Proof report
HFAO — Hugging Face Agent Observatory
Every verifiable claim behind this project, grouped by epistemic status, each with a source you can check — 7 claims, 7 source-linked.
Project status: In development · last verified 2026-08-28
← HFAO — Hugging Face Agent Observatory overview Repository ↗
Shipped
(7)The named artifact (test, doc, code path) exists in the repo today and is runnable. Per-claim, not per-project — a claim can be shipped while the parent project is still in development.
-
Storage plane (ClickHouse + DuckDB) with parity tests — the tests that need a live ClickHouse, including the DuckDB↔ClickHouse parity test, skip unless HFAO_TEST_CLICKHOUSE_DSN is set
-
Ingest plane (OTel GenAI + OpenInference normalizer, OTLP server, PII redaction, bounded buffer, body offload at 64 KiB)
-
SPEC.md v1.0.0 locked; every build-plan commit cites a spec section (§15.3 rule 1)
-
MCP-native queryability across observability primitives (FastMCP Streamable HTTP, §9.2 read surface + gated score_observation write)
packages/hfao/mcp_server/ + tests/acceptance/test_ac_9_mcp.py (v0.5.0, SPEC §9) ↗
-
Closed eval-trace loop (traces ↔ datasets ↔ scores ↔ monitors), with causal attribution, cost rollups, and retention
-
Stage 2 counterfactual replay — driver-pluggable, Tier-1 frameworks (LangGraph / OpenAI Agents SDK / Claude Agent SDK); tests use deterministic injected callables, not live framework or model calls
-
Proactive anomaly surfacing (Insight schema + AnomalyEngine) and rule-based insight routing (subscriptions)