Methodology

I. What we draw from

MarkTell's reporting starts from public records and primary documents. The newsroom monitors a curated set of fifty or more sources every day: SEC and CFTC enforcement actions, FTC complaints, DOJ criminal indictments, FINRA disciplinary actions, FBI white-collar bulletins, state attorney-general announcements, financial-press reporting, and topical aggregators that surface fraud-adjacent stories from across the wire.

We also accept reader tips at tips@marktell.com. A tip on its own is not a story. A tip pointing us at a filed document, an enforcement action, or a verifiable public record is the start of one.

We do not pay sources. We do not run stories built solely on anonymous social media posts, paid news services, or material we cannot trace to a primary record.

II. How a story gets chosen

Incoming items are scored against four criteria: how directly they map to one of our fraud taxonomies (pump and dump, Ponzi, course and guru, real estate, crypto, MLM and affinity, private placement, charisma stocks, senior and family, recovery scams); the strength of the underlying public record; the editorial potential of the narrative; and how useful the piece will be to the four reader states we write for (almost bought, already bought, protecting someone, curious).

The screening pass uses an AI model to draft the initial scoring and lane recommendation. A human editor reviews the queue and approves what moves forward. Items that score below threshold do not get drafted. Items that pass into drafting carry their lane assignment with them: Case File, Breakdown, Open File, or Feed and Tell.

The lane decision is the most consequential editorial choice we make. Lane discipline determines what language we are allowed to use, how the piece is structured, and how the legal review reads it. The right lane does not always look obvious from a single press release. When it is borderline, an editor picks the more conservative of the candidates.

III. How a story gets reported

Every story moves through the same production sequence.

Research. Primary documents are pulled and assembled. Court filings, regulator complaints, prior reporting, financial disclosures, and any prior coverage of the principals. The research output is a structured brief tied to the case, not prose.

Draft. The first pass writes the structural article from that brief: facts in the right order, citations where they belong, the lane discipline applied. We use Anthropic's Claude (Sonnet) for this pass.

Polish. A second pass rewrites the draft for editorial voice and rhythm: the cinematic register we are known for, the dramatic-beat sentences, the Roman-numeral section structure, the careful framing of allegations. We use OpenAI's GPT for this pass. The polish pass is constrained to never overwrite the citation set produced in the first pass.

Quality and legal review. A two-model compliance check runs against the polished draft before publication. It looks for unsupported assertions of guilt or liability, citation gaps, lane-discipline violations, and stylistic issues that have led to errors in the past. Failures do not auto-publish; they go back into the queue.

Human editor. A human editor reviews every piece before publication. The editor has authority to send a piece back to research, rewrite passages directly, kill a piece, change the lane, or hold for additional verification. Nothing goes live without that review.

IV. The Evidence Trail

Every published piece carries a numbered Evidence Trail at the foot of the article. It lists the documents, regulatory actions, court filings, and prior reporting we used to build the piece, in the order most relevant to the reader.

The Evidence Trail is not promotional. It is a working bibliography. A reader who wants to verify any claim in a piece, a regulator who wants to see what record we built on, an attorney advising a client about coverage of their company, all of them can read what we read. Where the original document is publicly accessible online, we link directly. Where it is paywalled or filed only in court, we cite the case and venue so the reader can find it.

If a piece does not deserve an Evidence Trail, the piece does not deserve to be published.

V. Reconstruction in practice

When a Case File depicts a scene the writer did not witness, the scene is reconstructed from the public record. We start from what filings and depositions specify (who was where, what was said on the record, what dates frame the events). We build outward only as far as the documents support, and we stop before invention.

A reconstructed dialogue inside a Case File is not a transcription. It is a faithful editorial rendering of what the documents establish was said, in the rhythm of how it was said, calibrated to the documented facts. We do not put words into a real person's mouth that contradict the record. We do not invent meetings. We do not change outcomes.

Where reconstruction is heavier than usual in a piece, we say so at the top, not just at the bottom. The reader is entitled to know how thick the reconstruction is before they read.

VI. Imagery

The lead image on a MarkTell story is generated by an AI model, prompted under what we call the cinematographer framework: high-contrast monochrome with a single point of warm color (oxblood or amber), a specific camera position, a specific time and weather, and a specific narrative object in frame. The references behind the visual style are documentary photographers (Todd Hido, Gregory Crewdson, Taryn Simon, Alec Soth, Philip-Lorca diCorcia), not crime-scene photography.

