Discovery
Preamble
Discovery is the Factory’s evidence gate. IdeaDB should find demand-backed wedges, but its first job is stopping rows, clusters, and fluent summaries from becoming founder confidence before the evidence can carry the weight.
Rows Can Look Like Insight
Discovery is where the Factory is most tempted to count motion. A scraper can run. A table can fill. A classifier can label thousands of records. A model can turn those records into a persuasive founder packet. None of that proves demand, pain, willingness to pay, incumbent weakness, reachability, or buildability.
Evidence And Inference Stay Apart
IdeaDB has to preserve the chain from raw artifact to founder claim. Raw evidence is a source quote, review, listing, pricing text, app metadata, trend value, page snapshot, or API record. A claim is a typed interpretation derived from that evidence. A cluster is a repeated pattern across claims. An opportunity is a wedge inferred from triangulated evidence. A recommendation is the next action.
Those objects cannot blur. App reviews can show pain or incumbent weakness under the right gates. They cannot prove market-level disconfirmation. Pricing pages can show seller expectation. They cannot stand in for user willingness to pay. Category breadth can orient the map. It cannot prove demand.
No Shit Before Scale
The current milestone is deliberately anti-volume. No source is migration-approved yet. Pilot candidates first require source contracts, input gates, extraction preservation, demand grading, review/apply simulation, founder-brief substrate checks, and judge/human calibration pass.
That order matters because automation compounds whatever enters the substrate. Bad data becomes bad packets. Bad packets become bad verdicts. Bad verdicts become products that should never have been built.
The Founder Packet Has To Survive Handoff
Discovery lands when Validation can pass, loop, or kill a packet without redoing basic research. A founder-ready packet should explain the wedge, the user, the workflow, the pain, the incumbent, the substitute, the demand proof, the missing proof, the disconfirmation state, and the next validation action. Missing evidence should stay visible. Unsupported claims should die before prose makes them comfortable.
IdeaDB exists because the Factory cannot scale from bad evidence. Its first job is to stop the machine from manufacturing founder confidence out of rows, packets, and fluent summaries.