Skip to content

SITREP ISR — Pipeline Docs

Technical documentation for how SITREP ISR actually works under the hood.

This site is aimed at:

  • Readers who want to know how reports are produced before deciding how much weight to give them.
  • Contributors who want to suggest sources, flag bugs, or reuse parts of the pipeline.
  • Future me, who will forget all of this inside a month.

What SITREP ISR is

A small, independent site that publishes structured situation reports ("SITREPs") on Israel and the wider Middle East, plus occasional scenario simulations. Reports are generated on a schedule (currently four times a day) by an AI agent that reads a curated list of open sources and synthesises them into a standardised format.

It is not a primary reporting outlet. Nothing on the site is first-hand journalism. The value is synthesis, cross-source framing, and cadence.

Start here

  • Architecture

    How the pieces fit together — public site, admin surface, agent, newsletter.

    High-level map

  • Pipelines

    The actual workflow: source collection, triage, analysis, writing, publication.

    Pipeline overview

  • Sources

    The canonical list of monitored sources lives in a separate public repo.

    Grounding-Sources

  • Operations

    How runs are triggered, what the admin surface does, where to look when something breaks.

    Triggering runs

Guiding principles

  • Open sources only. No private intel, no off-record briefings.
  • Adversary framing is signal. State-controlled outlets are ingested and flagged, not excluded.
  • Uncertainty is named. Forward-looking claims carry an explicit confidence rating.
  • Corrections are visible. Reports are back-editable and timestamped in place.

Repos

Repo What it is
SITREP_ISR Public site + agent pipeline (this is the main one).
SITREP-ISR-Admin Backoffice UI, newsletter send, manual SITREP trigger.
Grounding-Sources Public source inventory, contributor-editable.
Docs You are here.

Design inspiration. The pipeline-as-graph presentation in these docs borrows from Snowglobe and LLM Council by Andrej Karpathy. Neither is responsible for how the ideas are applied here. See Credits.