Generated imagery does not depict real persons, real events, or real locations. It is mood and metaphor. A photograph of an envelope on a doorstep illustrates a piece about a confidence game; it is not a photograph of any real envelope on any real doorstep. Where we publish an actual photograph (a court exhibit, a regulator-released image, a documentary photograph from a public source), the caption identifies it as such and credits the source.

We do not generate images of real defendants. We do not generate images of real victims. We do not generate images that could be mistaken for documentary evidence in any case we cover.

VII. The newsletter

The MarkTell daily newsletter goes out at 6:00 a.m. Eastern time. It is a digest of the most consequential pieces from the previous publishing cycle, selected by a human editor. Subjects are chosen for relevance to the reader who is about to be pitched, the reader who already sent the money, and the reader who is trying to protect a family member.

We do not buy email lists. We do not sell our subscriber list. The list is built only from people who chose to subscribe on this site. Unsubscribes are honored on the next send.

VIII. Comments and community

Comments on stories are open to readers signed in via LinkedIn. The sign-in is the moderation: comments are tied to a real LinkedIn identity, which raises the cost of bad-faith participation and lowers the cost of moderating it.

We remove comments that defame, threaten, or harass any person; that promote a fraud or attempt to recruit other readers into one; that disclose personal information about victims, defendants, or third parties beyond what the public record contains; or that violate the law in the reader's jurisdiction or ours.

We do not remove comments because they criticize our reporting. Honest criticism is welcome and often useful. If a comment identifies a factual error, that is a correction request. We treat it as one.

IX. When something is wrong

Errors are reported to corrections@marktell.com. Every report is read.

If the report identifies a factual error, the editor pulls the original sources, verifies the discrepancy, and runs a correction. Material corrections appear at the top of the affected story, dated, and they remain there. Substantive errors that change the meaning of a piece trigger a retraction, with the retraction notice replacing the affected text and explaining what went wrong.

We do not edit history without disclosure. The version of a piece that was wrong is part of the public record once published, and the corrections log explains what changed and when.

X. The team

MarkTell is staffed by a small editorial team led by an editor. Three senior contributors cover specialized fraud beats: a sales-floor and confessional-operator beat, a crypto and code-level beat, a securities and regulatory-law beat, and a financial-investigator beat. A research desk pulls primary documents.

All bylines on MarkTell are pen names. Coverage is built and reviewed under those bylines but published under the editorial responsibility of the publisher. The pen-name protocol is a deliberate choice: it lets the writers do their work without the reputational and personal-safety costs that named coverage of fraud subjects can carry. It does not reduce accountability. The publisher is the legal operator of the publication, registered with our hosting and payment providers, with our legal counsel, and with the relevant regulatory and tax authorities.

If you need to reach a contributor about their work, write to editor@marktell.com with the story URL and the question. Editor's-desk contact is the right channel for any reporting concern.

XI. What we do not do

This is the operational expression of the standards.

• We do not pay sources for stories.
• We do not give advance copies of stories to subjects of coverage.
• We do not accept gifts, trips, or any consideration from companies or individuals we cover.
• We do not hold financial positions in any company, security, token, or investment vehicle covered in our reporting.
• We do not run sponsorships from regulated investment advisers, broker-dealers, fund managers, cryptocurrency platforms, multi-level marketing organizations, or any business whose model includes selling investment opportunities to retail.
• We do not publish unverified rumors, no matter who is sending them or how confident the sender is.
• We do not let AI tools generate facts, citations, or attributed quotes. Every factual claim in a published piece traces to a real document the editor has seen.
• We do not delete stories on request from subjects whose only complaint is that the story exists.
• We do not silently edit history.

XII. A note on this page

This page is reviewed quarterly and updated as the operational practice changes. Where we add a new tool to the production workflow, change a verification step, or modify how a step works, the change shows up here. The revision date at the top of the page reflects the most recent update.

The point of publishing the methodology is the same as the point of publishing the Evidence Trail. A reader who wants to know how the work is made should be able to find out. A reader who finds something wrong with how the work is made should be able to point at the specific step and ask why.