true-crime/SKILL.md

Factual verification workflow for true-crime narratives grounded in court records, public documents, and journalistic investigation. This mode prioritizes source-tier classification, primary-record cross-checking, and systematic escalation strategies for unresolved claims.

Core workflow

Use these four steps in order for every claim in your manuscript:

1. Source-tier triage

Classify the claim's evidence quality:

• Tier 1 — Primary legal record (court dockets, DOJ press releases, sentencing transcripts, BOP records, indictments, FTC filings)
• Tier 2 — Direct subject statement (long-form interview transcript where subject speaks in their own words, with interviewer independently corroborating surrounding claims)
• Tier 3 — Journalistic paraphrase (NYT, Wired, Vice, NBC summarizing the subject's claims in the outlet's own words)
• Tier 4 — Marketing/bio copy (speaker bureau pages, podcast show-notes, consulting-firm bios — recycled self-description, usually not independently checked)

A claim appearing only at Tier 4, however many times it's repeated, stays [SINGLE-SOURCE]. Repetition across marketing copy is not corroboration.

2. Query construction

For subject-heavy cases, use specialized search patterns:

• Always pair surname with handle/alias (e.g., "gollumfun"), specific entity (e.g., "ShadowCrew.com"), or named associate
• Search court records specifically: "[name]" "[case-specific term]" indictment OR docket site:justice.gov
• Discard false positives aggressively; don't pattern-match on partial fits
• Log dead ends as [GAP — searched, not public] rather than [GAP] alone

3. Cross-source reconciliation

When multiple sources give different numbers or accounts:

• Convergence rule: If 5+ independently-produced sources converge on one figure, replace the original with convergent figure and cite the convergence
• Split evidence: If sources split evenly, keep both with per-outlet attribution ("X per [outlet], Y per [outlet]")
• Never average or guess — count independent sourcing lines instead

4. Escalation for persistent gaps

For facts that stay unverified after steps 1–3:

1. PACER direct docket search — by defendant name + district guess (if the district is known)
2. BOP inmate locator — historical only; live locator won't return released inmates
3. Local news archives — NewsBank, Google News Archive, specific time windows
4. Genuine gap disclosure — if none resolve it, the fact stays [GAP] with author's-note-level disclosure

---

Example workflow

Claim: "Brett Johnson served seven and a half years in federal prison."

Step 1 — Tier triage:

• Multiple journalist profiles (Tier 3) cite "seven and a half years"
• Johnson's own Darknet Diaries interview (Tier 2) mentions total time served
• DOJ press release (Tier 1) likely contains the sentencing order

Step 2 — Query construction:

• Search: "Brett Johnson" "gollumfun" ShadowCrew sentence site:justice.gov
• Search: "Brett Johnson" "gollumfun" indictment Newark PACER
• Search: "Darknet Diaries" "Gollumfun" transcript

Step 3 — Cross-source reconciliation:

• Journalist profiles: "seven and a half years total"
• Darknet Diaries Part 1: "roughly seven years, with escape adding time"
• One profile: "90 months for the original case, extended after escape"
• Result: Use "roughly seven and a half years total, following a sentence reported as 90 months for the core ShadowCrew case"

Step 4 — Escalation:

• If the exact sentencing court/date is still missing after these searches, mark as [GAP — PACER district narrowing needed]
• Next step: PACER search with "ShadowCrew defendants" + "Newark" to find the correct federal district

---

What each submodule covers

verification-engine.md

• Detailed tier definitions with examples
• How to distinguish Tier 2 from Tier 3 (the critical boundary)
• Why repetition is not corroboration
• Red flags for overcounting sources

query-construction-guide.md

• Common-surname poisoning and how to filter it
• Site-specific search operators for court records, journalism, academic databases
• Handle/alias/entity pairing strategies
• False-positive screening protocols

cross-source-reconciliation.md

• The 5+ convergence rule and its rationale
• How to handle 50/50 splits in evidence
• When to use footnotes vs. integrated attribution
• Precedent examples from journalism and academic history

escalation-protocols.md

• PACER search strategies and district inference
• BOP inmate locator workflow (and why it fails for released inmates)
• Local news archive search patterns by region and decade
• How to format [GAP] vs. [GAP — searched, not public] in your manuscript
• Author's-note disclosure templates for remaining gaps

---

Integration with narrative

This workflow produces sourcing annotations that feed directly into:

• Scene-by-scene outlines (which facts are Tier 1, which need reconstruction flags)
• Character portrayals (what dialogue can be quoted vs. summarized)
• Editorial checklists (where [SINGLE-SOURCE] items need corroboration)

See `../../../../shared/narrative-architecture.md` and `../../../../shared/editorial-checklist.md` for how to translate verification results into narrative structure.

---

When to stop searching

• Tier 1 fact with 3+ independent Tier 1 sources: Use it, cite converging Tier 1s
• Tier 2 fact with 2+ independent Tier 2 sources: Use it, note as Tier 2
• Tier 3 fact with 5+ independent Tier 3 sources: Use it, note convergence
• Fact appearing only at Tier 4: Mark as [SINGLE-SOURCE]; don't use without Tier 1–3 backup
• Fact unreachable after PACER, BOP, archive searches: Mark as [GAP] and disclose in author's note

Genuine gaps are not failures. They're opportunities for transparent writing.

---

Common pitfalls to avoid

1. Assuming interview transcripts are Tier 2 without corroboration — The transcript must show the interviewer independently verifying surrounding claims (e.g., "I read the police reports and it checks out")
2. Treating speaker-bureau bios as Tier 3 — They're Tier 4; they recycle the subject's self-description without independent checking
3. Counting repetition as corroboration — Five articles copying the same press kit is one source, not five
4. Skipping PACER when you have a quote — If the court record says something different, Tier 1 trumps Tier 2 or 3
5. Inventing specificity to fill gaps — If you don't know the prison name, don't guess; mark [GAP] and move on
6. Assuming "not found" means "doesn't exist" — Document your search attempts; "not found after 3 reformulations" is different from "never searched"

---

Resources

• PACER (Public Access to Court Electronic Records): pacer.uscourts.gov
• BOP Inmate Locator: bop.gov/inmateloc (historical archives available through Ancestry, some state corrections)
• NewsBank: newsbankarchives.com (local newspaper archives, typically institutional access)
• Google News Archive: news.google.com/newspapers (historical newspaper search, free)
• RECAP (Free PACER archive): recapthelaw.org

---

This module is part of @h4shed/skill-underworld-writer. See the root SKILL.md for the full skill overview and integration guide.