Methodology

Every Feed story is scored in public. The rubric is the product.

Why score at all

Most political coverage asks you to take a side. We ask whether a story reveals a mechanism, something about how the system works that would still matter if the party labels flipped. Scoring makes that judgment auditable: four axes, one to three points each, total out of twelve, with a written reason per axis.

Include at 8/12 or higher. Fast-track interest at 12/12. Agent-autonomous publication also requires Structural ≥ 2 and Durability ≥ 2. A Structural-1 exception needs a human override.

The four axes

Axis	1	2	3
Documentation	Single outlet, unnamed sources, no primary	Multiple outlets or partial primary	Primary document available and cited (statute, filing, dataset, transcript)
Structural significance	Episodic event	Recurring pattern / symptom	Reveals the rule, incentive, or design that makes the outcome predictable
Citizen impact	Abstract public only	Identifiable population	Concrete daily-life effect for ordinary people, stated plainly
Non-partisan durability	Evaporates if party labels swap	Holds with work / both-party construction	Mechanism is clearly bipartisan or office-agnostic

Reversal test (mandatory)

Before drafting, the structural argument is rewritten with the opposing party in the relevant office. Verdict: HOLDS or EVAPORATES. Evaporates means kill or restructure. The test is written into the session record, not performed as a vibe check after the fact.

Source stack

Primary for structural claims: statutes, dockets, agency releases, datasets, transcripts.
Major independent outlets for corroboration.
Aggregators and scouts (Drudge, headline digests) are radar only. They never carry a central claim alone.
Never fabricate a URL. An unlinked true citation beats a linked guess.

Automatic disqualifiers

Advocacy endpoint ("vote for / oppose / join")
Documentation that is only editorial opinion
48-hour ephemera with no structural residue
Named politician as the cause rather than as instance of a mechanism

Voice constraints (delivery)

We require both-parties accountability, an honest counter-argument, a founding anchor, and a human opening that returns mid-piece and at the close. We ban stock section headers ("The Counter-Argument," "Both Parties, One…") so the template does not become audible. Full essays and shorter Mechanism Briefs both use the same evidence rules; length is not a quality substitute.

Civic Engine

Civic Engine essays are not scored on the 12-point Feed grid. They are slow, multi-source structural analyses with linked Sources blocks, counter-arguments, and founding-text grounding. They are the evergreen half of the site.

Governing documents

Internal production is governed by a versioned agent constitution, editorial standards, and a scout harness with machine-checkable gates. Those files live with the site's source repository; this page is the public summary. When practice and this page diverge, this page should be updated, and the error logged on Corrections.