AI Services Market
AI Services Market › Fabrique de compétences › Flux de travail › Investigative Dossier
$4.99 · Sans inscriptionv1.0.0– vues

Investigative Dossier

Assemblez le tableau complet à partir d'éléments publics

Construisez des dossiers d'enquête à partir d'informations publiques : collecte OSINT, construction de chronologies, triangulation des sources et suivi de l'argent — avec des garde-fous éthiques (informations publiques uniquement, pas de piratage). Pour les journalistes, chercheurs et analystes de due diligence.

Aperçu de Investigative Dossier
Idéal pour
Journalistes, chercheurs, analystes, les profondément curieux
Temps nécessaire
~1 heure par dossier
Niveau
Avancé

Installez-la en 3 étapes

Sans inscription. Sans outils en ligne de commande. Choisissez votre IA, copiez une ligne, collez-la.

Pro — les 511 compétences

Vous préférez le faire vous-même ? Achetez la compétence ($4.99), téléchargez le ZIP depuis votre confirmation d'achat, décompressez-le dans le répertoire indiqué ci-dessus, redémarrez votre session IA — même résultat. Vous l'avez déjà achetée ? Votre assistant IA peut la télécharger pour vous.

Contenu du pack

Un fichier SKILL.md (le guide pratique) + un modèle de départ. Aperçu :

Investigative Dossier

When to use

  • You're a journalist chasing a story where the pieces are scattered across public records, old posts,
    filings, and interviews — and nobody has put them together yet
  • You're doing due diligence on a company, founder, or counterparty: what's on the record, what do
    their own words say, where do the stories conflict
  • You're a researcher mapping how a deal, scandal, or controversy actually unfolded — billboard
    campaigns, social-media stunts, secret profit arrangements — from what was said and done in public
  • You're vetting a claim someone made about themselves: resume lines, "sold the company" stories,
    funding announcements, awards that don't exist
  • You're preparing for an interview with a difficult subject and need every public contradiction of
    theirs in one document before you sit down
  • You're an analyst writing a background memo on a person or entity where accuracy matters more than
    speed — the kind of file a newsroom or a board would actually rely on
  • You're a citizen researcher or student learning to separate verified fact from viral rumor on a
    contested topic, and you want a method instead of a thread

