
+const dimensions = [
+    { id: "code", description: "Code changes", includes: ["implementation"], excludes: [] },
+    { id: "docs", description: "Documentation", includes: ["writing"], excludes: [] }
+];
+function cell(model, dimensionId, quality, options = {}) {
+    return {
+        model,
+        dimensionId,
+        suiteDigest: `suite-${dimensionId}`,
+        evidenceDigest: `evidence-${model}-${dimensionId}`,
+        quality: { passRate: quality, lowerConfidenceBound: quality, sampleCount: 20 },
+        failureRate: options.failureRate ?? 1 - quality,
+        ...(options.duration === undefined ? {} : { p95DurationMs: options.duration }),
+        ...(options.cost === undefined ? {} : { averageCostUsd: options.cost }),
+        unpricedCalls: options.unpricedCalls ?? (options.cost === undefined ? 20 : 0)
+    };
+}
+function snapshot(evidence, candidateModels = models) {
+    return {
+        version: 2,
+        generatedAt: "2026-08-17T00:00:00.000Z",
+        basisDigest: "definition-digest",
+        evidenceDigest: "evidence-digest",
+        classifierModel: "provider/classifier",
+        objective: { kind: "highest-quality" },
+        maximumUnknownWeight: 0.2,
+        dimensions,
+        candidateModels,
+        evidence
+    };
+}
+const decomposition = {
+    version: 2,
+    basisDigest: "definition-digest",
+    weights: [
+        { dimensionId: "code", weight: 0.54 },
+        { dimensionId: "docs", weight: 0.36 }
+    ],
+    unknownWeight: 0.1
+};
+const requirements = {
+    endpoint: "responses",
+    requiresTools: false,
+    requiresVision: false,
+    inputTokens: 4_000,
+    maxOutputTokens: 2_000
+};
+function available(model, overrides = {}) {
+    return {
+        model,
+        served: true,
+        endpoints: ["responses"],
+        supportsTools: true,
+        supportsVision: true,
+        maxInputTokens: 100_000,
+        maxOutputTokens: 8_000,
+        ...overrides
+    };
+}
+function run(evidence, objective = { kind: "highest-quality" }, options = {}) {
+    return scoreRoutingCandidates({
+        snapshot: snapshot(evidence, options.candidateModels),
+        decomposition: options.requestDecomposition ?? decomposition,
+        requirements,
+        objective,
+        availableModels: options.availableModels ?? models.map((model) => available(model))
+    });
+}
+const completeEvidence = [
+    cell(models[0], "code", 0.9, { cost: 3, duration: 100 }),
+    cell(models[0], "docs", 0.6, { cost: 1, duration: 300 }),
+    cell(models[1], "code", 0.75, { cost: 1, duration: 200 }),
+    cell(models[1], "docs", 0.8, { cost: 1, duration: 200 }),
+    cell(models[2], "code", 0.7, { cost: 0.5, duration: 80 }),
+    cell(models[2], "docs", 0.7, { cost: 0.5, duration: 80 })
+];
+test("highest-quality aggregates covered dimension weights and is candidate-order invariant", () => {
+    const first = run(completeEvidence);
+    const reversed = run([...completeEvidence].reverse(), { kind: "highest-quality" }, {
+        candidateModels: [...models].reverse()
+    });
+    assert.equal(first.selectedModel, models[0]);
+    assert.deepEqual(first.fallbackModels, [models[1], models[2]]);
+    assert.deepEqual(first.candidates.map(({ model, rank }) => ({ model, rank })), reversed.candidates.map(({ model, rank }) => ({ model, rank })));
+    assert.equal(first.candidates[0]?.quality, 0.78);
+    assert.ok(Math.abs((first.candidates[0]?.averageCostUsd ?? 0) - 2.2) < 1e-12);
+    assert.equal(first.candidates[0]?.p95DurationMs, 180);
+});
+test("hard requirements exclude unserved and incapable candidates before scoring", () => {
+    const result = run(completeEvidence, { kind: "highest-quality" }, {
+        availableModels: [
+            available(models[0], { supportsTools: false }),
+            available(models[1], { maxOutputTokens: 1_000 }),
+            available(models[2])
+        ]
+    });
+    const withTools = scoreRoutingCandidates({
+        snapshot: snapshot(completeEvidence),
+        decomposition,
+        requirements: { ...requirements, requiresTools: true },
+        objective: { kind: "highest-quality" },
+        availableModels: [
+            available(models[0], { supportsTools: false }),
+            available(models[1], { maxOutputTokens: 1_000 }),
+            available(models[2])
+        ]
+    });
+    assert.equal(result.selectedModel, models[0]);
+    assert.deepEqual(result.candidates.find(({ model }) => model === models[1])?.exclusionReasons, [
+        "output_token_limit_insufficient"
+    ]);
+    assert.deepEqual(withTools.candidates.find(({ model }) => model === models[0])?.exclusionReasons, ["tools_not_supported"]);
+});
+test("unknown numeric basis limits do not claim a served model is incapable", () => {
+    const result = run(completeEvidence, { kind: "highest-quality" }, {
+        availableModels: models.map((model) => available(model, {
+            maxInputTokens: undefined,
+            maxOutputTokens: undefined
+        }))
+    });
+    assert.equal(result.selectedModel, models[0]);
+    assert.deepEqual(result.candidates.map(({ exclusionReasons }) => exclusionReasons), [[], [], []]);
+});
+test("missing evidence and per-dimension quality floors fail candidates closed", () => {
+    const missingLastCell = completeEvidence.filter((entry) => !(entry.model === models[0] && entry.dimensionId === "docs"));
