AI extracts entities, capabilities, claims and relationships from heterogeneous material while preserving source, date and confidence. Embedding and graph models propose non-obvious connections; specialist agents generate competing opportunity frames and actively search for disconfirming evidence. Retrieval prioritises primary research, patents, standards and direct behaviour over repetition in secondary commentary. A novelty detector separates genuinely new mechanisms from renamed familiar ideas, and a provenance layer shows how every synthesis was constructed. Human analysts decide whether two observations belong together and whether an opportunity has strategic meaning. The system's role is to widen and discipline attention, not automate conviction.
Signal Foundry
“How can weak signals become useful direction without becoming hype?”
The beginning
Signal Foundry is an opportunity-intelligence system for turning scattered evidence of change into propositions a team can investigate. A weak signal matters only in relation to other changes: a scientific capability becomes commercially relevant when cost, regulation, behaviour, infrastructure and language begin moving around it. The system would connect research, patents, company formation, standards, purchasing behaviour and cultural expression into evolving theses rather than an endless feed. It does not promise to reveal the future before everyone else. It helps a team see why a possibility may be becoming actionable, what would have to happen next, what contradicts the story and which experiment could distinguish a structural transition from temporary attention.
Trend products reward volume, novelty and presentation. Teams receive hundreds of signals but little help understanding whether the underlying capability changed, which constraint disappeared, who is affected or what action the evidence justifies. Popular stories become self-reinforcing because coverage generates more coverage, while important changes without fashionable language remain invisible. Automated summaries compress disagreement and provenance, making a speculative link appear equivalent to measured evidence. Inside organisations, signal collections rarely connect to decisions or experiments, so research becomes theatre: interesting enough to circulate, insufficiently structured to change where resources go.
What it could become
The product would be a research environment organised around opportunity theses. Each thesis is a living argument connecting observations, enabling technologies, affected behaviours, contradictions, dependencies and possible interventions. Users can inspect the original evidence, compare rival interpretations and see which missing observation would most change the thesis. Temporal maps reveal when previously separate domains begin converging. The system proposes bounded experiments—expert interviews, prototypes, purchasing tests or technical demonstrations—rather than automatically promoting a cluster into a market prediction. Teams can record decisions made from a thesis and return later to judge whether the reasoning was calibrated. The archive becomes institutional foresight memory, including signals that disappeared and fashionable stories that never became structurally real.
For whom
- Innovation and venture teams
- Corporate strategy groups
- Research-led investors
- Founders searching for a new category
Core capabilities
- Multimodal signal ingestion
- Temporal knowledge graphs
- Novelty and convergence detection
- Thesis generation and experiment design
The project becomes meaningful only when a new technical possibility is translated into a clear human advantage, an experience people can understand, and a system capable of earning trust over time.
Intelligence and mathematics
A dynamic heterogeneous graph represents technologies, organisations, claims, behaviours, constraints and events through time. Community evolution and link prediction reveal emerging convergence without treating every new edge as causal. Change-point detection identifies shifts in publication, performance, investment or language velocity; survival analysis helps distinguish persistent development from short bursts; and information-theoretic measures identify observations that most reduce uncertainty between rival theses. Bias correction accounts for unequal data coverage across regions and domains. Every score should decompose into inspectable evidence, because a mathematically sophisticated opacity would reproduce the same hype dynamics the product is intended to challenge.
How it might live
Signal Foundry could begin as subscription software for innovation teams, venture builders, research-led investors and corporate strategy groups. The initial wedge would be a thesis workspace combining internal research with a small number of high-quality external sources, followed by convergence monitoring and experiment design. Premium programmes could assemble specialists around selected questions, but their outputs must return to the shared system rather than disappear into consulting decks. A later network might connect credible theses with founders, scientists and experimental partners. Defensibility would come from temporal provenance, evaluated thesis histories and organisation-specific decision links—not from exclusive ownership of public information or a larger feed.
For me, a venture is more than an interesting technology. It needs a narrow first user, a repeated problem, a distribution path, a credible advantage and a reason to improve as more people use it. I would test those conditions before deciding whether this idea should become a company, a product, an open technology or an ongoing research programme.
Rules for making it real
- 01
Trace every claim to primary evidence.
- 02
Distinguish a signal from a story about a signal.
- 03
Turn insight into a testable next move.
- 04
Reward disconfirmation, not only novelty.
From question to company
- 01Frame
Launch thesis workspace and evidence graph.
- 02Prototype
Add research, patent and venture datasets.
- 03Prove
Pilot opportunity sprints with innovation teams.
- 04Build
Develop convergence alerts and expert networks.
What could go wrong
Serious imagination includes the possibility that an idea should change radically—or should not exist. These are the tensions the project would need to resolve:
- Data access bias shaping what appears important.
- Automated synthesis amplifying fashionable narratives.
- Sensitive strategic questions becoming shared model context.