The most useful dataset we have ever published about Israeli technology is a measurement of our own ignorance. The Claw & Talon database tracks 1,437 companies — startups, public companies, acquired assets, defense primes, funds, and the wreckage of ventures that did not make it. This summer we ran the entire corpus through a hardened verification pipeline that asks one blunt question of every record: independent of anything the company says about itself, what evidence exists that the claims in this profile are true? The answer reshaped the site, and it is worth publishing in full, because nobody else who maps this ecosystem seems willing to say it out loud.
The distribution nobody publishes
Counting sources sounds trivial. It is not, once you define the rules honestly. In our pipeline, a company's own website never counts as a source. LinkedIn never counts, because it is self-authored. Duplicate links to the same URL count once. What remains — press coverage, regulatory filings, procurement records, index constituent lists, investor disclosures, conference documentation — is what we call an independent public source. Against that definition, the roughly 1,460 raw records in our data directory break down like this: 402 have zero independent sources, 234 have exactly one, 320 have exactly two, and about 508 have three or more.
Sit with the first number. More than a quarter of the companies in a professionally curated map of one of the world's most-watched technology ecosystems have no independent public evidence trail at all — no press that is not a reprinted announcement, no filing, no third-party documentation we could locate. These are, overwhelmingly, real companies with real employees. They are simply operating below the waterline of public verification, which is normal for early-stage and security-adjacent ventures and fatal for anyone who wants to write authoritatively about them.
What we did about it
The response was to rebuild the site around a three-tier policy, documented in full in our methodology. Roughly 170 profiles — those with at least three independent sources, at least 400 words of surviving editorial analysis, no verification flags on material claims, and either a high research-priority score or status as an established public company or defense prime — remain indexable flagship dossiers. Around 1,250 more remain browsable in the database as honestly labeled research leads, excluded from search engines. A final tier of unverifiable records is orphaned entirely. In practical terms: we voluntarily removed about three-quarters of our own pages from search indexes, because we could not stand behind them at the standard a search result implies.
The grading is enforced by code, not culture. The tier rules live in a single policy module consumed by the site build and by two independent audit scripts that re-derive every count from the raw data on each deployment; if an indexable page slips below any bar, the deployment fails. We also learned why automation must audit the auditors: an earlier editorial rewrite, tightening language across hundreds of profiles, silently dropped weakly-cited press references and de-indexed 31 profiles as a side effect. The pipeline caught it. A human reviewer would not have.
What the taxonomy reveals
Verification at scale also forced honesty about what these companies are. "Startup" turns out to be the wrong noun for a large minority of the ecosystem map: among records declaring an entity type, 807 are operating startups, but 223 are acquired assets — companies absorbed by acquirers, tracked because their technology and teams still matter — alongside 87 public companies, 24 defunct or wound-down ventures, 13 funds, 7 defense primes, and 16 records we classify, bluntly, as unverified. Each classification changes how a profile is written and what a reader should do with it. An acquired asset's dossier is about what survived the acquisition; a defunct company's dossier is a post-mortem; an unverified record is a warning label, not a profile.
The distribution also explains a market failure. Aggregator databases and press coverage concentrate on the same few hundred well-documented companies, because that is where the sources are — while the long tail gets described, when it is described at all, by recycling the companies' own marketing language. Our zero-source count is a measurement of exactly how large that recycled-marketing zone is. Any map of Israeli technology that does not grade its evidence is, for a quarter of its entries, a press release with a database skin.
What this means for readers
For an investor, the practical value is calibration. When a company appears in our flagship tier, the profile separates what independent evidence supports from what the company claims, source by source, with grading notes on what each citation can and cannot carry. When a company appears as a listed research lead instead, that placement is itself information: it tells you the public record is thin, that diligence must start from primary conversations rather than published material, and that confident third-party writeups you may find elsewhere are probably laundering the company's own copy. For the ecosystem, the finding cuts both ways. Israeli companies are often operationally secretive for good reasons — customers, adversaries, export rules. But the paper-trail deficit has a cost: it suppresses credible independent coverage, narrows the set of companies serious capital can research, and leaves the field to hype. Publishing our grading rules, and the unflattering distribution behind them, is our contribution to shrinking that gap. If you can improve a record — with a source we can check — the corrections process is open, and we use it.