fix(sync): fold re-keyed upstream duplicates into their existing id (#124)

* fix(sync): fold re-keyed upstream duplicates into their existing id

LiteLLM started listing Mistral Medium 3.5 under a second key that
differs only by punctuation (a dot instead of a dash). mergeCatalog
treated it as a brand-new model, and its id collided with the already-
published one once both were turned into a URL slug -- crashing
check-pages and failing every Sync pricing run on main since.

Reconcile any freshly-matched feed id that isn't already published
against the slug of ids we do publish, and fold it into the existing
id instead of minting a second, colliding one.

Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>

* docs: update published test counts for the new regression test

The new pipeline.test.ts case bumps this package to 434 tests and the
published total to 997 -- update the README badge, the README docs
table, and docs/TESTING.md's total and per-file row to match, per
check:test-badge.

Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>

* chore(mobile): align Expo SDK patch dependencies

expo, expo-constants and expo-font had drifted a patch version behind
what the installed Expo SDK 57 expects, tripping expo-doctor in the
mobile verify job -- unrelated to this branch's pricing-sync fix, but
failing on main regardless. `expo install --fix` resolves it; all 21
expo-doctor checks, the mobile test suite (115/115), and the iOS/
Android/web export all pass afterward.

Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>

---------

Co-authored-by: Claude Sonnet 5 <noreply@anthropic.com>

Author: AndrewAvery7

README.md

@@ -14,7 +14,7 @@
   <a href="LICENSE"><img src="https://img.shields.io/badge/license-MIT-green.svg" alt="MIT license"></a>
   <img src="https://img.shields.io/badge/models-74-2456E6.svg" alt="74 models tracked">
   <img src="https://img.shields.io/badge/providers-12-2456E6.svg" alt="12 providers">
-  <img src="https://img.shields.io/badge/tests-996-blue.svg" alt="996 tests">
+  <img src="https://img.shields.io/badge/tests-997-blue.svg" alt="997 tests">
   <img src="https://img.shields.io/badge/initial%20payload-87%20KB%20gzip-blue.svg" alt="87 KB gzip initial payload">
   <a href="https://github.com/AndrewAvery7/promptspend/actions/workflows/ci.yml"><img src="https://github.com/AndrewAvery7/promptspend/actions/workflows/ci.yml/badge.svg" alt="CI"></a>
   <a href="https://github.com/AndrewAvery7/promptspend/actions/workflows/sync-pricing.yml"><img src="https://github.com/AndrewAvery7/promptspend/actions/workflows/sync-pricing.yml/badge.svg" alt="Sync pricing"></a>
@@ -377,7 +377,7 @@ there is a `Ctrl`/`Cmd`+`K` command palette.
 | Document                                               | What is in it                                                                                            |
 | ------------------------------------------------------ | -------------------------------------------------------------------------------------------------------- |
 | [docs/ARCHITECTURE.md](docs/ARCHITECTURE.md)           | How the pipeline, the engine and the state layer work, and **why** each is shaped that way               |
-| [docs/TESTING.md](docs/TESTING.md)                     | What the 996 tests cover, the uneven coverage thresholds, and what the suite deliberately does not cover |
+| [docs/TESTING.md](docs/TESTING.md)                     | What the 997 tests cover, the uneven coverage thresholds, and what the suite deliberately does not cover |
 | [docs/TROUBLESHOOTING.md](docs/TROUBLESHOOTING.md)     | "The estimate does not match my bill", flagged prices, missing models, running it locally                |
 | [docs/PAGES.md](docs/PAGES.md)                         | The 163 generated pages: what is built, why the comparison set is curated, and the IndexNow pipeline     |
 | [docs/API.md](docs/API.md)                             | The public pricing API on `promptspend.dev` — endpoints, why it fetches rather than bundles, going live  |

apps/mobile/package-lock.json

@@ -8,7 +8,7 @@
     "@expo/vector-icons": "^15.0.2",
     "@promptspend/core": "file:../../packages/core",
     "@react-native-async-storage/async-storage": "2.2.0",
-    "expo": "~57.0.17",
+    "expo": "~57.0.18",
     "expo-clipboard": "~57.0.1",
     "expo-constants": "~57.0.12",
     "expo-file-system": "~57.0.4",

apps/mobile/package.json

@@ -15,7 +15,7 @@ to publish past it.
 
 ## What the suite covers
 
-**996 tests in all**: 433 unit and integration tests in this package, 115 in
+**997 tests in all**: 434 unit and integration tests in this package, 115 in
 `apps/mobile/`, 108 in `worker/`, 44 in `api/`, 47 in `mcp/`, 137 in `vscode/`,
 and 112 browser tests across four viewports. Those seven figures are the whole suite and they sum to the
 total. Two earlier revisions did not: one claimed 533, which was neither the sum
@@ -47,7 +47,7 @@ anyone adding them up.
 
