The bridge · AI-driven crime fiction

For Agents

This site is a craft reference; this page is its interface. It turns the chapters into two routines a writing agent runs — an analysis method that classifies and audits a crime story, and a generation method that constructs one. Every step points back to the chapter that justifies it.

Method A — Analyse & classify

Given a target text (or a brief), place it before you judge or imitate it.

  1. Locate the school. Match the text against the discriminating features — outsider vs. institution, violence on/off stage, whether order is restored. Split on the feature, never the mood.
  2. Set the mode. Decide hardboiled (code intact, order restored) vs. noir (the fall completes). See Chapter 6 — these are opposed, not adjacent.
  3. Find the spine. Reconstruct the hidden crime timeline (means, motive, opportunity) and the investigation that mirrors it — Chapter 1.
  4. Audit fair play. Is every clue the detective uses present to the reader by ~75–80%? For audit purposes the reveal opens at the first scene in which the detective acts on a settled solution — a clue first shown at or after that point does not count as planted. Was the culprit introduced early? Any coincidence, deus ex machina, or info-dump? — Chapter 4.
  5. Test the cast. Does every significant character carry means/motive/opportunity and a secret? Is the villain hidden in plain sight rather than flagged — or, in proof-gathering P.I. mode, is the killing (not the man) what stays discounted? — Chapter 5, §5.2.

Method B — Construct

Build the story so the reveal is "surprising yet inevitable."

  1. Know the ending first. Fix who, how, and why before seeding a single clue — Chapter 1.
  2. Choose voice and POV deliberately. Hardboiled first-person vs. Highsmith's third — a live trade-off, not a default. Make setting exert pressure — Chapter 3.
  3. Give the investigator a code, a wound, a stake. The reader's question is "will they hold the line," not only "who did it" — Chapter 2.
  4. Plant clues by the Rule of Three; deploy red herrings in waves. A careful reader should be able to solve it 2–3 chapters early — most won't — Chapter 1.
  5. Stage the reveal through deduction, not monologue. Resolve the crime and the protagonist's arc together (the cast and its stakes are Chapter 5); wrap subplots at or before the reveal — Chapter 4.
  6. Revise for the genre. Read backward to check the clue trail and timeline; time a beta reader's guess — Chapter 7.

Short-form calibration

The chapters calibrate to the novel. For a short story the percentage marks scale down cleanly; the count-based rules do not — inherited unchanged, they overload the story. The short form is also the pipeline's fast lane: a draft a human reviews in one sitting closes the construct→audit→revise loop in hours, so when the goal is improving the method rather than shipping a book, default to short.

ParameterNovel defaultShort story (~2–8k words)
Suspect pool Several suspects, each holding two of means/motive/opportunity; the culprit all three (§1.5) Three suspects: one full triple, two doubles. The lattice is unchanged — only smaller.
Red herrings Two (character-forward) or a fuller field (puzzle-forward) (§1.4) One, resolved before the reveal.
B-story A relationship subplot alongside the investigation (§1.6) None. Fold the emotional stake into the investigation itself.
Cast Suspect pool plus recurring cast around a morally grey victim (Ch. 5) At most six named characters; every name carries a function.
Timeline & POV Crime backstory may span years; POV choice is a live trade-off (Ch. 3) Two separate bounds: the narrated investigation is one continuous span, hours to days; the crime backstory may span months but must compress to the few timeline nodes the reveal actually needs. Same POV trade-off, but one POV — no switches.
Structure marks Discovery at 10–15%, midpoint ~50%, reveal opens by ~75% (§1.1) Unchanged — percentages scale. At 5,000 words the discovery lands on the first page.
Solvability A careful reader could solve it 2–3 chapters before the reveal (§1.3) Solvable from the page just before the reveal; must fully reconstruct on a reread.

Write in the vein of

When told to imitate a tradition, do not average it into generic grit. Pick one anchor and name the specific move.

Walter Mosley

postwar Black L.A.

Reinvents the P.I. across race, class, and era; voice + place + history as one instrument.

Devil in a Blue Dress

Sara Paretsky

Chicago

Reinvents the eye across gender; corporate/financial rot as the corrupt system.

Indemnity Only

Michael Connelly

Los Angeles

Procedural authenticity at scale; the case worked with real institutional texture.

The Black Echo

S.A. Cosby

rural US South

Neo-noir with race and class as the engine; setting as character, violence with weight.

Razorblade Tears

Steph Cha

Los Angeles

The Marlowe tradition seen through a young Korean-American woman — race, gender, generation.

Follow Her Home

Attica Locke

East Texas

Race and region; systemic racism as the case's real subject.

Bluebird, Bluebird

Dashiell Hammett

San Francisco / NYC

Gave murder "back to the people who commit it for reasons" — hardboiled realism, the operative's flat cold eye.

The Maltese Falcon

Raymond Chandler

Los Angeles

The mean-streets code and the built voice — wisecrack, surprising simile, description as characterization.

The Big Sleep

Patricia Highsmith

shifting / European

Noir from inside the culprit; third-person suspense; the amoral protagonist you follow anyway.

The Talented Mr. Ripley

James M. Cain

California

Pure doom-drive noir — desire and ruin, no detective needed, the fall as the whole arc.

The Postman Always Rings Twice

The German lineage — Arjouni, Ani, Schlink — sits on the Schools page, where the P.I. white space is argued.