Bench'd

The neutral benchmark for AI memory systems. Every score is independently run, cryptographically signed, and verifiable by anyone. Spend attestation powered by ProofMeter (Patent Pending).

Setup

pnpm install
pnpm dev

Open http://localhost:3000.

What's Built (v1)

Six launch pages with typed mock data:

Page	Route	Description
Homepage	/	Hero, latest verified runs feed, top 5 leaderboard, "Why two scores?" explainer
Leaderboard	/leaderboard	Compact/Detailed views, filters, search, sort, mobile card layout
System Profile	/system/[slug]	Score cards, Scores/History/Failures/Notes tabs, gap analysis traces
Receipt	/receipt/[run_id]	Full manifest JSON, Ed25519 signature, verification widget, share button
Methodology	/methodology	Long-form editorial with sticky TOC, Two-Score Model, Known Controversies
Trust	/trust	Signing keys, tabbed verification code (bash/Python/JS), key rotation log

Global Elements

• Header with nav and Cmd+K button
• Cmd+K command palette (search systems, benchmarks, receipts, pages)
• Sonner toast notifications for copy/verification actions
• Footer with signing key fingerprints and VerifiedState credit

Deferred

v1.5 (4-8 weeks post-launch)

• /benchmark/[slug] — Benchmark detail pages
• /compare/[slug-a]-vs-[slug-b] — Auto-generated comparison pages
• /run — Run benchmark page (BYOK + rate-limited)

v2 (when traffic justifies)

• /submit — Submit new system
• /claim/[system] — Vendor claim flow (DNS TXT, GitHub, OAuth)
• /opt-out — Vendor opt-out form
• /spec — Adapter specification

Mock Data

All data lives in lib/data/ with TypeScript types in lib/types.ts:

File	Contents
lib/data/systems.ts	8 seeded systems (Mem0, Letta, Zep, MemPalace, Mastra, ByteRover, OMEGA, MemMachine)
lib/data/benchmarks.ts	3 benchmarks (LongMemEval, LoCoMo, PersonaMem v2)
lib/data/runs.ts	16 sample runs with full manifests and fake Ed25519 signatures
lib/data/failures.ts	40+ failure traces with conversation histories, queries, expected answers

Each file exports typed arrays and lookup helpers (e.g., getSystemBySlug, getRunById).

Swapping Mock Data for Supabase

Every component receives data as typed props. To connect Supabase:

1. Install @supabase/supabase-js and @supabase/ssr
2. Create lib/supabase.ts with your client
3. Replace the mock data imports in each page's server component:

   - import { systems } from "@/lib/data/systems";
   + import { createClient } from "@/lib/supabase";
   + const supabase = createClient();
   + const { data: systems } = await supabase.from("systems").select("*");

4. Component props stay the same — no component changes needed
5. Types in lib/types.ts map directly to the database schema in the master brief

Design System

Colors

Token	Hex	Usage
Primary/Amber	#FFB800	Verified scores, CTAs, active states, Vendor-Verified tier
Verified Green	#00C87A	Cryptographic verification success only
Listed	#6B7280	Listed tier badge
Self-Reported	#DC2626	Unclaimed Self-Reported tier badge
Community	#3B82F6	Community-Verified tier badge
Partner	#7C3AED	Partner-Audited tier badge

Fonts

Font	CSS Variable	Usage
Inter	--font-sans	UI chrome, headings, labels
Merriweather	--font-serif	Editorial/methodology body copy
JetBrains Mono	--font-mono	Scores, IDs, manifests, code, timestamps

Key Components

Component	Path	Usage
TrustTierBadge	components/bench/trust-tier-badge.tsx	Trust tier with icon + color
ScorePair / ScoreCard	components/bench/score-pair.tsx	Verified + Nuance score display
Sparkline	components/bench/sparkline.tsx	SVG sparkline for score trends
CodeBlock	components/bench/code-block.tsx	Monospace code with copy/download
FailureRow	components/bench/failure-row.tsx	Expandable failure trace row
CommandPalette	components/bench/command-palette.tsx	Cmd+K search overlay

Tech Stack

• Next.js 15 (App Router, Turbopack)
• TypeScript (strict mode)
• Tailwind CSS v4
• shadcn/ui (new-york style)
• pnpm