Model Benchmarks
Introducing DiscoLike Supernova: A Purpose Built ICP Fit Model
Supernova builds on Stardust, our existing company discovery model, with a focus on finding companies that meet your ideal customer profile (ICP).
From Stardust to Supernova
Stardust, DiscoLike’s first company discovery model, uses information from business websites to find companies similar to a target description or an existing customer. It established the foundation for our company search and lookalike discovery.
Supernova builds on that foundation as a purpose built ICP fit model. It focuses discovery on whether a company meets the requirements in an ideal customer profile, including criteria that similarity alone may miss.
For example, a seller targeting manufacturers with their own production facilities needs to distinguish those businesses from equipment distributors and agencies serving factories. All three may discuss the same products and industries, but only the manufacturer meets that requirement. The benchmark below measures how often the retrieved companies fit the target profile.
Stardust vs. Supernova: benchmark results
We tested Supernova across millions of ICP prompts and fit validations. In DiscoLike’s reported October 2026 internal benchmark, 94.1% of Supernova’s top 100 results were judged an ICP fit, compared with 78.1% for Stardust, our previous company discovery model. That is a gain of 15.9 percentage points and a 20.4% relative improvement. Fit was assessed by an automated language model judge.
ICP Fit
Share of results judged an ICP fit.
More of the first companies you review match your requirements. At 1,000 results, Supernova still returned 80.8% fits versus 63.7% for Stardust.
Lookalike
Share of results judged similar to the starting business across a random and diverse company set.
Supernova returned 97.3% similar companies in the top 100, compared with 77.5% for Stardust, when searching the full company index.
Effective Fit
Average assessed fit strength on a 0-100 scale.
Shows how strongly results match your profile, including partial matches that a fit count alone does not describe.
Score separation
How well a result’s similarity score separates fits from non-fits (AUC). 0.50 is chance, 1.00 is perfect. A random non-fit outranks a random fit 79% less often with Supernova.
Supernova scores more reliably distinguish matching companies from non-matches, helping you prioritize which accounts to review.
View data as tables
| Result depth | Stardust | Supernova | Relative gain |
|---|---|---|---|
| Top 10 | 84.6% | 98.2% | +16.1% |
| Top 100 | 78.1% | 94.1% | +20.4% |
| Top 1,000 | 63.7% | 80.8% | +26.9% |
| Result depth | Stardust | Supernova | Relative gain |
|---|---|---|---|
| Top 10 | 84.6% | 98.6% | +16.5% |
| Top 100 | 77.5% | 97.3% | +25.5% |
| Top 1,000 | 64.9% | 94.2% | +45.1% |
| Top 10,000 | 47.9% | 87.0% | +81.6% |
| Result depth | Stardust | Supernova | Relative gain |
|---|---|---|---|
| Top 10 | 66.4 | 83.4 | +25.5% |
| Top 100 | 60.0 | 76.5 | +27.5% |
| Top 1,000 | 48.3 | 61.7 | +27.7% |
| Result depth | Stardust | Supernova | Relative gain |
|---|---|---|---|
| All ICP Fit searches | 0.70 | 0.94 | 79% fewer misranked |
Relative changes use the reported scores before display rounding; results vary by search.
Bringing ICP fit into discovery
Checking results for fit can remove mismatches, but it cannot recover qualifying companies that search never returned. That is why ICP fit needs to be part of discovery.
Supernova advances that direction by improving fit at discovery. For sales teams and prospecting agents, the measured benefit is a higher share of matching companies in the results they review. The benchmarks compare search quality with Stardust; they do not establish complete coverage of every qualifying company in a market.
Why DiscoLike is in a unique position to build this
DiscoLike’s proprietary corpus combines customer ICP searches, company fit validations, and simulated evaluations using VibeCheck, our specialized fit validation model. Together, these connect the criteria users specify with evidence about which companies meet them. This combination puts us in a unique position to build discovery models around ICP fit.
Supernova builds on Stardust’s company discovery foundation and this proprietary corpus. Public company information alone does not provide the same history of ICP requirements and qualification outcomes.
Evaluation methodology
All evaluations ran in October 2026 and compare Supernova with Stardust, DiscoLike’s previous company discovery model.
The reported benchmarks are specific comparisons from that broader evaluation work. An automated language model judge assessed company fit against each target profile; these scores are automated assessments rather than human qualification decisions.
ICP Fit measures the share of results that match the target profile. Effective Fit measures average fit strength, including partial matches. Both cover the top 10, 100, and 1,000 results. Score separation measures how reliably similarity scores distinguish fits from non-fits in the ICP Fit evaluation.
The top 100 ICP Fit rate increased from 78.1% for Stardust to 94.1% for Supernova: a gain of 15.9 percentage points, or 20.4% relative to Stardust. Changes are calculated from the unrounded scores. This measures the share of returned companies judged a fit, not the share of the total market found.
For Lookalike, we picked 100 random active companies as starting points and retrieved the 10,000 most similar companies for each with Stardust and Supernova. Retrieval covered active companies available to both models across the full company index. An automated fit judge assessed whether each result was similar to the starting business.
Supernova scored higher in the top 100 for 84 of the 100 starting companies, tied for 14, and scored lower for 2. In the top 10,000 results, it scored higher for every starting company.
These internal evaluations measure search quality. Qualification for your campaign still matters; time and token savings were not measured.
Start with your target market
Describe the businesses you want to reach in the DiscoLike ICP generator. Include what they do, where they operate, and what would disqualify them. The generator turns those requirements into a structured profile and a prompt you can use for company search.
For agent workflows, see the MCP setup guide or company search API.
Related reading:
Your market, your criteria
Try an ICP for your market.
Adapt the manufacturing example below, or describe your own target market. The generator will turn your requirements into a profile and search prompt.
“Try this ICP” opens the generator with the example filled in. Review it before generating your profile.
Connect your agent through MCP · Explore the company search API