Sourceframe

Workflow pattern / Briefing

Build a daily news and signal briefing

Assemble a short, source-linked daily brief from a defined set of public sources on a defined topic.

Intermediate / Digest / Daily

The problem

Manual daily scanning is expensive and inconsistent, while unfiltered feeds produce volume rather than signal.

Who it is for

Teams that need a defensible daily read on a specific topic.

Teams: Research / Editorial / Strategy

Inputs

  • Topic definition and exclusions
  • Source list plus discovery queries
  • Brief length limit
  • Delivery time

Conceptual process

  1. 01Collect

    Gather candidates from defined sources and a fixed query set.

  2. 02Filter

    Drop items outside the topic definition and items already covered.

  3. 03Retrieve

    Fetch and normalize each surviving item.

  4. 04Summarise

    Write one or two sentences per item, each carrying its source link.

  5. 05Rank

    Order by relevance to the stated topic, not by recency alone.

  6. 06Deliver

    Emit a fixed-length brief and archive the full candidate list.

Flow diagram

  1. 01Sources
  2. 02Collect
  3. 03Dedupe
  4. 04Retrieve
  5. 05Summarise
  6. 06Rank
  7. 07Brief

Example output

Illustrative output
# Daily brief — topic: logistics automation 1. A regional operator published a new API for shipment tracking.   https://example.com/newsroom/tracking-api (retrieved 07:02)2. A standards draft on freight data exchange entered public comment.   https://example.org/drafts/freight-exchange (retrieved 07:04) _4 further items archived as below-threshold._

Data-quality considerations

  • Deduplicate syndicated copies of the same item before summarising.
  • Keep the archive of rejected candidates so the filter can be audited.
  • Never summarise from a snippet you did not retrieve.

Failure modes

  • Same story counted three times from three republishers.
  • Topic drift as the query set ages.
  • Summaries that add claims not present in the source.

Suggested architecture

  • Source registry
  • Query runner
  • Dedupe by content similarity
  • Normalizer
  • Summariser with citation binding
  • Digest delivery

What to test first

  1. 01Run silently for a week and compare against what a human would have picked.
  2. 02Check that every summary sentence is supported by its linked source.

Related reading