How this is made

Why a reader should care how a newsletter is made: because the question every careful reader is silently asking is “who is telling me this, and how do they know.” The honest answer is on this page, in plain English, once. There is no per-issue badge and no buried disclosure.

Who writes it

Joseph Gomez writes and edits every issue. The byline on every piece is his real name. The bio, including dated operator-side experience and contact, is at /about/joseph-gomez/. Joseph is one human, one editor, one byline. There is no “The Editors” collective and there is no ghost writer behind a brand voice.

How an issue gets made

The publication runs the same four steps every day:

1. Source selection. Joseph picks the topic and the primary source. The roster of sources rotates so the publication does not become an echo of a single feed. The roster is reviewed quarterly.
2. Angle and outline. The angle, the outline, the chart concept, and the call-out decisions are Joseph's. This is the editor's job and it does not get delegated.
3. Drafting and illustration. The first draft of prose and the first pass of the bespoke diagram are produced with the help of Claude (Anthropic). Claude is a large language model that the publication uses as a drafting and rendering assistant. It does not pick the angle, it does not pick the source, and it does not get the last word on any claim.
4. Review and verification. Every draft is run through a published anti-slop rubric and a multi-model reviewer panel before send. The deterministic part of the review checks that every numeric claim has a primary source URL with an access date, that the sample size is named, that there are no em-dashes in customer-facing copy, and that the lede leads with purpose, not with feature. The judging part runs a parallel three-model panel (Claude, GPT, Gemini), and the panel either agrees the piece is shippable or the piece goes back to Joseph. The rubric and the panel are described in more depth in the publication's reviewer harness documentation.

What AI is used for, and what it is not

Used for: turning Joseph's outline into a first prose draft, rendering the bespoke chart from a brief Joseph wrote, catching slop patterns before they ship, and acting as one of three independent judges on the reviewer panel.

Not used for: picking the source. Picking the angle. Inventing a statistic. Making up a quote. Standing in for a human accountability claim. Writing a correction. Writing the bio.

If the AI hallucinates a number or a name, the deterministic fact-checker catches it (numeric claims must have a same-sentence or same-section primary-source URL), and if the fact-checker misses it, the multi-model reviewer panel catches it, and if the panel misses it, the one-in-five human spot check catches it, and if all of those miss it, the correction goes on the corrections page with the date and the source. That layered review is the trust infrastructure.

Why daily, why AI-assisted

One operator cannot ship a high-quality daily newsletter without help. The honest options were three: ship a weekly, hire a team, or ship a daily with AI assistance on the drafting step. The publication picked the third option because the daily cadence is the right reading rhythm for the audience (contractors who check their inbox in the morning, between site visits) and because the trade-off, AI-assisted drafting under a published review process, is the one a careful reader can audit on this page.

The whole governance layer in one paragraph

Named editor with a real name and a public bio. Public corrections page with a severity definition. Public reviewer rubric. Multi-model reviewer panel from issue one. Deterministic pre-ship checks that fail the build on missing sources, em-dashes, or anti-slop violations. A one-in-five human spot check. A quarterly audit of how often the rubric is being passed in letter but failed in spirit. A reserve bank of evergreen issues so a bad day can be a rest day instead of a slop day. That is the system. It is not a marketing claim, it is the build.

Contact

Questions about how an issue was made: joseph@ownthebid.com or (747) 745-5837.