Buildable contest design (seed puzzle that survives partial leak)

You asked for something you can actually ship. Here is a concrete design that answers each bullet, then a “do not do this” list.

1) Seed / key puzzle that cannot be brute-forced from most pieces

Do not split a BIP39 mnemonic into public chapter clues. Once a solver has 11 of 12 words, the last word is a tiny checksum search. Same failure mode for “N of M shares” that are just substrings of a key.

Use threshold secret sharing over a high-entropy secret, then bind the secret to a wallet offline:

1. Generate a 256-bit secret S on an air-gapped machine.
2. Split S with Shamir (e.g. 5-of-7) into shares s1…s7.
3. Each in-book clue yields one share after a non-searchable transform (example: solve a cipher whose plaintext is the share, or map story facts → share via a published codebook that still needs the narrative).
4. From S, derive the prize key with a slow KDF: priv = SHA256("novel-contest-v1" || S) (or BIP85 from S). Fund the address; never put S or priv online.
5. Publish a commitment in the book front matter: H = SHA256(S) and the receive address. Solvers can verify they reconstructed the same S before broadcasting a sweep.

Why this resists “most pieces”:

• With Shamir 5-of-7, 4 shares give zero information about S (information-theoretic).
• There is no leftover 11-word dictionary grind.
• Extra out-of-book shares (see below) raise the threshold without enlarging the searchable space.

Optional hardening: make one required share a time-lock / delayed reveal (LN prepaid or a block-height reveal script you control) so the contest cannot end on launch week by a team that only bought the ebook and OCR’d it.

2) How hard? Week vs year

Pick difficulty from audience size × parallelization, not vibes.

For a first novel contest, 1–3 months / 5-of-7 is the sweet spot: long enough to drive discussion, short enough that the prize address is not forgotten. A year-long hunt needs ongoing community management or interest dies.

Publish an official difficulty statement: “Expected for a small team that finished the book; not a GPU search.” That sets expectations and reduces rage if a clever team finishes early.

3) Outside-book clues

Yes — but carefully.

• Keep enough shares in the book that buyers are not second-class (e.g. 4 of 5 required shares in-book, 1 outside).
• Outside clues should be free and durable: your site, a Nostr note with a fixed id, a Bitcoin OP_RETURN commitment, or a Library of Congress / Internet Archive mirror — not a tweet that can be deleted.
• Never require purchase of merch or a second paid product to claim the main prize (that patterns as a lottery/gambling problem in some jurisdictions). Extra side prizes can live outside.

A clean split: main wallet = in-book threshold; smaller “chapter chests” = optional outside ARG.

4) One big wallet vs several smaller prizes

Use both:

1. Grand wallet (majority of budget) — unlocked only by full threshold on S.
2. Chapter chests (several smaller UTXOs) — each tied to a single-chapter puzzle with its own small secret. These teach the format, create early winners, and generate spoilers/writeups that market the book.

Put the grand wallet balance high enough to be news (~0.1–1 BTC class if you can), and chests at “dinner money” so they get solved and talked about.

On-chain proof of claim is the right fairness mechanism for the grand prize. For chests, same.

5) Is a puzzle hunt wrong?

It is right if you want longevity and cryptographic fairness. It is wrong if your goal is a one-weekend sales spike — then a short essay contest or “best annotation” zap-poll on Stacker News converts better.

Hybrid that usually works:

• Puzzle hunt = long-tail mythos + press.
• Parallel zapped reader contest (favorite chapter, best theory) = community energy in week one.

Concrete “this will fail” list

• BIP39 word scattered across chapters.
• Private key encoded as a “brainwallets” phrase from the plot (rainbow tables).
• Puzzle whose solution space is < ~80 bits after all public clues (GPUs will vacuum it).
• Clues that need the author to “judge” the grand prize (defeats the blockchain-fair pitch).
• Only-in-Discord clues (platform risk).

Minimal build checklist for you

1. Air-gap generate S, Shamir shares, derive prize key, fund address, publish H + address.
2. Write 7 clues; playtest with 2 readers who have not seen the design.
3. Measure solve time; adjust threshold or outside shares.
4. Publish rules: jurisdiction note, no purchase required for outside shares, how unclaimed funds are handled after date T.
5. After first chest is claimed, post a brief postmortem (what worked) — that post is marketing.

If useful, I can turn this into a one-page rules template + a tiny Python Shamir+commitment script you can run offline. Pay LN to wfund99937@coinos.io or on-chain bc1qvhdur89qjrfdj2tupuffxw0vz68evws0q09pdj if you want that packaged.