refactor(db): Prisma → Drizzle — the fleet has one ORM

.env.example

@@ -1,2 +1,2 @@
-# Postgres connection for Prisma. Copy to .env and fill in.
+# Postgres connection for Drizzle. Copy to .env and fill in.
 DATABASE_URL="postgresql://user:password@localhost:5432/biaslens"

.github/workflows/ci.yml

@@ -1,10 +1,10 @@
 # Golden CI floor — see dotfiles/templates/ci/README.md.
-# biaslens: npm + vitest + Prisma. Floor = verify (lint + typecheck + test);
-# the 14-case suite now runs on every push & PR. Prisma has no postinstall, so
-# the client is generated explicitly before typecheck needs its types. `next build`
-# is hermetic (no live DB): all DB access is inside dynamic route handlers, so the
-# prerender stage never connects — verified locally against an unreachable DB. It
-# runs after verify to catch build-only breakage.
+# biaslens: npm + vitest + Drizzle. Floor = verify (lint + typecheck + test);
+# the 14-case suite runs on every push & PR. Drizzle needs no codegen — types
+# derive from src/lib/db/schema.ts directly, so nothing runs between install and
+# verify. `next build` is hermetic (no live DB): all DB access is inside dynamic
+# route handlers and the client pool is lazy, so the prerender stage never
+# connects. It runs after verify to catch build-only breakage.
 name: CI
 
 on:
@@ -35,15 +35,12 @@ jobs:
       - name: Install
         run: npm ci
 
-      - name: Prisma generate (types for typecheck)
-        run: npx prisma generate
-
       # SSOT: lint + typecheck + test, defined once in package.json `verify`.
       - name: Verify
         run: npm run verify
 
       # Build gate: catches build-only breakage that verify (lint+typecheck+test)
       # can miss. Hermetic — no live DB needed (all queries are in dynamic route
-      # handlers). Reuses the client generated by the prisma-generate step above.
+      # handlers; the Drizzle client initializes lazily on first request).
       - name: Build
         run: npm run build

.gitignore

@@ -6,6 +6,5 @@ build/
 .env*.local
 *.log
 .DS_Store
-/prisma/*.db
 next-env.d.ts
 *.tsbuildinfo

.prettierignore

@@ -29,3 +29,5 @@ yarn.lock
 # is where it is most opinionated and least useful, and it would bury the real
 # diff. Remove this line when you want docs formatted too.
 *.md
+# drizzle-kit generated migration snapshots — machine-written, not hand-formatted
+drizzle/meta/

AGENTS.md

@@ -6,7 +6,7 @@ Media bias analysis engine. See `CLAUDE.md` for the product contract and
 ## Stack
 
 - Next.js (App Router) + TypeScript (strict)
-- Prisma / PostgreSQL — `prisma/schema.prisma` is the SSOT for domain types
+- Drizzle ORM / PostgreSQL — `src/lib/db/schema.ts` is the SSOT for domain types
 - Vitest for tests
 - Node >= 20
 
@@ -15,8 +15,8 @@ Media bias analysis engine. See `CLAUDE.md` for the product contract and
 ```bash
 npm install
 cp .env.example .env        # set DATABASE_URL
-npm run db:generate         # prisma generate — run before typecheck/build
-npm run db:migrate          # prisma migrate dev (needs a live Postgres)
+npm run db:generate         # drizzle-kit generate — new migration from schema changes
+npm run db:migrate          # drizzle-kit migrate (needs a live Postgres)
 npm run dev                 # local dev server
 npm run build               # next build (production compile; no live DB needed)
 npm run verify              # SSOT gate: lint + typecheck + test
@@ -28,16 +28,19 @@ npm run verify              # SSOT gate: lint + typecheck + test
 "green". CI calls it verbatim; run it locally before declaring any change done.
 `next build` also runs in CI after verify to catch build-only breakage.
 
-Prisma has no `postinstall`, so run `npm run db:generate` (or `npx prisma
-generate`) before `typecheck`/`build` — the generated client provides the types.
+Drizzle has no codegen: types derive from `src/lib/db/schema.ts` at typecheck
+time, so `npm install` is the only prerequisite for `typecheck`/`build`.
+`npm run db:generate` is only needed after changing the schema (it writes a new
+SQL migration into `drizzle/`).
 
 ## Layout
 
