Kscope · disruption · 2026-10-07
Yes. An analysis, an opinion, or a model is a product we can build when the inputs are filings, news, RSS, and open crawl. We already parse funding, mergers, people, and pay out of that stream, including about 100,000 RSS feeds in 16 languages. The first scoring pass treated that model layer as if it were a panel. It is not. What keeps its price is an input nobody publishes.
Crunchbase, CB Insights, and Tracxn move to now. The private graph is funding and M&A parsed from every news article, RSS feed, and open crawl, with Form D as the filed slice. kbrain is behind this file. PitchBook and Dealroom stay a slice, and the wall is narrower: financials a company submits only to them, founder submissions, and partner feeds. A prediction on the graph we have is our model.
BoardEx, Equilar, and Altrata move to now on the people side. The private and non-US relationship file is the RSS crawl, in 16 languages, plus the proxy and the 8-K. Equilar's public pay was already first-party. Private pay that gets published rides the same feeds. A survey answer or an interview that never becomes a document stays theirs.
Credit opinions, ESG scores, vote recommendations, covenant scores, controversy indexes, quant scores, citators, and written industry forecasts move to now or fast wherever the input is a corpus we hold or can crawl. Moody's letter, MSCI's badge, and ISS's client policy are their brand. The view we publish is ours. Visible Alpha, Zacks, and Estimize stay never, because those numbers are other people's unpublished estimates. Training our own model produces a different product, and it does not reproduce a broker consensus.
The dataset is one entity graph. A person, a company, a fund, and a security are the same ids across Form 4, the proxy, 13F, N-PORT, ADV, Form D, the 990, the agreement, the 8-K, and the facts pulled from news, RSS, and open crawl. Every filing edge can point at an accession. That is the MARS graph, plus a private layer the notes have not caught up to. Console is a FactSet-style front end Raul is building on top of it. The MCP and the public API are the other doors. Insider Focus is the proof that this graph already wins a market: donor wealth, where the number that matters is verified capacity and a liquidity trigger.
SEDAR continues through QuoteMedia. Image-only OCR on old EDGAR exhibits is the agreements hole that is still open, and it is a short technical job.
The graph is the disruption, because a buyer currently pays several vendors to assemble one question: who owns this, who sits on the board, which fund they advise, which contract they signed, who holds the bond, who just raised, and who just had a sale under a 10b5-1 plan. An LLM can answer any one of those hops from a web search. It cannot compute the join, keep it point-in-time, and cite the source. A model on top of that join — a credit view, a vote recommendation, a seller-intent flag, a covenant score — is the same shape of product. The scarce piece is the resolved network, and the model is how we charge for a computed outcome across it.
Extraction speed is the advantage, so it should be spent where the output joins a graph we already have. A new free corpus that does not attach to a person or a company we already resolve is a science project. A free corpus that does is a product.
Hebbia, Rogo, Fiscal.ai, Hudson Labs, Intelligize, and BamSEC are scored now. Their buyers are already asking an agent to read filings. We are the data under that question, with the join those tools do not have. AlphaSense is a slice: we take the filing search, and we leave the Tegus library. The work is packaging, not extraction. Point-in-time files, a data dictionary, and the MCP family keys Console already uses. Console can keep growing as the human front end. It does not have to be finished for this to sell.
DonorSearch, iWave, and WealthEngine are scored now. We already lead on the public-company prospect. Windfall, GOBEL's clinical file, Blackbaud's house file, and EverTrue's engagement graph stay slice, because the household file, the EMR, and the school's own gifts are inputs we do not hold. The fast adds that make the same person more valuable are the ones that join: finish the county property records we have started, and take FEC, LDA, and congressional trades. Those are free, and they hang off a person id we already mint. Do not buy a consumer file to look more like WealthEngine.
