Bioinformatics Intelligence Radar

A reusable ChatGPT Skill and deterministic evidence backend for source-grounded bioinformatics intelligence and concise Persian newsroom coverage.

Protocol

Radar 3.1.0 separates two responsibilities:

1. Intelligence Engine discovers, verifies, identifies, deduplicates and preserves evidence.
2. Newsroom Engine decides what deserves attention, chooses an evidence-safe angle and writes readable Persian news.

The new scripts/ layer makes identity, evidence gates, state transitions and News Value scoring reproducible without replacing web discovery or editorial judgment.

What it monitors

Radar can cover peer-reviewed literature, official software releases, databases, datasets, infrastructure and security events, benchmarks, corrections and retractions. Scholarly items require positively verified peer review. Preprints may support identity resolution only; they never appear in user-visible output as eligible papers.

3.1 capabilities

• JSON Schemas for Evidence Card, Story Card and run state;
• durable canonical identity and material event fingerprints;
• within-run deduplication and cross-run suppression;
• publication/release transition and correction classification;
• explainable six-dimension News Value scoring;
• CLI validation and processing;
• golden fixtures and regression tests;
• example configuration and operations guidance;
• compatible Persian newsroom and Telegram Handoff contracts.

Quick start

python3 -m unittest discover -s tests -v
python3 -m scripts.radar_cli validate tests/fixtures/items.json
python3 -m scripts.radar_cli run tests/fixtures/items.json \
  --state-out /tmp/radar-state.json \
  --output /tmp/radar-run.json
python3 -m scripts.render_output /tmp/radar-run.json \
  --markdown /tmp/radar-report.md \
  --handoff /tmp/telegram-handoff.json

The CLI consumes normalized JSON items from a discovery adapter. It does not browse the web, infer peer review from a DOI, or invent claims. The Skill and reference policies remain the authority for source interpretation and Persian newsroom writing.

Default newsroom output

The visible report is intentionally compact: خبر اول, ارزش دنبال‌کردن, an operational warning when needed, an evidence-backed signal when justified, editorial selections, Deep Dive questions, a short watchlist and observed-only telemetry. A quiet day is valid; there is no filler quota.

Repository layout

• SKILL.md — control plane and release gates.
• references/ — source, peer-review, evidence, newsroom, scoring, state and handoff policies.
• scripts/radar_core.py — deterministic identity, gating, deduplication, state and scoring.
• scripts/radar_cli.py — validation and processing CLI.
• scripts/render_output.py — Persian Markdown renderer and Telegram Handoff v1 builder.
• schemas/ — machine-readable Evidence Card, Story Card and state contracts.
• tests/ — regression tests and golden fixtures.
• config/radar.example.json — safe example configuration.
• docs/ — architecture, operations and contribution guidance.

Design principles

Primary sources come first. Event, publication, observed and effective dates remain distinct. Evidence is a gate, not a proxy for newsworthiness. Claims retain provenance and limitations. Headlines lead with the news, not the method. Association is not causation, prediction is not measurement, preclinical is not clinical, and author-reported benchmarks remain attributed.

See MIGRATION-3.1.md for adoption steps and docs/ARCHITECTURE.md for extension points.