How it works

Four stages, one mandate: only surface what changes obligations.

regflow runs a daily pipeline against U.S. regulatory publishers. Each stage is idempotent — if anything fails, the next run picks it up. The numbers below come from the live database.

classified
164
last 24h
0
matches today
0
live feeds
6

§ 01 Pipeline

STAGE 01

Fetch

Federal Register API. Pages, normalizes, dedupes by document number. Stores raw payload + content hash. Today: ~200 docs/day.

STAGE 02

Classify

Each doc → GPT-5.5 with a strict mandate. Output is schema-validated JSON: summary, materiality 1-5, topics, entities, rationale.

STAGE 03

Match

Hard gate on materiality threshold + keyword/topic overlap. Composite score 40/60 keyword/topic. Idempotent — no duplicates.

STAGE 04

Deliver

RSS 2.0, JSON API, email digest (when shipped), Discord/Slack/generic webhook. One alert per match per channel.

§ 02 The materiality scale

The same scale across every document, every agency. Calibrated for compliance officers triaging a daily queue — not for academic classification.

M5 Major 2 docs · 1%

A new rule with substantial obligations. Withdrawal of a major rule. Enforcement-posture shift. Expansion of regulated activity.

M4 Significant 31 docs · 19%

Compliance-obligation change. New filing or reporting. Deadline shifts that move money or staffing.

M3 Notable 29 docs · 18%

Definition shifts, new exemptions, low-impact procedural changes worth reading.

M2 Minor 89 docs · 54%

Clarifications, format changes, withdrawals of stale notices, references.

M1 Noise 13 docs · 8%

Typos, scheduling, non-substantive corrections, ceremonial. Default-skip.

Percentages above reflect the live corpus. M2 dominates in any honest regulatory feed — most documents are procedural. The point of regflow is to surface the M3+ minority without manual triage.

§ 03 Topics extracted from the live corpus

Topic tags are LLM-derived per document — specific enough to filter, not generic buckets. These are the top tags across the current corpus, verbatim from the classifier.

  • paperwork-reduction-act-review · 12
  • foreign-military-sales · 9
  • paperwork-reduction-act-clearance · 6
  • medical-device-classification · 6
  • defense-export-controls · 5
  • arms-sales-notification · 5
  • antidumping-duty-reviews · 4
  • countervailing-duty-reviews · 4
  • arms-sales-notifications · 4
  • china-imports · 4
  • paperwork-reduction-act-icr · 3
  • nih-grant-peer-review · 3
  • customs-cash-deposits · 3
  • sunset-reviews · 3
  • trade-remedy-proceedings · 3
  • paperwork-reduction-act · 3
  • paperwork-reduction-act-renewal · 3
  • ferc-docket-filings · 3
  • class-ii-special-controls · 3
  • fda-special-controls · 3
  • ferc-intervention-deadlines · 2
  • postal-negotiated-service-agreements · 2
  • omb-information-collection · 2
  • bank-holding-company-applications · 2
  • bank-merger-review · 2
  • esa-listed-salmon-take · 2
  • hatchery-genetics-management-plans · 2
  • marine-mammal-incidental-take · 2
  • environmental-monitoring-reporting · 2
  • paperwork-reduction-act-comments · 2

Don't see your industry? View the source roadmap or request a topic.

Ready to wire it in?

Pick a vertical, subscribe to its RSS feed, watch it for a week. Pro tier is free for 30 days.

Browse feeds See pricing