Workflow

  1. The Question (input: a topic or hunch → output: one-page investigation brief)
    Write the question before collecting anything. Not "look into X" — a real question: "Did Company
    A conceal the terms of its 2024 licensing deal?" or "What is the actual employment history behind
    Founder B's public biography?" Then add:
  • Scope: the people and entities involved, named explicitly. If a name isn't on this list,
    it's out.
  • Time window: the dates that matter. Open-ended investigations never end; bounded ones do.
  • Success criteria: what "answered" looks like — "a company-by-company table of officerships,
    each row source-cited."
  • Out of scope: say it in writing — family members, unrelated business lines, speculation
    about motives. The out-of-scope list is what keeps a dossier from becoming a fishing expedition.
  • A dossier built on a vague question becomes a pile, not a file — this brief is what every later
    step answers to. If the question changes mid-investigation (it will), revise the brief and date the
    revision. Never let the question drift silently.
  1. Collection (input: brief → output: evidence inventory)
    Collect only what is public or on the record. Work through the source ladder from strongest to
    weakest:
  • Primary documents: corporate filings, court records, regulatory disclosures, property and
    trademark registries, campaign-finance data, archived official pages. These anchor everything.
  • The subject's own words: their posts, interviews, press releases, books, testimony,
    investor decks. People contradict themselves more often than enemies contradict them — their own
    archive is usually the richest source in the file.
  • Secondary reporting: reputable outlets' reporting, each logged with its own sourcing note
    (did the outlet see the document, or is it citing another outlet?).
  • Context material: industry data, expert background, historical timelines — useful for
    interpretation, never for claims.
  • Log every item the moment you find it: what it is, where it came from, the exact URL or record
    number, and the date you captured it. Screenshot or archive anything that could disappear — the
    internet forgets on schedule, and public figures delete on purpose. A claim with a dead link is an
    uncorroborated claim.
  • Hard boundaries: nothing behind a login you don't own, nothing obtained by deception or
    pretexting, nothing hacked or leaked through illicit means, no private messages you weren't a party
    to. If a source can't be cited openly, it doesn't go in the file. If you're unsure whether a source
    is legitimate, it stays out until you're sure.
  1. Timeline (input: evidence inventory → output: chronological event map)
    Lay every dated fact on one timeline — one row per event, in order, no commentary yet. Public
    statement on March 3, filing on March 14, contradictory post on April 2. Then:
  • Mark each entry with its source, so the reader can see at a glance which dates rest on solid
    filings and which rest on a single post.
  • Flag divergences explicitly: anywhere the subject's public story and the paper trail visibly
    separate, mark it with a ⚑. This is where documentaries find their spine — the divergence is usually
    the story.
  • Note the silences too: long gaps where a normally vocal subject went quiet, filings that should
    exist but don't, periods where the record thins. Absences are data.
  • Keep two layers if the case is complex: a public-facing timeline (what was said and done in the
    open) and a records timeline (what the documents show). Laying them side by side is where
    contradictions become undeniable.
  1. Triangulation (input: timeline + inventory → output: confidence-graded claims)
    Turn observations into claims — one claim per row — and grade every claim with a fixed rubric:
  • Confirmed: three independent sources, or one primary document that directly establishes it.
    This is the only grade that can carry a headline.
  • Probable: two independent sources pointing the same way. Usable with attribution and the
    evidence shown.
  • Uncorroborated: a single source. Keep it in the file, label it honestly, and never build
    another claim on top of it alone.
  • Disputed: sources contradict each other. Present both sides, strongest evidence first, and
    name the contradiction instead of smoothing it over.
  • Independence is the whole game: three outlets quoting the same press release is one source, not
    three. A subject's own claim repeated by ten outlets is still the subject's own claim. Count sources
    by origin, not by headline.
  • For each claim, write the "steelman against": what's the most innocent explanation consistent
    with the evidence? If you can't articulate it, you haven't understood the evidence yet. Claims that
    survive their steelman are the ones worth keeping.
  1. Gap Analysis (input: graded claims → output: unknowns ledger + fill plan)
    Write down what you still don't know — explicitly, in one ledger, no matter how embarrassing the
    gaps are. "Unknown: who authorized the payment." "Unknown: whether the second entity was ever
    registered." "Unknown: what the subject did between 2019 and 2021." For each gap, decide:
  • Fillable by more collection: loop back to step 2 with a targeted search plan (which
    registry, which archive, which records request).
  • Answerable only on the record: an interview request, a public-records/FOIA-style request, a
    direct question to the subject. Write the exact ask.
  • Unanswerable from public sources: mark it as such and stop. Chasing unanswerable gaps is
    how investigations turn into conspiracy boards.
  • An honest unknowns section is what separates a dossier from a hit piece. It also tells you when
    to stop: when the remaining gaps can't be filled legally or ethically, the dossier is done — and the
    gaps say so, in writing, instead of being quietly dropped.
  1. Assembly (input: brief + inventory + timeline + claims + unknowns → output: the dossier)
    Assemble the file in this order, and don't improvise the structure — readers of dossiers
    (editors, lawyers, boards) expect to find things in the same place every time:
  • Executive summary (one page): findings first, each with its confidence grade, then the
    one-paragraph bottom line. If the reader stops here, they still know what you found and how sure you
    are.
  • Question and method: what was asked, what was searched, what was excluded, and the revision
    history of the brief. Method transparency is what makes the findings checkable.
  • Timeline: the spine of the file, source-marked, divergences flagged.
  • Key findings: ordered by confidence (confirmed first), each claim with its evidence
    attached and its steelman-against noted.
  • Unknowns ledger: the fill plan and the unanswerables, stated plainly.
  • Source inventory: every source with a reliability note — primary document, subject's own
    statement, secondary reporting, single-source.
  • Ethics statement: what was and wasn't used, and what the dossier cannot prove.
  • Write findings as what the evidence supports — "records show X," not "X is guilty of Y."
    Adjectives are not evidence: if "shady," "explosive," or "bombshell" is doing the work, the evidence
    isn't.

Aperçu du modèle

Investigative Dossier Pack

Investigation brief

  • Question (one precise sentence): […]
  • Scope — people/entities involved: […]
  • Time window: […]
  • What "answered" looks like: […]
  • Explicitly out of scope: […]

Evidence inventory

_Archive or screenshot anything that could be deleted — a claim with a dead link is uncorroborated._

Timeline

  • [date] — [event] — source: […]
  • [date] — [event] — source: […]
  • ⚑ Divergence flagged: [public claim vs. record — dates, sources]

Claims register

  • [Confirmed / Probable / Uncorroborated / Disputed]: [claim]
  • Support: (1) […] (2) […]
  • Note: […]
  • […]: [claim]
  • Support: […]

Le pack de compétence lui-même (SKILL.md + modèle) est rédigé en anglais.

Pro — les 511 compétences

Vous aimerez aussi