This Is Not a Chatbot
The Straight Record is produced by a structured editorial intelligence system. It is not someone typing a question into an AI and publishing whatever comes back. It is a multi-stage pipeline with defined research protocols, adversarial fact-checking, locked production standards, and human editorial control at every decision point.
The system has four layers. Each layer is designed to catch what the previous one missed. The publisher, Brian Louden, controls the process from start to finish and makes the final decision on what goes live.
Layer 1: The Editorial Framework
Before a single word of any edition is written, the system operates within a permanent editorial framework. This is not a prompt typed fresh each session. It is a structured set of project-level instructions, production rules, design specifications, and editorial standards that persist across every edition and cannot be overridden by any individual session.
The framework includes locked word count budgets for every section of every story. It specifies which outlets to monitor, how to distinguish print coverage from broadcast commentary, when to use the Developing Story protocol, and how to handle the Democracy and Governance Watch tracker. It defines the publication's voice, its sourcing standards, and its rules for headline neutrality. It contains design specifications down to hex color values and font weights.
These are not suggestions. They are constraints. The production engine operates within them the way a newspaper operates within its style guide, except these rules are enforced programmatically rather than by editorial memory.
Layer 2: The Production Engine
Each edition begins with the publisher confirming the edition type, date, and any stories carrying forward from previous editions. The production engine then executes a structured research process.
First, broad monitoring across wire services and major outlets identifies the news cycle's dominant stories. Second, targeted searches of specific right-leaning and left-leaning outlets document how each side framed each event. This is not a summary of coverage. It is a forensic examination: What headline did Fox News run? What did MSNBC's host say in the opening segment? What word did the Washington Post choose that the Wall Street Journal did not? What did Breitbart put in the headline versus what CNN buried in paragraph twelve?
Third, broadcast and commentary sources are searched separately. The same outlet's written news desk and its primetime show often tell different stories. When they diverge, the system notes it.
Fourth, every factual claim is cross-referenced against primary sources, official records, and fact-checking organizations. The verified facts in The Record section come from wire services, government data, court filings, and primary documentation. No editorial language is permitted in this section.
Fifth, the system drafts each story in the publication's four-section format and subjects every draft to an automated word count audit against the locked budgets. If a story exceeds its ceiling, it is cut. The framing examples are never what gets cut. The framing examples are the product.
The production engine uses Claude, an AI model made by Anthropic, with live web search capabilities. It operates within the editorial framework described above. It does not freelance.
Layer 3: Adversarial Review (Red Team)
Before any edition reaches the publisher for final review, it passes through an independent adversarial verification process called the Red Team.
The Red Team's job is to doubt everything. It takes the completed edition and subjects every factual claim to independent research. It does not trust what the production engine found. It re-searches. It challenges. It looks for contradictions, outdated information, misattributions, and errors of emphasis. It checks whether the framing examples attributed to specific outlets are accurately represented. It verifies that The Gap section's criticisms are supported by the documented record.
When the Red Team identifies an error, it produces a correction report. The production engine implements every correction, searches the entire document for other instances of the same error, updates the edition, and logs the verification record in the edition's footer. The Red Team report is then reviewed by the publisher as the final check.
This is the layer that separates The Straight Record from AI-generated content. AI models can misread sources, miss context, or produce errors of emphasis. A system that uses AI without adversarial review is hoping those errors don't happen. This system assumes they will happen and builds the process to find them.
Layer 4: Publisher Review
Brian Louden reviews every edition before publication. He reads the stories, checks the framing examples against his own knowledge of the outlets involved, evaluates whether The Gap section is fair and evidence-based, and confirms the overall editorial judgment. He decides what publishes. He decides what gets held. He decides what gets killed.
The AI does not select stories. The AI does not decide what conclusions to draw. The AI does not publish. The journalist does.
What the AI Does and Does Not Do
The AI researches across dozens of outlets simultaneously. It identifies specific framing choices, documents them with citations, cross-references claims against primary sources, drafts analysis within locked editorial constraints, and formats the output. It makes the scope of this publication possible for an independent journalist. The alternative to AI-assisted production is not a larger newsroom. The alternative is that this publication does not exist.
The AI does not editorialize. It does not assume how an outlet covered something. If it cannot find specific coverage, it says so. It does not manufacture divergence. It does not publish without passing through adversarial review and human sign-off.
The Technology
The production system uses Claude (Anthropic) with web search, operating within a persistent project environment that maintains editorial standards, production rules, design specifications, and continuity data across every edition. The adversarial review layer operates as an independent verification process with its own research capabilities. The entire system runs on documented, reproducible protocols.
Privacy
This site uses Plausible Analytics, a privacy-friendly, cookieless tool that counts page views without tracking individuals. No cookies. No personal data. No visitor profiles. No advertising. No data sold or shared. The complete Plausible privacy policy is at plausible.io.
Corrections
When The Straight Record gets something wrong, it gets corrected. Corrections are logged in the edition's footer with the date, what changed, and the corrected value. If a correction changes the substance of a story's framing analysis, the edition is updated and the change is noted prominently. To report an error: Email Brian.