InfoliticoInfolitico

How Infolitico works

The automated daily-news pipeline: source evidence, optional verse selection, drafting, guards, and publication.

The default daily-news lane is an automated source-to-publication pipeline. It admits source-backed events, drafts and guards a story, then either publishes it or rejects it. A separate owner-gated script supports hand-authored Inspirational Thought articles; it is not part of this pipeline.

  1. Source evidence

    A curated feed event must carry a nonempty factual summary — not just a headline.

  2. Faith-angle judgment

    The theme extractor decides whether the event has a natural faith/values angle at all.

  3. Verse selection

    When recommended, a curated NLT verse is selected by hashing the source event ID within the eligible theme bucket; recent-use exclusions can change that bucket.

  4. Outline and draft

    Structure the arc, then draft toward the 300–500-word editorial target, with reflection and prayer.

  5. Guard and review

    Content gates, headline novelty, and person checks run before anything can publish.

  6. Publish

    The image gate clears and the story goes live on the website and public RSS feed; newsletter delivery is separate.

From a real news event to a published article: every stage either advances the story or rejects it. Rejections are expected — a story with no genuine faith connection is never forced into a devotional frame.

The source repository includes a separate digest sender and timer. Its dated operations record logs one past production run that skipped an edition; that does not verify the current schedule or a delivered digest. See the newsletter guide.

The stages

Source-evidence preflight

Before any model work, the pipeline requires a nonempty factual source summary that is more than the headline repeated. Feed rows without real evidence are rejected (source_event_summary_missing_or_too_thin) and never consume a generation slot.

Faith-angle judgment (theme-extractor)

The extractor reads the source event and decides whether it has a natural faith/values angle at all. The most common rejection is no_natural_faith_connection — in one observed day, 16 of 20 candidates were rejected on this ground and none were forced through. When the verdict is write, the stage also extracts the theme (for example stewardship or perseverance) and produces the factual news summary used downstream.

Verse selection (verse-selector)

When recommended, a Bible verse is selected deterministically: the index into the curated bible_verses database is derived from a SHA-256 hash of the source event's ID. The candidate bucket can first exclude verses used in unretracted stories during the default seven-day recency window. If that set would be empty, or the recency lookup fails, selection falls back to the full theme bucket. Thus the event ID determines the index for a fixed candidate set, but a changed recency set can change the selected verse. When a verse is not recommended, the article publishes with a reflection but no verse blockquote. Optional theologian commentary quotes are attached in code, capped at three candidates, and never trusted from a model response.

Outline and draft

The outline stage structures the arc: headline, news paragraphs, reflection arc, and prayer focus. The draft stage writes toward the 300–500-word editorial target from that outline. The verified NLT verse block is the only full-verse quotation; the reflection discusses its meaning primarily in original prose and may weave in at most one commentary quote — verbatim, attributed, never invented.

Guard and review

The polish-and-guard stage is the last-line content gate. Editorial review, headline-novelty checks, and a pre-publish person check run next. The image gate clears, and the story publishes.

Determinism and safety by design

  • Article structure is reproducible. Verse choice is stable only while its eligible candidate set is unchanged; recent-use exclusions can change it.
  • Scripture text is reproduced exactly from the curated database; the pipeline never fabricates or modifies a verse (see editorial standards).
  • Verified data (verses, commentary quotes, protected-person policy) is attached in code between stages, never trusted to a model's own output.
  • Every story must clear the guards before publication; a story that fails a gate is rejected, not repaired by prompt.
Was this page helpful?

On this page