Address pre-publish review findings for corbits-embedding (#1)

* Address pre-publish review findings for corbits-embedding

Reject batchSize below 1 at the embedTexts entry boundary, document installation and transport exports, and complete the publish metadata.

* docs: install from git until @corbits/embedding is on npm

npm view @corbits/embedding 404s; git is the install path. Default export
is dist/; native Node does not load extensionless TypeScript source.

* fix: emit .js specifiers in dist and integer batchSize

Node cannot import extensionless relative specifiers from packed dist.
Source relative imports now carry .js so tsc emits Node-loadable ESM.
EmbedConfigSchema and embedTexts both require batchSize integer >= 1,
rejecting 0.5 and NaN.

Author: TheGreatAxios

README.md

@@ -22,6 +22,18 @@ from an embedding endpoint therefore behaves exactly as one from a chat
 endpoint, and a `credential_failure` aborts immediately rather than retrying a
 bad key.
 
+## Installation
+
+Not published to npm yet. Until a registry publish, `npm install @corbits/embedding`
+404s. Git is the install path.
+
+```bash
+bun add github:corbitsdev/corbits-embedding
+```
+
+Requires Node >= 20 or Bun >= 1.2. The default export is built `dist/`; native Node
+does not load this package's extensionless TypeScript source.
+
 ```ts
 import { createDefaultScheduler } from "@intx/inference";
 import { embedTexts, probeEmbedDims } from "@corbits/embedding";
@@ -59,6 +71,19 @@ copy of this class until the shared transport is upstreamed, so `instanceof`
 does not hold across the two. Catching both? Discriminate on
 `error.name === "ModelRequestError"`.
 
+## Transport exports
+
+The barrel re-exports the transport `embedTexts` is built on, for sibling
+one-shot JSON clients that want the same classified, retried request path
+without touching `@intx/inference` internals:
+
+- `runJSONRequest` — one JSON POST with error classification and retry.
+- `extractRetryAfterMs` — the default `Retry-After` reader, overridable per
+  call via `extractRetryAfterMs`.
+- `ModelRequestError` — the error every failure mode raises, carrying the
+  classified `InferenceError` as `reason` and the URL.
+- Types `RunRequestOptions` and `RetryAfterExtractor`.
+
 ## Versioning
 
 Semver. Released manually with `bun run build && npm publish` once CI is green.

package.json

@@ -3,6 +3,7 @@
   "description": "Embedding client for OpenAI-compatible /v1/embeddings endpoints, built on the Interchange inference primitives",
   "version": "0.1.0",
   "license": "LGPL-2.1-only",
+  "author": "Sawyer Cutler <sawyer@dirtroad.dev>",
   "type": "module",
   "exports": {
     ".": {
@@ -29,7 +30,11 @@
     "prettier": "3.6.2",
     "typescript": "5.9.3"
   },
-  "files": ["dist", "README.md", "LICENSE"],
+  "files": [
+    "dist",
+    "README.md",
+    "LICENSE"
+  ],
   "sideEffects": false,
   "publishConfig": {
     "access": "public"
@@ -38,7 +43,12 @@
     "type": "git",
     "url": "git+https://github.com/corbitsdev/corbits-embedding.git"
   },
+  "homepage": "https://github.com/corbitsdev/corbits-embedding",
+  "bugs": {
+    "url": "https://github.com/corbitsdev/corbits-embedding/issues"
+  },
   "engines": {
-    "bun": ">=1.2.0"
+    "bun": ">=1.2.0",
+    "node": ">=20"
   }
 }

src/embed.test.ts

@@ -1,7 +1,13 @@
+import { type } from "arktype";
 import { describe, expect, it } from "bun:test";
 import type { Dependencies } from "@intx/inference";
 
-import { embedTexts, probeEmbedDims, type EmbedConfig } from "./embed";
+import {
+  embedTexts,
+  probeEmbedDims,
+  EmbedConfigSchema,
+  type EmbedConfig,
