Editorial standards

How an issue is made

The pipeline pulls every morning from SAM.gov, USASpending, published GAO bid protest decisions, the Federal Register, GSA's prospectus library and lease inventory, and the agency budget documents. It stores the source artifact for every record, scores what is new, and maintains a register of every open construction solicitation and its response date. On Friday it assembles the Monday issue from that week's records.

An independent verification pass then checks every factual claim in the draft against the stored source document, without access to the reasoning that produced the draft. Anything it cannot confirm is removed. The rules it enforces are code, not judgement: every proper noun, dollar figure, date, PIID and solicitation number must resolve to a stored record; no claim about a named firm that is not in the award record; no characterisation of why a firm won or lost.

The assembled issue is then held. It does not send until a person has read it in full — every deadline with its source link, every company named, and every flag the verifier raised — and approved it. If nobody approves it, the issue does not go out. That is the intended failure mode. A missed issue is recoverable; a wrong one is not.

Where AI is used, and where it is not

Machine assembly does the ingestion, the parsing, the scoring, the drafting and the verification pass. This is disclosed because it is true and because a disclosure that tells you what a human actually checked is worth more than a label.

What a model never does here:

• It never produces a number. Dollar figures, dates, counts and identifiers are parsed from the source record. If a figure appears in an issue, it came out of a document, not out of a model.
• It never sends anything. A person approves every issue before it goes.
• It never answers a legal message. Contact on the legal route is read by a person, same day, and answered by a person. It is never acknowledged automatically beyond the confirmation on the form.
• It never handles a correction or a complaint. Any contact from a company named in an item, any correction request, and any contact from a law firm, an agency or a journalist goes straight to a person.

What we will not print

• A claim about a named firm that is not in the award record.
• An explanation of why a company won or lost a competition.
• The name of a private individual.
• Occupant identity for law-enforcement or defense-tenanted leases.
• Anything sourced to an unnamed person. There is no unnamed sourcing here, because there is no reporting by conversation — the sources are documents.
• Another publication's article, reproduced or reconstructed through paraphrase.

House style

Short declarative sentences. No adjectives doing an argument's work. No speculation about who will win. Every item ends with a link to the document it came from, so the claim can be checked rather than believed.

Independence

No ads, no sponsored items, no affiliate links, no sponsored anything — in the issue, on the site, or in the data files. No money from any firm we cover. The publication is funded by readers buying data products and annual subscriptions.

The publisher, K2 Business Strategies, also does consulting work in the federal market. Where a data product or an item would touch a client engagement, the item is not published. We would rather leave a story out than write around a conflict.

Corrections

Corrections run at the top of the next issue and stamp the archived item. Nothing is silently edited. The running log is at /corrections and every entry states what was wrong, what is correct, and what in the method allowed it through.

Reaching a person

Two routes, both on the contact section: administration for subscriptions, billing, data products, privacy requests and corrections; legal for notices, licensing and formal demands. There is no email address published on this site and no chat widget, because both attract more scraping than correspondence. Every message is answered.