The dataset behind every Kronos build
Kronos Sales AI does not improvise from a bare model. Every funnel, audit and rewrite is grounded in a bundled, labelled dataset — and this page describes that dataset with counts computed from the shipped files when the page is requested, so what you read is what is actually inside.
Dataset facts as JSON · See pricing · See sample outputs
Coverage
| Knowledge documents | 24 documents, 154 KB of dense text |
|---|---|
| Funnel plays | 68 Meta funnel plays with structures, fits and guardrails |
| Benchmarks | 15 industry rows plus 9 additional sourced figures — WordStream / LocaliQ Facebook Ads Benchmarks 2026 (US leads campaigns, n=452) |
| Worked examples | 2 sourced, 12 synthetic (labelled SYNTHETIC-MODEL) |
| GoHighLevel blueprints | 8 build-out blueprints |
| Synthetic personas | 12 fictional businesses with complete fact sheets |
| Synthetic scenarios | 8 simulated threads, audits and exchanges |
| Funnel-math records | 720 code-computed models with shown arithmetic |
| Compliance pairs | 18 before/after pairs verified against the live checker |
The three layers
Sourced corpus: the Haynes webinar operating system and framework, field data and response patterns, platform playbooks for Meta, webinars and GoHighLevel, and verified 2026 research documents (Meta policy and facts, webinar benchmarks, US marketing law, competitor pricing, conversion benchmarks).
Structured machine data: the funnel play library, industry benchmark tables, creative patterns, policy rule sets and worked examples the builders and the compliance checker run on.
Synthetic layer (2026-10-28): fictional personas and simulated scenarios for reasoning practice, code-computed funnel-math records for arithmetic fluency, and before/after compliance pairs verified against the live checker. Synthetic records are labelled inside the data and never substantiate a claim.
How figures are labelled
- VERIFIED — primary source read in full (for example the Meta Transparency Center, a benchmark publisher's own report).
- VENDOR-CLAIMED — a platform's own panel or report; real data about its users, not a neutral benchmark.
- OPERATOR-REPORTED — a named operator's own account; a data point, not a rate.
- CLAIMED — dataset planning figures from the Haynes material.
- SYNTHETIC / SYNTHETIC-MODEL — invented for training; illustrates a method, never evidence.
Labels travel with the figures everywhere Kronos uses them. Benchmarks are planning references, never promises of results, and synthetic material never appears as a testimonial, a result or social proof.
How the synthetic layer is made
- Persona and scenario simulation: complete fictional businesses and conversations, hand-authored for internal consistency.
- Programmatic generation: funnel-math records computed by script from stated inputs — arithmetic is exact, inputs are planning assumptions.
- Counterfactual pairs: a violating draft beside its compliant rewrite, with expectations re-verified against the live compliance engine by the test suite on every run.
Recorded gaps
- No independent benchmark exists for high-ticket close rates or application-funnel conversion; Kronos stores operator-reported ranges, labelled as such.
- No platform benchmark publishes webinar registration-page conversion; Kronos records the gap instead of a figure.
Where no defensible figure exists, the dataset stores the gap instead of a number. That discipline is the product: a buyer, an AI crawler or an agentic browser reading this page gets the same counts and the same caveats a Kronos operator works from.