Biennale Radar · Living Document · v1.0

Operational Manifesto

Mission Statement

"To serve as the definitive, intelligent conduit for the global art biennial ecosystem by autonomously curating, verifying, and distributing institutional signals with editorial rigor — ensuring Biennale Radar remains the most trusted, current, and comprehensive knowledge hub for cultural partners and professionals."

Accuracy over Volume — we remove data that lacks confidence rather than publish noise.
Architect over Operator — humans set the rules; agents execute at scale.
Telemetry over Intuition — every failure is logged, every fix is systematic.
Trust as a Product — our reputation is the provenance and accuracy of every record we deliver.
Editorial Rigor — we publish what is verified, contextualized, and meaningful to the global art community.
Institutional Primacy — primary sources (biennale offices, museum press, official art fair channels) always outrank secondary coverage.
Continuous Improvement — after every pipeline cycle, we measure, learn, and recalibrate.
Auto-PublishTier 1 — Autopilot

Official biennale websites, vetted top-tier art press (Artforum, The Art Newspaper, Frieze, Ocula, e-flux), museum institutional feeds.

Verified source trust score ≥ 80. Content flows directly through enrichment and to publication with no agent hold.

EditorInChief AgentTier 2 — Agent Review

Emerging galleries, analytical reviews, secondary press, newsletter signals from subscribed institutional inboxes.

Trust score 50–79. EditorInChief Agent summarizes content into a digest. Human Architect approves batch — not individual items.

Archive-OnlyTier 3 — Quarantine

Unknown domains, PR spam, transactional senders, sources with 3+ consecutive classification failures.

Auto-archived. Never published. Source paused and flagged for Architect review.

These numeric guardrails govern all automated pipeline decisions. Update via the EditorialEthos entity — no code changes required.

Minimum Relevance to Auto-Publish

Items below this relevance_score are archived, not published.

65/ 100

Minimum Confidence to Auto-Publish

Items below this confidence_score are held for agent review.

60/ 100

Minimum Editorial Score for Featured

Only items reaching this score are promoted to Featured status.

85/ 100

Max Consecutive Source Failures

Source is auto-paused after this many consecutive pipeline failures.

3failures

Target Category Precision

% of items correctly classified beyond 'other'. Below this triggers re-enrichment.

75%

Max Hours in Enrichment Stage

Items stuck longer than this are automatically reset to pending.

2hours

Every record delivered to the Biennale Platform must meet all of the following criteria.

Every record is geolocated (lat/lng verified)
Every record has an editorial summary (18–28 words)
Every record has a verified start_date and end_date
Every record carries source provenance and data_confidence classification
Every record has been deduplication-checked before publish
High-confidence records carry accessibility notes where available

The Architect's role is to set policy, not operate the pipeline. These are your four control surfaces.

Rules Registry

Edit EditorialEthos entity

Adjust quality thresholds, source tiers, and mission. No code changes required.

Curator Dashboard

Review AI Digest

Review agent-prepared summaries of Tier 2 content. One-click batch approve or reject.

Operations Dashboard

Monitor Pipeline Health

View green/yellow/red pipeline health. If it's green, the system is winning.

Exception Review

Handle Escalations

Edge cases surfaced by agents that require human judgement — rare by design.

This document is a living standard. All pipeline functions, agents, and entities operate in alignment with these principles. Last reviewed: September 12, 2026.