How CassandraBot Works
CassandraBot applies narrative diffusion theory, information cascade models, and computational linguistics to detect early-stage signal formation before it reaches mainstream visibility. Every detection is grounded in a peer-reviewed theoretical foundation — the eight hypotheses below each operationalize a specific academic framework.
We poll diverse sources — Hacker News, Bluesky, Mastodon, Arctic Shift, and RSS feeds — continuously. Each source is assigned a community tier (niche expert, enthusiast, or mainstream) based on its audience composition.
Incoming text is evaluated against the eight hypotheses by a large-language-model analyst operating under a theory-grounded system prompt. Each hypothesis is scored for confidence, tagged with its theoretical basis, and supported by extracted linguistic markers and escalation indicators.
When a signal is detected, we track its strength over time, its trajectory (accelerating, peaking, stable, decelerating, or fading), and its spread across community tiers. Signals reaching Tier 2 or Tier 3 gain analytic priority.
Active signals are matched against a historical archive of resolved events (e.g., the SVB collapse, the GME squeeze, Brexit). The prediction engine assigns probability distributions to resolution archetypes and generates observable confirmation indicators.
Strength is derived from hypothesis confidence, community tier spread, velocity of mentions, and sentiment intensity. A signal reaching Critical does not mean certainty — it means the narrative is moving fast and widely.
Theoretical Foundations
Every hypothesis CassandraBot evaluates operationalizes a published theory from economics, communication studies, network science, or behavioral psychology. The system is an instrument for applying these frameworks to live online discourse at scale.
How viral stories spread through populations and drive economic behavior, decoupled from underlying fundamentals.
- Terminology Velocity
How ideas and behaviors spread through social networks — Centola for multi-exposure adoption across community boundaries, Rogers for the S-curve of adopter categories.
- Cross-Community Diffusion
- Adoption-Cycle Position
How rational actors imitate predecessors, producing fragile herding equilibria that can collapse abruptly when new evidence appears.
- Catastrophizing Decay Rate
How conceptual metaphors and selective emphasis structure perception — Lakoff for metaphor-level frames, Entman for institutional framing contests.
- Frame Substitution
- Institutional-Grassroots Sentiment Gap
How media and institutional attention determine which issues the public considers important — and the lag between grassroots salience and institutional uptake.
- Regulatory/Policy Lag Indicator
How easily recalled events distort probability judgments, producing availability cascades when high-salience but emotionally restrained signals tip into mass attention.
- Affect Suppression Signal
The Eight Hypotheses
Every piece of text CassandraBot analyzes is scored against these eight hypotheses. A single text may trigger multiple hypotheses. The more hypotheses firing with high confidence, the stronger the overall signal.
The same story or frame appearing across ideologically distinct communities — multi-exposure adoption rather than within-cluster echo. Complex-contagion models predict that ideas requiring social reinforcement only cross community boundaries once enough independent ties carry them; observing that crossing is a leading indicator of broader diffusion.
- cross-community terminology overlap
- shared framing across political divides
- same frame in financial + tech + health spaces
Niche, technical, or domain-specific vocabulary appearing at unusual frequency — specialist words spreading before journalistic abstraction. Shiller's narrative-economics framework treats viral vocabulary as the measurable surface of a narrative epidemic; rising terminology velocity precedes institutional coverage by weeks.
- domain jargon in general forums
- technical terms in non-technical communities
- expert language trending on social platforms
A measurable gap between official/institutional framing and grassroots emotional expression — distrust of the official narrative, or emotional tone contradicting stated facts. Entman's framing framework identifies these gaps as active frame contests: the ground is shifting beneath an institutional consensus.
- media reports calm / comments report panic
- official framing vs. community anger
- official statistics contradicted by lived-experience accounts
The dominant conceptual metaphor or frame for a referent has been replaced. Home as investment becomes home as trap. Crypto as future becomes crypto as casino. Work as career becomes work as survival. Lakoff's metaphor-frame analysis treats frame substitution as a leading indicator of behavioral change.
- new dominant metaphor emerging in discourse
- old metaphor used ironically or rejected
- metaphor crosses from one domain to another
What stage of Rogers' S-curve do the questions indicate? 'What is X' = innovators/early. 'How do I X' = early majority/growth. 'Why did X fail' = late majority/disillusionment. 'How do I recover from X' = laggards/bottom. We map the question landscape to locate the topic on the adoption curve.
- shifting from what → how → why → recovery
- questions becoming more emotional
- advice requests peaking then declining
High engagement volume paired with suppressed or flat emotional expression — high salience but restrained valence. Kahneman and Tversky's availability work predicts that this tension state often precedes sharp availability cascades in either direction.
- high engagement, flat sentiment scores
- measured language in emotional situations
- suppressed reactions before a floodgate opens
Technical or grassroots vocabulary that institutional agenda-setters (regulators, mainstream press) have not yet adopted. McCombs and Shaw's agenda-setting framework predicts that this agenda lag translates into policy, regulation, and market response delays of roughly 12–24 months.
- regulators using outdated terminology
- mainstream media lacking vocabulary for a trend
- institutions reacting to language from 1–2 years ago
Catastrophizing or intense language that appears to be peaking or decaying into disengagement, suggesting an information cascade nearing exhaustion. Bikhchandani's and Banerjee's cascade models predict that fragile herding equilibria reverse abruptly once new evidence breaks the chain — decaying catastrophizing often precedes the reversal.
- peak catastrophizing followed by silence
- users expressing burnout about a topic
- former activists shifting to resignation or apathy
Limitations & Uncertainty
CassandraBot is a probabilistic tool, not a crystal ball. It can detect patterns that historically preceded major events, but correlation is not causation. False positives occur — a signal may fade without ever resolving into a headline event. The system is designed to surface narratives early, which means it will sometimes be wrong. That is the trade-off of early detection: higher recall, lower precision. Every forecast includes a confidence note and a list of what would confirm or invalidate the prediction.