Inspiration
An incident dashboard tells you that something is wrong. Learning to respond means deciding which evidence matters, choosing an intervention and checking whether it actually helped. Deadlock Manor turns that reasoning into a haunted investigation: the failing system is the mystery, and recovery is the way out.
What it does
Players enter one of three fictional outages: a broken release in The Phantom Release, abandoned database connections in The Connection Crypt, or a retry feedback loop in The Echoing Ballroom. They inspect logs, traces and runbooks, then spend limited response capacity on interventions. Actions have causal effects and side effects. Improving a graph is not enough: the player must repair the underlying problem and complete the scenario's simulated recovery check.
The Causal Replay Lab revisits a recorded decision with exactly the evidence, capacity and state available at that moment. It compares one alternative step, supports deliberately continuing that branch, and preserves a return path to the original shift. A downloadable case report records the decisions, outcome and score. The haunted casebook, evidence cards and midnight-outage setting are our Costume Contest treatment of a familiar incident-response drill.
How it was built
The application uses native browser modules with HTML and CSS, and a pure JavaScript simulation engine. Immutable transitions make saved histories replayable. The Analyst, Skeptic and Commander use transparent deterministic rules; the running game needs no model or external service. Node's built-in test runner checks the scenarios, replay integrity and persistence fault handling. The project contains 27 passing local tests, a static build and MIT-licensed source.
The original game, content and .kiro planning material were created with coordinated Codex assistance. On November 9, the authenticated official Kiro CLI 3.0.0 completed a 92.614-second read-only review, trusting only fs_read. It identified overwritten branch-save warnings and silently ignored backup-read failures. Codex reproduced both warning issues against the actual application handlers, adapted the fixes and added regressions. One speculative aliasing finding was rejected. Codex also independently found and fixed premature backup deletion during a failed restoration save. Kiro did not write the original application, edit these fixes or run the tests. The exact prompt, response, trace and assessment are published in the repository.
Challenges and lessons
Replay must retain what was known at the decision point instead of borrowing clues discovered later. Persistence also needs precise promises: a failed save must stay visible, and an existing original backup must survive until its replacement is saved. The review was useful because its suggestions were reproduced and assessed individually, rather than accepted as proof.
What's next
The edited gameplay walkthrough is now public on YouTube. The next step is collecting feedback on how clearly players can explain their interventions. Further scenarios should preserve the same evidence trail and explicit distinction between containment, repair and verification.
