Cold Case

Cold Case is a self-generating noir detective game built end-to-end using Codex with GPT-5.6. The description states it produces unique murder mysteries from shareable seeds, ensuring solvability through backward generation of cases (killer first, then alibis, contradictions). It includes features like evidence boards, timeline auto-flagging, and suspect interrogation. The game is fully offline, single-file HTML, and designed to be playable without backend or dependencies.

The project appears to be a prototype or proof-of-concept built in one session by two developers. No revenue, customers, or traction are evidenced. The description claims the game guarantees solvability and fairness, but these claims are unverified.

The description states that Cold Case is a "noir murder mystery that generates itself." It builds cases from shareable seeds so that the same seed always produces the same crime. Features include:

• Searching rooms for evidence
• Interrogating suspects
• Pinning cards to an evidence board with red string
• Timeline auto-flagging contradictions
• Accusation with proof
• Three wrong guesses and the culprit walks

It is built using Codex + GPT-5.6, with no scaffolding, templates, or dependencies. The result is a single HTML file (~40KB) that runs offline in any browser.

Inference The product is a procedural game generator designed to produce fair, solvable mysteries programmatically.

The description states the inspiration was to create a detective game where "once you finish, it's over" — and they wanted one that could give players a new murder every time they opened it. The key claim is that every case is provably solvable before being seen, with no dead ends or gotchas.

They also state that the game guarantees fairness by generating cases backwards: killer first, then alibis, then contradictions. This approach ensures that each case has a unique logical solution.

Inference The positioning is as a procedural mystery generator with an emphasis on solvability and fairness — not just variety.

The description does not state who the target customer or ideal customer profile (ICP) is. It only describes the game's mechanics and generation process.

Not evidenced

There is no evidence of a business model or pricing structure in the description. The project is presented as a hackathon submission with no mention of monetization, licensing, or distribution plans.

Not evidenced

The description states that Cold Case was built end-to-end using Codex with GPT-5.6, and that it required no scaffolding, templates, or dependencies. It is a single HTML file (~40KB), runs offline, and has no backend or APIs.

Key technical details include:

• Backward case generation (killer first)
• Verification step (verifySolvable()) to ensure solvability
• Dialogue system with six suspect personalities
• Evidence board with draggable cards and red string
• Timeline auto-flagging contradictions
• Deterministic randomness via seeded PRNG

Inference The technical approach shows a high degree of automation using AI tools, but lacks evidence of scalability or production-grade infrastructure.

The description states that the team stress-tested thousands of generated cases across every difficulty and claims zero unsolvable cases. It also says it's "genuinely shippable in one session" and includes a save system, title screen, and accuse-and-resolve loop.

However, there is no evidence of actual users, downloads, or adoption beyond the authors' own testing.

Not evidenced

The description references detective games like Ace Attorney, Return of the Obra Dinn, and Sherlock Holmes puzzle books. It positions Cold Case as a procedural mystery generator that avoids the "fatal flaw" of being over once played.

It also mentions that the hard part of procedural games isn't generating variety, but generating fairness — implying it's competing with other procedural or puzzle games in terms of logic and design.

Not evidenced

• Unverified claims: The description states that every case is provably solvable, but this has not been independently verified.
• No traction or user data: There are no signs of actual players or usage beyond the authors’ own testing.
• Prototype nature: The project is described as a hackathon submission and a one-session build — not a mature product.
• Limited scope: It currently supports only 8 settings and 24 weapons, with no indication of expansion plans.
• AI dependency: Heavy reliance on Codex + GPT-5.6 may create fragility if those tools change or become unavailable.

1. How many actual test cases were generated and verified? What was the process for ensuring solvability?
2. Is there any evidence of user feedback or playtesting beyond internal testing?
3. What is the long-term vision for the product — is it intended to be a commercial game, a tool, or something else?
4. How would you scale this approach beyond a single developer or small team?
5. Are there plans to monetize or distribute the game beyond the current prototype?