// agent surface v1.18.0
A machine surface for reading l0g without scraping.
Agent Surface v1.18.0 exposes the French and English l0g corpus as static JSON and NDJSON: bilingual catalog, shared search index, evidence graph, at most three structural typed claims per article, primary sources, freshness, Risk Diff, Black Box Recorder, named and versioned signal series, fingerprints, Agent Bench results and an OpenAPI contract. Translations carry canonical and alternate identifiers plus freshness status. Claims remain canonical in French; English analyses link to them through evidenceRef. Heuristic typing remains distinct from canonical review.
// endpoints
/agents.jsonCapability manifest, endpoints, usage rules and proof policy.
Contract/openapi.jsonOpenAPI 3.1 description of machine-consumable endpoints.
Catalog/api/v1/catalog.jsonBilingual corpus map with language, canonicalId, translationOf, alternateUrl and translationStatus.
Shared search index/api/v1/search-index.jsonCanonical bilingual search index used by Agent Surface, MCP and WebMCP.
English corpus/llms-full-en.txtFull English analyses and guides, kept separate from the French context.
Catalog NDJSON/api/v1/catalog.ndjsonLine-by-line catalog for streaming ingestion, RAG or incremental indexing.
Claims/api/v1/claims.jsonAt most three structural claims per article, typed as fact, estimate, inference or scenario, with references and dates when detectable.
Evidence graph/api/v1/evidence-graph.jsonArticles, claims, references, hosts, primary sources and datasets as nodes and edges.
Sources/api/v1/sources.jsonPrimary sources and cited hosts actually used by the corpus.
Freshness/api/v1/freshness.jsonFreshness policy, last content updates and signal timing metadata.
Integrity/api/v1/integrity.jsonCanonical SHA-256 fingerprints of M2M surfaces.
Changefeed/api/v1/changes.jsonMachine-readable publication and revision feed.
Risk Diff/api/v1/risk-diff.jsonRisk changes over 1, 7 and 30 days.
Black Box/api/v1/black-box.jsonPoint-in-time hashed frames for replaying public risk states.
Named series/series/Stable identifiers, citable names, methodology versions and dedicated downloads for each gauge.
Signal history/api/v1/signals/history.jsonPoint-in-time signal observations, level changes, coverage and policy.
Signal NDJSON/api/v1/signals/history.ndjsonLine-by-line signal history for agents, watchers and incremental ingestion.
Signal CSV/api/v1/signals/history.csvFlat signal observations for pandas, R, DuckDB or spreadsheets.
// recommended flow
- Discover the surface with /agents.json.
- Read /api/v1/freshness.json to date the corpus and dashboard signals.
- Verify fingerprints with /api/v1/integrity.json when a response depends on a fixed snapshot.
- Watch /api/v1/changes.json for new content and revisions.
- Use /api/v1/risk-diff.json to understand what changed in risk over 1, 7 or 30 days.
- Use /api/v1/black-box.json to replay a public date without filling historical gaps.
- Choose a named series in /series/, then use its dedicated CSV, JSON or NDJSON export.
- Use /api/v1/signals/history.csv or .ndjson to replay every series without look-ahead bias.
- Use /api/v1/evidence-graph.json to traverse article → claim → source relationships.
- Switch to NDJSON feeds for line-by-line ingestion.
- Use /api/v1/search-index.json to search the bilingual corpus shared with MCP and WebMCP.
- Use /llms-full.txt for French or /llms-full-en.txt for English without loading both corpora.
- Quote factual claims via /api/v1/claims.json when the answer depends on a precise assertion.
- For an English translation, follow evidenceRef to the canonical French claims and evidence graph.
- Preserve the canonical URL, date fields and claim kind in generated answers.
// guardrails
Fields such as kind, dateLabel, references and confidence are not decoration. They prevent a scenario or estimate from being turned into an observed fact, and remind consumers that extracted relations remain verifiable in the canonical page.
// minimal contract
// example
curl -s https://l0g.fr/api/v1/claims.json// integration
For a RAG agent, use /api/v1/search-index.json for bilingual retrieval, then /api/v1/evidence-graph.json or /api/v1/claims.ndjson for sensitive assertions. /api/v1/changes.ndjson avoids rescanning the whole corpus. /openapi.json is the typed entry point for client generation.