+    const result = scoreRoutingCandidates({
+        snapshot: snapshot(missingLastCell),
+        decomposition,
+        requirements,
+        objective: { kind: "highest-quality" },
+        availableModels: models.map((model) => available(model)),
+        constraints: { minimumDimensionQuality: { code: 0.72 } }
+    });
+    assert.equal(result.selectedModel, models[1]);
+    assert.deepEqual(result.candidates.find(({ model }) => model === models[0])?.exclusionReasons, [
+        "missing_evidence:docs"
+    ]);
+    assert.deepEqual(result.candidates.find(({ model }) => model === models[2])?.exclusionReasons, [
+        "quality_below_dimension_floor:code"
+    ]);
+});
+test("lowest-cost excludes partially unpriced candidates rather than treating cost as zero", () => {
+    const evidence = completeEvidence.map((entry) => entry.model === models[2] ? { ...entry, unpricedCalls: 1 } : entry);
+    const result = run(evidence, { kind: "lowest-cost", minimumQuality: 0.65 });
+    const unpriced = result.candidates.find(({ model }) => model === models[2]);
+    assert.equal(result.selectedModel, models[1]);
+    assert.equal(unpriced?.costStatus, "unavailable");
+    assert.equal(unpriced?.averageCostUsd, undefined);
+    assert.deepEqual(unpriced?.exclusionReasons, ["cost_unavailable"]);
+});
+test("lowest-latency applies the aggregate quality floor and requires latency evidence", () => {
+    const evidence = completeEvidence.map((entry) => entry.model === models[2] && entry.dimensionId === "docs"
+        ? cell(models[2], "docs", 0.7, { cost: 0.5 })
+        : entry);
+    const result = run(evidence, { kind: "lowest-latency", minimumQuality: 0.72 });
+    assert.equal(result.selectedModel, models[0]);
+    assert.deepEqual(result.candidates.find(({ model }) => model === models[2])?.exclusionReasons, [
+        "quality_below_minimum",
+        "latency_unavailable"
+    ]);
+});
+test("balanced objective normalizes metrics and honors explicit weights", () => {
+    const result = run(completeEvidence, {
+        kind: "balanced",
+        minimumQuality: 0.65,
+        weights: { quality: 0.2, cost: 0.2, latency: 0.6 }
+    });
+    assert.equal(result.selectedModel, models[2]);
+    assert.equal(result.candidates[0]?.utility, 0.8);
+    assert.ok((result.candidates[1]?.utility ?? 0) < (result.candidates[0]?.utility ?? 0));
+});
+test("pareto ranks each non-dominated frontier by the declared preference", () => {
+    const evidence = [
+        cell(models[0], "code", 0.9, { cost: 3, duration: 100 }),
+        cell(models[0], "docs", 0.9, { cost: 3, duration: 100 }),
+        cell(models[1], "code", 0.8, { cost: 2, duration: 200 }),
+        cell(models[1], "docs", 0.8, { cost: 2, duration: 200 }),
+        cell(models[2], "code", 0.7, { cost: 4, duration: 300 }),
+        cell(models[2], "docs", 0.7, { cost: 4, duration: 300 })
+    ];
+    const result = run(evidence, {
+        kind: "pareto",
+        minimumQuality: 0.6,
+        preference: "cost"
+    });
+    assert.equal(result.selectedModel, models[1]);
+    assert.deepEqual(result.fallbackModels, [models[0], models[2]]);
+});
+test("invalid decompositions and balanced weights fail with a typed input error", () => {
+    assert.throws(() => run(completeEvidence, { kind: "highest-quality" }, {
+        requestDecomposition: { ...decomposition, unknownWeight: 0.2 }
+    }), (error) => error instanceof RoutingScoringError && error.code === "invalid_input");
+    assert.throws(() => run(completeEvidence, {
+        kind: "balanced",
+        minimumQuality: 0,
+        weights: { quality: 0.5, cost: 0.5, latency: 0.5 }
+    }), (error) => error instanceof RoutingScoringError && error.code === "invalid_input");
+});
+test("no eligible model returns sanitized candidate reasons in the typed error", () => {
+    assert.throws(() => run(completeEvidence, { kind: "lowest-cost", minimumQuality: 0.99 }), (error) => {
+        assert.ok(error instanceof RoutingScoringError);
+        assert.equal(error.code, "no_eligible_models");
+        assert.equal(error.candidates.length, 3);
+        assert.ok(error.candidates.every((candidate) => !candidate.eligible));
+        return true;
+    });
+});

package/package.json

@@ -0,0 +1,47 @@
+{
+  "name": "@velum-labs/routekit-eval-core",
+  "private": false,
+  "version": "1.0.0",
+  "repository": {
+    "type": "git",
+    "url": "git+https://github.com/velum-labs/routekit.git",
+    "directory": "packages/eval-core"
+  },
+  "description": "Effect evaluation execution and aggregation for RouteKit.",
+  "license": "Apache-2.0",
+  "type": "module",
+  "exports": {
+    ".": {
+      "types": "./dist/index.d.ts",
+      "default": "./dist/index.js"
+    },
+    "./effect": {
+      "types": "./dist/effect-api.d.ts",
+      "default": "./dist/effect-api.js"
+    }
+  },
+  "files": [
+    "dist",
+    "LICENSE"
+  ],
+  "publishConfig": {
+    "registry": "https://registry.npmjs.org",
+    "access": "public",
+    "provenance": true
+  },
+  "dependencies": {
+    "effect": "4.0.0-rc.108",
+    "@velum-labs/routekit-eval-contracts": "1.0.0",
+    "@velum-labs/routekit-runtime": "1.0.0"
+  },
+  "keywords": [
+    "routekit",
+    "eval",
+    "effect"
+  ],
+  "scripts": {
+    "build": "tsc -b",
+    "clean": "tsc -b --clean",
+    "test": "node --test \"dist/test/*.test.js\""
+  }
+}