 ```
-prisma/schema.prisma   SSOT for domain types (Outlet, Article, Claim, Evidence, Narrative, ...)
+src/lib/db/schema.ts   SSOT for domain types (Outlet, Article, Claim, Evidence, Narrative, ...)
+drizzle/               SQL migrations generated by drizzle-kit
 src/app/               App Router pages + API route handlers
 src/lib/domain/        pure scoring logic (explainable, reproducible) + tests
-src/lib/db/prisma.ts   the single PrismaClient (SSOT for DB access)
+src/lib/db/client.ts   the single Drizzle client (SSOT for DB access)
 src/lib/api/           HTTP result shape
 src/lib/config/        verification config
 ```

CLAUDE.md

@@ -13,9 +13,9 @@ contract every change follows. It mirrors the FleetCrown project brief.
 
 ## Conventions
 
-- Next.js (App Router) + TypeScript strict + Prisma/Postgres.
-- `prisma/schema.prisma` is the SSOT for domain types — derive TS types from it,
-  never redefine them.
+- Next.js (App Router) + TypeScript strict + Drizzle/Postgres.
+- `src/lib/db/schema.ts` is the SSOT for domain types — derive TS types from it
+  (`$inferSelect`), never redefine them.
 - Facts live in `Claim`/`Evidence`; framing lives in `Narrative`/`Frame`. Do not
   collapse the two.
 - Scores are functions of stored evidence — never hard-coded — so they recompute
@@ -29,9 +29,9 @@ pass. (FleetCrown's DoD gate judges against this.)
 
 ## Roadmap (current milestone first)
 
-1. **Core data model** — seeded in `prisma/schema.prisma` (Outlet, Article, Claim,
-   Evidence, Narrative, EditorialDna, HomepageSnapshot). Acceptance: `prisma
-   migrate dev` runs clean and the client generates.
+1. **Core data model** — seeded in `src/lib/db/schema.ts` (Outlet, Article, Claim,
+   Evidence, Narrative, EditorialDna, HomepageSnapshot). Acceptance: `npm run
+   db:migrate` runs clean against a fresh Postgres.
 2. Extraction + Claim agents: parse an article into structured claims with sources.
 3. Bias + Evidence scoring engine — explainable scores.
 4. Hourly Homepage Loop: crawl + compare outlets.

README.md

@@ -25,7 +25,7 @@ Argument Arena vision.
 
 ## Stack
 
-Next.js + TypeScript + Prisma/Postgres. Multi-agent pipeline for
+Next.js + TypeScript + Drizzle/Postgres. Multi-agent pipeline for
 crawl/extract/score. FleetCrown for orchestration + loops; OrangeCat for funding
 + public transparency.
 
@@ -40,8 +40,9 @@ and on-demand Article Loop (extract → analyze → report → feedback).
 ## Roadmap
 
 1. ~~**Define the core data model** — Outlet, Article, Claim, Evidence, Narrative,
-   Editorial DNA as Prisma schemas (this scaffold seeds `prisma/schema.prisma`).~~
-   **✅ DONE** (`0d46066`) — App Router root + initial Prisma migration applied;
+   Editorial DNA as schema definitions (now `src/lib/db/schema.ts`, Drizzle ORM —
+   migrated from the original Prisma scaffold with byte-identical tables).~~
+   **✅ DONE** (`0d46066`) — App Router root + initial migration applied;
    build, lint, and `tsc` all green.
 2. Build the Extraction + Claim agents to parse an article into structured claims.
 3. Bias + Evidence scoring engine — every score explainable.
@@ -54,7 +55,7 @@ and on-demand Article Loop (extract → analyze → report → feedback).
 ```bash
 npm install
 cp .env.example .env   # set DATABASE_URL
-npx prisma migrate dev
+npm run db:migrate     # drizzle-kit migrate — applies drizzle/*.sql
 npm run dev
 ```
 

drizzle.config.ts

@@ -0,0 +1,19 @@
+/**
+ * Drizzle Kit configuration — fleet house pattern (see hirnli, vitareba).
+ *
+ * `src/lib/db/schema.ts` is the SSOT for domain types; migrations are generated
+ * into `drizzle/` with `npm run db:generate` and applied with `npm run
+ * db:migrate`. DATABASE_URL comes from the environment (no dotenv magic —
+ * export it or prefix the command).
+ */
+
+import { defineConfig } from 'drizzle-kit';
+
+export default defineConfig({
+  schema: './src/lib/db/schema.ts',
