How A Signal Becomes A Tell
Seven stages take raw public records to an inferred strategic read, and the center of that path is a disagreement: two AI readers who analyze the same evidence on their own terms. Every number below is computed from this build’s snapshot.
148
Signals analyzed12
Companies47
Strategic themes19
Source types100
Debates57
Hypotheses- Stage 01
Discovery
For every tracked company, the platform generates targeted search queries and suggests subreddits, then resolves candidate sources — newsrooms, filing repositories, blogs, job boards. Discovery is LLM-driven: as the investigative questions change, so does the source list.
12 companies trackedSee the companies - Stage 02
Collection
Scrapers pull the raw material — SEC filings, earnings transcripts, news, social posts, job postings, patents, court records, lobbying disclosures. Every fetch is cached, rate-limited, and attributed to the source that produced it.
19 source types · 148 signals in this snapshotBrowse the signals - Stage 03
Enrichment
Before analysis, each company is anchored: website probes, blog and social discovery, and SEC ticker resolution. Enrichment gives the scrapers a canonical home per company instead of a name to search for.
Website, blog, social, and ticker anchors resolved for all 12 companies - Stage 04
Understanding (NLP)
A local NLP layer turns raw pages into structured material: vector embeddings, entity extraction, sentiment classification, keyphrase extraction, language detection, and a quality gate. Models run in worker threads on the machine — this stage does not ship text to a third-party inference API.
Local inference · no external embedding API - Stage 05 · The differentiator
Dual-Agent Analysis
Two readers take on every signal independently. The Analyst weighs numbers, dates, and named sources. Gossip Girl reads subtext, executive behavior, and what the disclosure omits. Where they disagree, the system stages a structured debate: each position, its evidence, the points of contention, and a synthesis.
The AnalystNumbers, dates, named sources. Reads the disclosure for what it states.
vsGossip GirlSubtext, behavior, omission. Reads the disclosure for what it avoids.
100 debates — 53 signal · 47 clusterSee a live debate - Stage 06
Correlation
Related signals cluster into strategic themes. Each theme carries momentum, a synthesized implication, and key facts deduplicated across its member signals, so a pattern becomes visible across companies and weeks rather than one article at a time.
47 strategic themes, each with momentum and a synthesized implicationExplore the themes - Stage 07
Delivery
The result is the feed, a company dossier with an inferred strategic intent and the evidence behind it, and briefing artifacts meant to be forwarded. Everything on this site renders from a single frozen pipeline run.
This demo is a static snapshot — the pipeline above ran once, at October 15, 2026. No live data, no logins.