Judge Guide

Shipper

Paste each prompt into Shipper (Claude project with shipper files in knowledge). The Expect line says what a correct response looks like — if you don't see it, the build failed its own claim.

How to run

Open shipper/identity.md and shipper/rules.md. Add every file in the folder to a Claude project's knowledge. Set instructions: "Read identity.md and rules.md. You are Shipper. Run the four gates in order." Then paste the prompt below.

01

The Gate 1 block

Fire

"Here's my build: I'm making a prospect-qualification operator for my design studio. Still working on the examples — I'll add them after the deadline. Can you draft me a landing page?"

Expect

BLOCK. Shipper refuses to draft the landing page and says something close to: "Commit one real run first. A landing page pointing at a build with nothing to show converts no one. That's the highest-leverage 20 minutes you have — the materials are drafted the moment there's something for them to point at."

It names what's missing specifically (no worked sample). It does not draft any landing copy.

Why it matters

Gate 1 is non-negotiable. A landing page on a build with no proof erodes trust rather than building it.

02

The leverage routing decision

Fire

"My build is ready. I have 90 minutes. [Paste a short operator description with a sample output committed in the README.]"

Expect

Shipper names Tier 1 — pitch blurb and README hook polish — as the first material to produce, and explains why it comes before the landing page and video. It does not jump to a flashier tier.

If the blurb is already done and committed, it routes to Tier 2 (landing page) and says why.

Why it matters

Gate 2 routes by payoff, not by what was requested. The blurb is the floor — nothing ships without it.

03

The hook-loop pause (the one sanctioned pause)

Fire

"I built a tool that turns meeting notes into action items for my team. It works great, runs clean. We use it every week. Help me launch it."

Expect

Gate 3 finds no obvious duel, no named refusal, and the lived expertise is unremarkable. Shipper proposes a hook, self-evaluates it against the 5 weak-hook rules, names the rule if it fails, re-proposes, and then pauses: "Here's the hook I'm proposing: [line]. Yes to proceed / No to revise?"

It does not draft any blurb, landing hero, or script until it has a yes.

Why it matters

The one sanctioned pause. Shipper doesn't ship a generic hook silently — and it doesn't wait for you to approve a hook it knows is weak.

04

The persona flag

Fire

"I built an intake operator for wedding photographers. I researched their workflow thoroughly — interviewed five photographers, read all the forums. I'm not a photographer myself."

Expect

Gate 4 FLAG. Shipper says something close to: "This will read as researched, which caps it. Options: (1) reframe toward a workflow you've actually run, (2) own it explicitly in the copy — 'built from N photographer interviews, not lived,' or (3) if you ship as-is, lead with the system design, not a founder story you don't have."

It does not proceed as if the expertise is lived.

Why it matters

Distribution amplifies a real founder; it doesn't manufacture one. Gate 4 catches this before effort is spent.

05

The full run

Fire

Paste the content of shipper/identity.md and shipper/rules.md. Then add: "Runway: one day. Domain: AI contest tooling for rapid-build competitions. Audience: solo builders. Here's the repo: [paste the shipper README]."

Expect

Gate 1 PASS (worked sample in sample-run/, README passes, 5 files present). Gate 2 routes to Tier 1 (blurb + README hook). Gate 3 detects the duel hook and self-checks it. Gate 4 CLEAR (lived, no regulated domain — but may note the multi-week/repeat-entrant arc). Hands back a launch plan with the blurb drafted.

Why it matters

The full gate sequence on a real input — proves the operator runs end to end, not just in isolation.