Discretionary versus 10b5-1 insider sales, Form D dated to the EDGAR acceptance, and the filing classifiers are already extracted. The go-to-market note says funds will ask for point-in-time history, identifiers, a dictionary, and one backtest each. That is the bottleneck. None of it is a new crawl. This is how the graph gets paid while Console is still a front end, and it is the honest alternative-data product: signals a terminal column does not have, because the terminal did not parse the footnote.
pfactorial is now, with SEDAR continuing and OCR as the leftover. Covenant Review is now: the indenture is in the agreement graph, and the lawyer's score is a model on that text. CreditSights and the rating agencies are fast, because a credit view on filings and agreements is the same kind of model. Octus, 9fin, and Debtwire stay slices: we publish the view, and their newsroom, their loan marks, and their unpublished private-credit reports stay with them. S&P LCD stays never. A credit buyer who wants the document, the capital structure, and our view, joined to the holder and the insider, is a sale. A credit buyer who wants a bid is a different sale.
WhaleWisdom, Fintel, and GuruFocus are now. 13F, Form 4, 13D/G, and N-PORT are in the book. N-PX is free XML. ISS and Glass Lewis are now too: the ballot is on the proxy, and a recommendation is a model on that text plus the votes. Their advantage with institutions is that clients have adopted their policy. The model itself is ours to ship, under our name.
BoardEx, Equilar, and Altrata are now. Public pay is first-party proxy extraction. Private and non-US people are already arriving through about 100,000 RSS feeds in 16 languages. The product is the path: this director sits here, filed a Form 4 there, advises this fund, showed up in a foreign-language funding story, and chairs this foundation. A relationship score on those edges is a model. A survey response or an interview that never becomes a document stays theirs. ZoomInfo and Proxycurl stay never, because a direct dial and an email are a contributed contact file. Buying one to look like them does not extend this graph.
Quiver is fast because contracts, lobbying, and congressional trades are bulk files, and Form 4 is already done. That bundle attaches to people and companies we resolve. ImportGenius is also fast, and PatSnap's document layer is fast, but bills of lading and patents join a company more weakly than a lobbying filing joins a person. If we run one new crawl, run the government-records bundle. Job-posting panels (LinkUp, Thinknum, Revelio) are fast too, and they are a second bet, not the first.
2iQ is fast. The US file is done. Canada comes with the QuoteMedia SEDAR feed. The UK and the other regimes that publish insider reports are the same shape of work, one parser at a time. That is a product international desks pay 2iQ for, and it lands in the person graph instead of in a new silo.
48 of the 153 companies are never, and 39 are slice because a contributed or licensed input sits beside a core we can take. The rest, 66, are companies whose core we have, can crawl, or can model from that corpus. The never list is the discipline, and it is a list of inputs. Real-time and evaluated prices. Broker working models, broker consensus, and crowd estimates. Expert-call libraries. Card, location, device, and satellite panels. Commercial credit files. CoStar and RCA. Burgiss and Preqin cashflows. Premium squawks and wires. Street-name proxy distribution. Direct-dial and email graphs. Questionnaire answers and a nonprofit's own gift file. Mergermarket's pre-announcement story.
A model we train on filings and the crawl belongs on the now and fast side of that line. Quoting it against Visible Alpha, or against a card panel, describes a different product. The Excel add-in that already runs FactSet formulas is still the right wedge for the filing columns: sit beside the terminal and replace the modules whose inputs we hold. Estimates that are a resale of the street stay in the terminal. Estimates that are our model can sit in the add-in under our name.
The fraction-of-the-price claim holds for the now and fast rows, including the models, and for the crawled half of every slice. It holds there because the cost of a model collapsed and the corpus is already in house. It does not hold for the never rows. The buyers who feel the difference first are donor shops, fund analysts, bank coverage teams, credit desks who want a document and a view, governance teams who want a ballot and a call, and anyone wiring an agent to the graph. They can drop a filings vendor, a people file, or a research shop without dropping Bloomberg.