How the digest is built
The MAHA / HHS weekly digest is a structured record of activity at the U.S. Department of Health and Human Services under Secretary Robert F. Kennedy Jr., aggregated from 22 vetted sources every Sunday at midnight (Mountain). This page documents what's in the digest, where it comes from, what isn't, and how to read it.
What this is
A weekly, sourced record of federal health-policy activity covering MAHA movement statements, HHS Department announcements, and sub-agency action at CDC, FDA, NIH, CMS, SAMHSA, HRSA, AHRQ, and the White House. Items are extracted from primary government press, the Federal Register, movement-aligned advocacy, balanced press coverage, and policy think-tanks — then organized into ten topic lanes plus advocacy and press sections.
Every item links back to its primary source. Items reported by multiple outlets are clustered into a single lead with "Also covered by" citations beneath. The digest tracks named initiatives across six persistent ledgers (Executive Orders, HHS Announcements, Personnel, Budget, Regulatory, Research). The pipeline runs every Sunday at midnight (Mountain) and publishes to this page.
Source tiers
Sources are classified into four tiers by editorial role. Tier 1 is the source of truth; everything
else is context, balance, or interpretation. Tier-2 advocacy items are explicitly tagged
[MAHA-aligned] and never blended into the main lanes — they live in their own section
so the reader can weigh framing.
| Tier | Role | Treatment |
|---|---|---|
| Tier 1 | Government primary — HHS press, sub-agency newsrooms, Federal Register, White House actions | Source of truth. Items go directly into the relevant topic lane. |
| Tier 2 | Movement-aligned advocacy — MAHA Action, Children's Health Defense, Heritage Foundation Health | Tagged as advocacy. Live in a separate "Notable from Advocacy" section. |
| Tier 3 | Health-policy news — STAT, KFF Health News, Politico, The Hill, Axios, NPR Health | Editorial coverage. Live in "Notable from Press" section. |
| Tier 4 | Watchdog & think-tank — KFF, Health Affairs, ProPublica, Brookings, AEI | Analysis and longform. Live in "Notable from Press" section, distinguished by source. |
The 22 sources
Every source is fetched fresh on each run via RSS (when available) or HTML, parsed for the past 14 days, and clustered for cross-source duplication.
Pipeline cadence
The pipeline runs every Sunday at 23:55 (Mountain) via Windows Task Scheduler. A secondary scheduled job inside the AI Swarm fires the same pipeline at midnight as a redundant path. Each run takes ~3 seconds end-to-end:
- Fetch — 22 sources in parallel; httpx first, falls back to curl_cffi (Chrome TLS impersonation) for Cloudflare-protected sites.
- Extract — RSS items and HTML pages parsed structurally (title, date, link, summary). The 14-day window is computed in Mountain Time with a 1-day forward buffer.
- Cluster — items are grouped by normalized title and URL across sources via union-find. Each cluster picks a lead by tier preference; secondary coverage is attached as
also_covered_by. - Classify — heuristic keyword routing into 10 topic lanes and 6 ledger types (Executive Orders, HHS Announcements, Personnel, Budget, Regulatory, Research).
- Publish — POST to
/api/maha/publishwith bearer auth; the live page renders the latest payload.
Inclusion criteria
An item appears in the digest only if all are true:
- Date falls within the past 14 days (Mountain Time, 1-day forward buffer)
- Comes from one of the 22 vetted sources or replaces a known-down source from the reserves list
- For sitewide RSS sources (Axios, AEI, Brookings, ProPublica), title or summary matches at least one health-relevance keyword
- Has a parsable date and a primary URL pointing to an authoritative document
Routine procedural noise is filtered out of the executive summary (Federal Register information-collection notices, OMB review submissions, meeting cancellations) but remains visible in the full lane view for completeness.
How to read the page
The page has four layers of reader controls. They compose:
- Audience Mode pills — coarse-grained: All News, Government Only, Quick Brief, or Movement view.
- Filter chips — fine-grained multi-select across Topic Lane, Source Tier, Sub-agency, and Date range. State mirrors to
?lane=&tier=&agency=&since=URL params for shareable views. - Reader tools — Saved items (★ on each item), Tracked terms (highlighted on the page), and a "new since last visit" ribbon.
- Per-item share — hover over any item to copy its permalink or share to X / LinkedIn.
Known limitations
- Heuristic classification. Topic lane and action-type assignment use keyword matching. Items can land in slightly off lanes when wording is unusual; the section is sourced and dated, so reclassification is straightforward to spot.
- Single-machine cron. The pipeline currently fires from one machine; a temporary network or power outage on Sunday night skips that week's run. The status pulse on the page indicates staleness.
- 14-day window. Items dated more than two weeks before publication time are excluded, so slow-rotating sources (Health Affairs journal feed) may report zero items in a given week.
- Cross-source dedup is exact-match only. The same story reported with substantially different wording across two outlets can still surface as two items. We err on the side of duplicates over false merges.
- No editorial review. No human edits the digest before publish. Items are aggregated mechanically from primary sources, then organized via heuristics. The reader is the last line of editorial defense.