 | Suite                                  | Tests | Guards                                                                                                                                                                                                                                                                                                                                                        |
 | -------------------------------------- | ----- | ------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- |
-| `scripts/lib/pipeline.test.ts`         | 62    | The whole pipeline against fixtures: the trust ladder, both sanity thresholds, cold start, stale retention, flag-once semantics — including a standing flag whose _figures_ drift, which is what re-opened a pull request every morning — full-field diffing, coverage versus a price move versus a source correction, the content hash, changelog rendering. |
+| `scripts/lib/pipeline.test.ts`         | 63    | The whole pipeline against fixtures: the trust ladder, both sanity thresholds, cold start, stale retention, flag-once semantics — including a standing flag whose _figures_ drift, which is what re-opened a pull request every morning — full-field diffing, coverage versus a price move versus a source correction, the content hash, changelog rendering. |
 | `scripts/lib/readme-badges.test.ts`    | 3     | The daily catalog sync updates both badge URLs and accessible alt text, leaves unrelated prose untouched, and fails closed if a README redesign removes or duplicates a derived claim.                                                                                                                                                                        |
 | `src/lib/contrast.test.ts`             | 49    | Every accent × theme × canvas combination against every surface it can appear on, plus the status colours on their own soft fills — the pairing the accent matrix does not reach.                                                                                                                                                                             |
 | `src/lib/engine/cost.test.ts`          | 44    | Compounding history, per-turn long-context tiers, cache reads and writes, promotional windows, reasoning multipliers, margins, break-even, and the scenarios that cannot physically run.                                                                                                                                                                      |

docs/TESTING.md

@@ -20,6 +20,7 @@
  */
 import type { Model, PricingCatalog } from '../../src/lib/pricing/types';
 import { SCHEMA_VERSION } from '../../src/lib/pricing/types';
+import { modelSlug } from '../../src/lib/seo/slug';
 import { pricingChanged } from './diff';
 import {
   comparisonKey,
@@ -105,7 +106,20 @@ export function mergeCatalog(input: MergeInput): MergeResult {
   }
   const overrideById = new Map(overrides.map((o) => [o.id, o]));
   const previousById = new Map((previous?.models ?? []).map((m) => [m.id, m]));
-  const feedById = new Map(litellm.map((r) => [r.id, r]));
+
+  // Upstream occasionally re-keys a model with different punctuation (a dot
+  // where a dash used to be, or vice versa) without it being a new model at
+  // all. Ids are permanent once published (see src/lib/seo/slug.ts), so when
+  // a feed row we don't already publish collides on URL slug with one we do,
+  // treat it as that same model rather than minting a second, colliding id.
+  const publishedIdBySlug = new Map<string, string>();
+  for (const id of previousById.keys()) publishedIdBySlug.set(modelSlug(id), id);
+  const reconciledLitellm = litellm.map((rate) => {
+    if (previousById.has(rate.id)) return rate;
+    const knownId = publishedIdBySlug.get(modelSlug(rate.id));
+    return knownId ? { ...rate, id: knownId } : rate;
+  });
+  const feedById = new Map(reconciledLitellm.map((r) => [r.id, r]));
   const review: ReviewItem[] = [];
   const corrections: CorrectionItem[] = [];
   const staleIds: string[] = [];

scripts/lib/merge.ts

@@ -257,6 +257,70 @@ describe('mergeCatalog — the trust ladder', () => {
     expect(warm.review.map((item) => item.id)).toEqual(['moonshot-kimi-k2.6']);
   });
 
+  it('folds a re-keyed upstream duplicate into the existing id instead of colliding on URL slug', () => {
+    // LiteLLM occasionally lists the same model under a differently-punctuated
+    // key (a dot where a dash used to be). Both are valid catalog ids on their
+    // own, but they collapse to the identical URL slug — assertUniqueSlugs in
+    // src/lib/seo/slug.ts is what catches that at build time — so a same-priced
+    // dot-form arriving later must be folded into the id already published,
+    // not minted as a second "new" model.
+    const mistralAllowlist: Allowlist = {
+      ...ALLOWLIST,
+      providers: [...ALLOWLIST.providers, { id: 'mistral', name: 'Mistral AI', country: 'FR' }],
+      families: [
+        ...ALLOWLIST.families,
+        {
+          id: 'mistral',
+          providerId: 'mistral',
+          include: ['^mistral/mistral-medium-\\d(\\.\\d|-\\d)?$'],
+          stripPrefix: 'mistral/',
+          tokenizer: { kind: 'approx', charsPerToken: 3.7, cjkCharsPerToken: 1.8 },
+          capabilities: { reasoning: false, vision: true },
+        },
+      ],
+    };
+    const previous: PricingCatalog = {
+      schemaVersion: SCHEMA_VERSION,
+      generatedAt: '2026-08-01T06:00:00.000Z',
+      providers: mistralAllowlist.providers,
+      models: [
+        {
+          id: 'mistral-mistral-medium-3-5',
+          providerId: 'mistral',
+          displayName: 'Mistral Medium 3.5',
+          status: 'current',
+          contextWindow: 262_144,
+          pricing: { input: 1.5, output: 7.5 },
+          tokenizer: { kind: 'approx', charsPerToken: 3.7, cjkCharsPerToken: 1.8 },
+          capabilities: { reasoning: false, vision: true },
+          provenance: { source: 'litellm', lastVerified: '2026-08-01' },
+        },
+      ],
+    };
+    const feed = fromLiteLLM(
+      {
+        'mistral/mistral-medium-3.5': {
+          mode: 'chat',
+          input_cost_per_token: 1.5e-6,
+          output_cost_per_token: 7.5e-6,
+        },
+      },
+      mistralAllowlist,
+    );
+    const { catalog, review } = mergeCatalog({
+      litellm: feed,
+      openrouter: new Map(),
+      allowlist: mistralAllowlist,
+      overrides: [],
+      previous,
+      generatedAt,
+    });
+    const mistralModels = catalog.models.filter((m) => m.providerId === 'mistral');
+    expect(mistralModels).toHaveLength(1);
+    expect(mistralModels[0]!.id).toBe('mistral-mistral-medium-3-5');
+    expect(review).toEqual([]);
+  });
+
   it('keeps a previously published model when the feed stops listing it', () => {
     // One truncated upstream response used to be enough to delete most of the
     // catalog — and because removals did not trip the review rule, the deletion