A short, useful body of documentation.
Three documents cover everything required to integrate ASGE, defend its output to a research committee or an intelligence reviewer, and map its capabilities onto concrete applications. They are intentionally short.
Submit your first passage in under five minutes.
Each session token is provisioned against a tier. Once issued, a single POST request returns a complete analysis: lexical-density timeline, peak concentration moments, the dominant semantic-conflict theme, a structured analytical summary, and the graduated alert layer.
1. Obtain a session token
API access is provisioned per organisation. Sandbox access is available on request and runs the full three-stage pipeline on a single language for evaluation. Write to [email protected] with the tier you intend, the contracting entity, and the expected use case.
2. Submit a passage
The endpoint is POST https://api.llulllab.com/asge/v1/analyse. The full schema is on the API page. The minimum request carries the text, a language code, and a session token.
Line breaks segment temporal units. For passages that lack natural minute structure (essays, articles), break the text into paragraphs roughly equivalent to one minute of speech and the temporal density map will read naturally.
3. Read the response
The response is a single JSON object. The most useful fields, in order of how analysts typically consume them:
- analysis.peak_minute — where to look first.
- analysis.dominant_theme — the opposition the passage is organised around.
- analysis.conflict_summary — two to three sentences naming the sustained tensions.
- analysis.alert_level — graduated 1 to 4. Higher values warrant closer attention.
- analysis.trend — minute-by-minute density to overlay with other timelines.
How ASGE actually arrives at an analysis.
ASGE detects, maps, and analyses ambivalence structures in natural language: the sustained co-presence of opposing orientations toward the same object. The concept is developed in anthropology and political science. Sustained ambivalence structures act as early indicators of neurotic crisis at the individual level, offensive escalation at the collective level, and criminal action across both — before these stabilise into observable behaviour.
Each passage is processed through a three-stage computational pipeline. Every stage's output is part of the public contract and is returned in the response.
-
Real-time lexical detection
Proprietary multilingual lexical databases identify opposing orientations as they emerge across the passage. Each detection feeds the temporal map and contributes to the graduated alert layer.
-
Temporal density mapping
The system constructs a minute-by-minute density map of detected ambivalence. Three visualisations are available: wave (continuous flow), bar (discrete per-minute counts), and delta (rate of change). Peak concentration moments are surfaced explicitly.
-
Deep contextual analysis
A deep contextual analysis layer goes beyond dictionary-level pairs to identify contextual ambivalence: opposing orientations that the surface vocabulary does not name. The stage produces the dominant theme, the structured analytical summary, and the alert level.
What the output gives an analyst
Standard sentiment classifiers flatten ambiguity into positive, negative, or neutral. The signal that actually precedes escalation or crisis sits in the gap those models discard. ASGE adds a layer that existing OSINT, risk, and intelligence toolchains do not provide: a structural reading of sustained co-presence of opposing orientations, mapped against time, with a written summary and a graduated alert.
What is not on record
The infrastructure underneath the deliberation is proprietary and is not part of the public contract. The output is the contract: temporal density maps, peak concentration moments, dominant semantic-conflict themes, structured analytical summary, and the graduated alert layer.
Where ASGE adds a layer that existing toolchains do not provide.
Three buyer profiles deploy ASGE in production. Each uses the same API surface; the difference is the timeline and the source material.
Anthropology, political science, and computational social science. Comparative analysis of speeches, interviews, and ethnographic transcripts where the ambivalence structure is the unit of analysis.
Open-source intelligence units, threat-assessment teams, and risk analysts. Sustained ambivalence structures act as early indicators of escalation. ASGE adds a structural reading alongside existing sentiment, network, and event-based pipelines.
Strategic communications, public-affairs analytics, and media-research teams. Tracking how an audience's orientation toward an object shifts across a media cycle, where the binary positive-or-negative reading flattens the relevant signal.
Peak minute: highest-density timestamp. Dominant theme: opposition the passage is organised around. Alert level: 1 to 4. Classic vs contextual: dictionary detection vs deep contextual reading.
The same material, in printable form.
Each section above is also available as a standalone PDF, formatted for circulation inside an organisation.
Sandbox access available on request.
Integration enquiries, sandbox access, contract negotiations, on-premise discussions, or anything that requires more than a documentation page.
Paris-based research laboratory. Written enquiries are answered within two business days.