ferravex
GitHub

Paper Alibi

A complete six-case detective adventure about a local newspaper following a paper trail through its city. Examine places, interview people, compare records, and publish conclusions that explain who, how, and why.

The investigations are authored rather than randomly assembled. They have 18 suspects, 18 locations, 18 physical records, 18 witness statements, six signed confrontations, six explanatory debriefs and a final newspaper ending. Independent scene records establish facts; a suspect's testimony can contradict them. Every case provides two independently supported innocent alibis and one contradiction that identifies the remaining suspect. Later cases use physical traces and cross-referencing in addition to recordings.

The six dossiers

  1. The Last Carbon - a stolen payroll ledger in a locked newspaper safe.
  2. The Silent Bell - a museum theft behind unbroken glass.
  3. The Blue Lantern - lighthouse sabotage and a suspicious sea passage.
  4. The Borrowed Voice - a forged radio announcement assembled from wax.
  5. The Winter Orchard - substituted seeds and an unjust land valuation.
  6. The Glass Mayor - a final evidence-case swap before a public hearing.

Each case can be read at your own pace. No real-time timer punishes reading or planning. The cases form a connected story about diverted public funds and a council registrar's grey seal, while retaining their own suspects and solutions.

How to investigate

The screen explains itself with icons rather than instructions. Four icon tabs (map, people, notebook, target) show their progress as dots and checks, and the tab that holds your next useful step pings softly.

The header tracks the case: six progress pips, the case name, three WHO / HOW / WHY slots that light as you earn them (the liar's face appears once the contradiction is connected) and four hearts of credibility. Pointing at a ready Accuse button makes the heart at risk flash.

Mouse clicks select tabs and rows; hovering a tab names it. Clicking a theory answer cycles it. The large bottom buttons perform their displayed actions. Away from Notebook, X opens it. Held movement or held Space does not repeat deductions: actions use press edges.

The shared host provides Enter to begin, Esc/P to pause, H for the four-bullet field guide, R for a new six-case campaign, M for sound and F2 for control remapping. Space/Enter, X/E and C/Q share their respective primary, secondary and third-action bindings. Key caps on each button show the defaults.

Credibility, scoring and replay

You begin with four credibility points, shown as hearts. A formal accusation with the wrong suspect, method or motive breaks one heart. Four false reports end the campaign. An unsupported accusation cannot be filed and does not spend credibility. Exploring an unproductive clue connection is safe; keep investigating.

New physical evidence and interviews award 25 points each, a valid contradiction 100, and its signed confrontation 75. A closed case awards 500 plus 50 for each remaining credibility point. Printing the final edition awards 500. A spotless campaign scores 6,650 and earns Editor of Record. Repeated examinations, links, confrontations or terminal commands cannot award extra points.

Try to solve the cases without false accusations. Notebook pinning is a reasoning aid: compare a concrete denial with an independently verified fact. The innocent alibis matter as much as an incriminating trace. Best scores and completed-run records stay on the current device through the shared host; unfinished investigations are not saved.

Run and verify

cargo run -p paper-alibi --offline --locked -j 2
cargo run -p paper-alibi --offline --locked -j 2 -- --headless
cargo test -p paper-alibi --offline --locked -j 2
cargo clippy -p paper-alibi --all-targets --offline --locked -j 2 -- -D warnings

The headless command completes the full campaign with ordinary player commands, without a display. Its scripted investigator gathers all records, makes the connection, confronts the suspect and selects the authored solution through the same input path as a human. It wins at tick 4,248 with four credibility points and 6,650 score. It is a reproducible solution, not an independent human test of mystery difficulty.

For parent-owned native graphical checks, useful autoplay frames in the first case are around 100 (places), 250 (people), 445 (two pins on the board), 470 (contradiction), 515 (signed response), 710 (accusation) and 732 (edition). The standard runner accepts --autoplay --smoke-frames N --screenshot PATH; FERRAVEX_ARCADE_FAST=40 with about 220 frames reaches the victory card. Demonstration runs use transient personal records.

Implementation and limits

model.rs contains renderer-independent rules with fixed arrays: seven current records, two pins, three theory answers and four view selections. Each executed tick also keeps a bounded list of the commands it accepted or refused (a new record, a pin, a contradiction, a wrong witness, a retracted accusation). The list is cleared at the start of every update, ignored by equality and never read by the rules. content.rs contains the original case writing. view.rs paints the paper dossiers, procedural portraits, engine pictograms for every location, the red-string pinboard, the header's WHO / HOW / WHY slots and hearts, and the icon results card. It turns the tick's events into effects where they happened: bursts on collected records, a stamp slam, a "Contradiction!" callout over the pinboard, a heart breaking in the header, and small shakes with one-word popups for refusals. The shared ferravex-arcade runner owns timing, input sampling, menus, storage, audio and the effect pool. No external packages, assets, requests or services are needed.

Twenty-three behavioral tests and one executed documentation example pass. They cover a legal full-campaign win, credibility failure, immutable endings, one-shot rewards, invalid actions, clue pinning, every verdict field, mouse button semantics including the Clear button, campaign progress pips, per-tick events, effect requests for a contradiction, a retraction and a refusal, layout fits for every authored line, and the help word budget. Strict scoped Clippy and formatting pass. Browser/Wasm integration, actual graphical layout, device audio, gamepad/touch play and platform portability are parent-owned validation; these simulation checks do not establish them.