+} from "./embed";
 import { extractRetryAfterMs, ModelRequestError } from "./request";
 
 const CONFIG: EmbedConfig = {
@@ -125,6 +131,19 @@ describe("embedTexts", () => {
     expect(vectors.map((v) => v[0])).toEqual([0, 1, 0]);
   });
 
+  it("rejects a batchSize that is not an integer >= 1 instead of looping forever", async () => {
+    // `batches` advances by `i += size`, so a size of 0 never advances and
+    // grows the batch list until OOM. Fractional sizes would also stall or
+    // slice wrongly. Schema and runtime both require integer >= 1.
+    for (const batchSize of [0, -1, 0.5, Number.NaN, 1.5]) {
+      const { deps: d, calls } = deps([embedding(3)]);
+      await expect(
+        embedTexts(["a"], { ...CONFIG, batchSize }, { deps: d }),
+      ).rejects.toBeInstanceOf(RangeError);
+      expect(calls).toHaveLength(0);
+    }
+  });
+
   it("rejects a malformed response rather than returning junk vectors", async () => {
     const { deps: d } = deps([
       Response.json({ data: [{ index: 0, embedding: true }] }),
@@ -135,6 +154,26 @@ describe("embedTexts", () => {
   });
 });
 
+describe("EmbedConfigSchema", () => {
+  const base = { baseURL: "https://embed.example/v1", model: "m" };
+
+  it("accepts an integer batchSize >= 1 and an omitted batchSize", () => {
+    expect(EmbedConfigSchema({ ...base, batchSize: 1 })).toEqual({
+      ...base,
+      batchSize: 1,
+    });
+    expect(EmbedConfigSchema(base)).toEqual(base);
+  });
+
+  it("rejects a fractional, non-integer, or sub-1 batchSize", () => {
+    for (const batchSize of [0, -1, 0.5, Number.NaN, 1.5]) {
+      expect(EmbedConfigSchema({ ...base, batchSize })).toBeInstanceOf(
+        type.errors,
+      );
+    }
+  });
+});
+
 describe("retry behaviour inherited from the inference policy", () => {
   it("retries a 429 and succeeds on the follow-up", async () => {
     // The hand-rolled client this replaced had no retry at all: a single 429

src/embed.ts

@@ -6,7 +6,7 @@ import {
   runJSONRequest,
   type RequestDependencies,
   type RetryAfterExtractor,
-} from "./request";
+} from "./request.js";
 
 /**
  * `/v1/embeddings` is the one wire format worth supporting — OpenAI, Ollama,
@@ -30,7 +30,7 @@ export const EmbedConfigSchema = type({
    * Inputs per request. Providers cap this — OpenAI rejects arrays over 2048 —
    * and a smaller batch is also how you stay under a per-request token ceiling.
    */
-  "batchSize?": "number > 0",
+  "batchSize?": "number.integer >= 1",
   "timeoutMs?": "number > 0",
 });
 export type EmbedConfig = typeof EmbedConfigSchema.infer;
@@ -165,8 +165,15 @@ export async function embedTexts(
 ): Promise<number[][]> {
   if (texts.length === 0) return [];
 
+  const batchSize = config.batchSize ?? DEFAULT_BATCH_SIZE;
+  if (!Number.isInteger(batchSize) || batchSize < 1) {
+    throw new RangeError(
+      `batchSize must be an integer >= 1, got ${batchSize}: a batch size below 1 would never advance through the input`,
+    );
+  }
+
   const vectors: number[][] = [];
-  for (const batch of batches(texts, config.batchSize ?? DEFAULT_BATCH_SIZE)) {
+  for (const batch of batches(texts, batchSize)) {
     const request = buildRequest(config, batch);
     const body = await runJSONRequest(request, {
       deps: options.deps,

src/index.ts

@@ -4,11 +4,11 @@ export {
   EmbedConfigSchema,
   type EmbedConfig,
   type EmbedOptions,
-} from "./embed";
+} from "./embed.js";
 export {
   runJSONRequest,
   extractRetryAfterMs,
   ModelRequestError,
   type RetryAfterExtractor,
   type RunRequestOptions,
-} from "./request";
+} from "./request.js";