+  out: './drizzle',
+  dialect: 'postgresql',
+  dbCredentials: {
+    url: process.env.DATABASE_URL!,
+  },
+});

drizzle/0000_nosy_thor_girl.sql

@@ -0,0 +1,78 @@
+CREATE TABLE "Article" (
+	"id" text PRIMARY KEY NOT NULL,
+	"outletId" text NOT NULL,
+	"headline" text NOT NULL,
+	"author" text,
+	"section" text,
+	"url" text NOT NULL,
+	"body" text NOT NULL,
+	"publishedAt" timestamp (3),
+	"createdAt" timestamp (3) DEFAULT now() NOT NULL
+);
+--> statement-breakpoint
+CREATE TABLE "Claim" (
+	"id" text PRIMARY KEY NOT NULL,
+	"articleId" text NOT NULL,
+	"text" text NOT NULL,
+	"speaker" text,
+	"verificationStatus" text DEFAULT 'pending' NOT NULL,
+	"confidence" double precision,
+	"createdAt" timestamp (3) DEFAULT now() NOT NULL
+);
+--> statement-breakpoint
+CREATE TABLE "EditorialDna" (
+	"id" text PRIMARY KEY NOT NULL,
+	"outletId" text NOT NULL,
+	"dimensions" jsonb NOT NULL,
+	"updatedAt" timestamp (3) NOT NULL
+);
+--> statement-breakpoint
+CREATE TABLE "Evidence" (
+	"id" text PRIMARY KEY NOT NULL,
+	"claimId" text NOT NULL,
+	"stance" text NOT NULL,
+	"source" text NOT NULL,
+	"excerpt" text,
+	"url" text,
+	"createdAt" timestamp (3) DEFAULT now() NOT NULL
+);
+--> statement-breakpoint
+CREATE TABLE "HomepageSnapshot" (
+	"id" text PRIMARY KEY NOT NULL,
+	"outletId" text NOT NULL,
+	"capturedAt" timestamp (3) DEFAULT now() NOT NULL,
+	"items" jsonb NOT NULL
+);
+--> statement-breakpoint
+CREATE TABLE "Narrative" (
+	"id" text PRIMARY KEY NOT NULL,
+	"articleId" text NOT NULL,
+	"frame" text NOT NULL,
+	"summary" text,
+	"createdAt" timestamp (3) DEFAULT now() NOT NULL
+);
+--> statement-breakpoint
+CREATE TABLE "Outlet" (
+	"id" text PRIMARY KEY NOT NULL,
+	"name" text NOT NULL,
+	"country" text,
+	"ownership" text,
+	"funding" text,
+	"isPublic" boolean DEFAULT false NOT NULL,
+	"createdAt" timestamp (3) DEFAULT now() NOT NULL
+);
+--> statement-breakpoint
+ALTER TABLE "Article" ADD CONSTRAINT "Article_outletId_fkey" FOREIGN KEY ("outletId") REFERENCES "public"."Outlet"("id") ON DELETE cascade ON UPDATE cascade;--> statement-breakpoint
+ALTER TABLE "Claim" ADD CONSTRAINT "Claim_articleId_fkey" FOREIGN KEY ("articleId") REFERENCES "public"."Article"("id") ON DELETE cascade ON UPDATE cascade;--> statement-breakpoint
+ALTER TABLE "EditorialDna" ADD CONSTRAINT "EditorialDna_outletId_fkey" FOREIGN KEY ("outletId") REFERENCES "public"."Outlet"("id") ON DELETE cascade ON UPDATE cascade;--> statement-breakpoint
+ALTER TABLE "Evidence" ADD CONSTRAINT "Evidence_claimId_fkey" FOREIGN KEY ("claimId") REFERENCES "public"."Claim"("id") ON DELETE cascade ON UPDATE cascade;--> statement-breakpoint
+ALTER TABLE "HomepageSnapshot" ADD CONSTRAINT "HomepageSnapshot_outletId_fkey" FOREIGN KEY ("outletId") REFERENCES "public"."Outlet"("id") ON DELETE cascade ON UPDATE cascade;--> statement-breakpoint
+ALTER TABLE "Narrative" ADD CONSTRAINT "Narrative_articleId_fkey" FOREIGN KEY ("articleId") REFERENCES "public"."Article"("id") ON DELETE cascade ON UPDATE cascade;--> statement-breakpoint
+CREATE UNIQUE INDEX "Article_url_key" ON "Article" USING btree ("url");--> statement-breakpoint
+CREATE INDEX "Article_outletId_idx" ON "Article" USING btree ("outletId");--> statement-breakpoint
+CREATE INDEX "Claim_articleId_idx" ON "Claim" USING btree ("articleId");--> statement-breakpoint
+CREATE UNIQUE INDEX "EditorialDna_outletId_key" ON "EditorialDna" USING btree ("outletId");--> statement-breakpoint
+CREATE INDEX "Evidence_claimId_idx" ON "Evidence" USING btree ("claimId");--> statement-breakpoint
+CREATE INDEX "HomepageSnapshot_outletId_capturedAt_idx" ON "HomepageSnapshot" USING btree ("outletId","capturedAt");--> statement-breakpoint
+CREATE INDEX "Narrative_articleId_idx" ON "Narrative" USING btree ("articleId");--> statement-breakpoint
+CREATE UNIQUE INDEX "Outlet_name_key" ON "Outlet" USING btree ("name");