Triad — Benchmark Intelligence

The standing record of who is fastest, cheapest, and most correct in AI.

Triad is a three-layer benchmark intelligence service for AI engineers. We aggregate LLM, Agent, and Harness benchmarks across sources, normalise them into a single cross-source scoreboard, and publish a footnoted weekly digest plus on-demand custom slices.

• Live: benchmark-intel.prin7r.com
• Notion opportunity: LLM benchmark reports — Wave 2 scope expansion
• GitHub: prin7r-projects/benchmark-intel
• Brand: Triad / IBM Plex / paper-ink-signal-cinnabar — see DESIGN.md

What's in here (Wave 2 batch 1)

Path	Purpose
DESIGN.md	Canonical design + style guide (15 sections, ShadCN-first, screenshots)
docs/01-brand-identity.md … docs/10-pitch-deck.md	The ten strategy/design docs
docs/pitch-deck.html	Self-contained 10-slide pitch deck (open in any browser)
apps/landing/	Next.js 15 + Tailwind landing site, hand-authored
apps/app/	Stub for the Wave 3 authenticated reader app
Dockerfile.landing	Multistage Next.js standalone build for production
docker-compose.yml	Single landing service with Traefik labels, env_file: .env
.env.example	Public surface for env variable names; live values live on storage-contabo
docs/screenshots/	Production screenshots of the landing (desktop + mobile)

Three layers

• Layer 1 — LLMs: Anthropic, OpenAI, Google, Meta, xAI, DeepSeek, Mistral, Qwen, open-weights. Tracked across ArtificialAnalysis, lmsys arena, HELM.
• Layer 2 — Agents: SWE-bench, GAIA, AgentBench, terminal-bench, OSWorld, plus product-mode agents (Claude Code, Codex CLI, Cursor agent, Aider, Devin-class).
• Layer 3 — Harnesses: same model + same benchmark across different scaffolds (Aider vs. Claude Code vs. Cursor vs. plain SDK vs. custom). The layer no public site measures.

Cross-source aggregation: every score lands in the canonical {model_id, harness_id, benchmark_id, score, score_kind, n, ci, source_url, retrieved_at} row, and the cross-source aggregator publishes a layer-aware percentile rank.

Quickstart (landing, locally)

cd apps/landing
pnpm install
cp ../../.env.example .env.local      # NOWPayments keys optional locally; mock checkout works without
pnpm dev                              # http://localhost:3000

Deploy

Deploy target: storage-contabo (root@161.97.99.120). Reverse proxy: Dokploy / Traefik (host network, ACME letsencrypt). DNS: wildcard *.prin7r.com.

ssh storage-contabo
mkdir -p /opt/prin7r-deploys/benchmark-intel
cd /opt/prin7r-deploys/benchmark-intel
git clone https://github.com/prin7r-projects/benchmark-intel.git .
# Populate /opt/prin7r-deploys/benchmark-intel/.env with NOWPayments keys
docker compose build
docker compose up -d
curl -sI https://benchmark-intel.prin7r.com    # → HTTP/2 200

Payments

Triad uses NOWPayments as the default crypto rail (Reader $29/mo, Custom-slices $199/mo). Enterprise tier accepts wire transfer or USDT direct invoice via founder@prin7r.com. See docs/07-sales-strategy.md for full pricing logic.

• apps/landing/app/api/checkout/nowpayments/route.ts — server-side hosted invoice creation
• apps/landing/app/api/webhooks/nowpayments/route.ts — IPN webhook with HMAC-SHA512 verification
• Secrets live on storage-contabo:/opt/prin7r-deploys/benchmark-intel/.env, referenced by env_file: .env in docker-compose.yml. Never committed.
• Reference implementation: /Users/keer/projects/prin7r/payments-prototypes/.

Screenshots

Captured against the deployed URL (https://benchmark-intel.prin7r.com) with headless Chromium.

License

MIT. See LICENSE.