Coverage
The Field Manual covers software architecture, AI in production, security and privacy, fintech mechanics, startups, infrastructure, programming languages and tools, and tech history that explains current systems. Articles are picked because the editor has something specific to say, not because a topic is trending.
Sourcing
Every factual claim ties back to a primary source where one exists. Statistics quote the original report, not a secondary summary. Court cases cite the actual decision and distinguish ratio from dicta. Code claims tie back to documentation, source, or a reproducible test.
Sources are listed at the bottom of every article, with publication date and link. The sources panel is rendered from the database, not from inline prose, so the audit trail is queryable.
Fact-checking
Before publish, every article runs through an automated fact-checking pipeline that verifies every quantitative claim, date, court citation, and named entity against external sources using grounded web search. Findings are stored alongside the article with an applied-at timestamp. Articles do not ship while findings remain unresolved.
The master review pipeline scores every article on five dimensions (intellectual depth, technical insight, narrative voice, practical takeaway, brand alignment). The target threshold for publish is 48 out of 50.
Corrections
When an article is found to contain an error after publish, the article is edited in place to remove the error. We do not annotate with “Update: corrected X” prose inside the running text. The audit trail of the correction lives in the fact-check findings table, with timestamps. This is the greenfield-corrections policy: the published article is the source of truth at any given time, and prior errors are not memorialized in user-visible text.
If a correction is structural enough to change the article’s thesis, the updated field is set and an explanation appears in the article’s “Update note” banner.
AI disclosure
Articles are written by the editor. Drafts may be reviewed by AI tools against a master review rubric, and the editor may incorporate suggested fixes for specific scoring dimensions. Fact-checking uses an AI assistant with grounded web search. Hero images for articles are AI-generated line drawings. No article is published that was generated end-to-end by an AI tool without editor authorship and review. All published content is under the editor’s name and the editor’s accountability.
Tracking and privacy
No tracking pixels in emails. No third-party analytics on the site. No behavioral retargeting. The Sunday digest is double opt-in (Gmail and Yahoo 2024 sender rules) with one-click unsubscribe (RFC 8058). Subscriber data lives in a Cloudflare D1 database under the publisher’s control. See the privacy policy for full data handling.
Voice rules
Short sentences. Plain words. No marketing fluff. No em-dash overuse. No banned adjective stacks (“robust,” “seamless,” “leverage,” “AI-powered,” etc.). No scare quotes around normal words. The coffee-shop test applies: if you wouldn’t say it across a table from another senior engineer, it doesn’t ship.
Diversity and fairness
The publication welcomes guest analysis from operators of any background. Coverage is judged on technical accuracy and operator credibility, not on the author’s affiliations. When two operators disagree on a technical question, both readings get aired with the evidence each cites.
Feedback and engagement
Reader corrections, additional sources, and counter-evidence are welcomed at editor@roamingpigs.com. Confirmed errors are corrected per the policy above, usually within 48 hours. Reader letters worth publishing are surfaced in the Sunday digest with attribution and consent.