Reproducing our numbers
Recompute the synthetic scoring example and see the evidence needed to substantiate a product measurement
On this page — 5 sections
This is a synthetic reference estate, not customer data or a measured pilot result. Its 42 assets and score of 75/100 illustrate the published method; they do not predict findings or accuracy on your systems.
Recomputation is available
You can recompute the published scoring example from the weights, cohort counts and means on /methodology. This checks that the arithmetic follows from the stated inputs. It does not validate a customer outcome or the choice of weights.
| Input or result | Value | Interpretation |
|---|---|---|
| Dimension weights | 30 / 25 / 20 / 15 / 10 | Declared weights summing to 100. |
| Worked asset | 86.5 → 87 | Illustrative weighted sum, rounded half-up. |
| Reference estate | 42 assets | Synthetic example. |
| Outside Contained | 36 assets | Includes classical weakness and quantum exposure. |
| Estate calculation | 6295 / 84 → 75 | Criticality-weighted mean, rounded half-up. |
A check on the estate tier
Changing a sufficiently influential cohort can change the estate tier. For example, changing the first cohort mean from 90 to 30 reduces the weighted total by 24 × 60. The resulting (6295 − 1440) / 84 rounds to 58, in Watch. This is an arithmetic scenario, not a measured remediation result.
What a timing comparison needs
Earlier review-time figures have been withdrawn as a substantiated product benchmark. The supporting measurement records have not been established for publication. We make no numerical speedup claim here.
- The same input set and agreed scope for manual and automated review.
- Documented staffing, allowed tools, start and stop conditions, and elapsed time versus person-hours.
- Software and detector versions, hardware, configuration and run logs.
- Repeated runs with variability and failures reported.
- An independent reference and checks that the outputs are comparable in coverage and quality.
What an accuracy claim needs
The earlier classifier figures came from a synthetic evaluation. They do not establish accuracy on customer estates. An externally useful evaluation needs a documented dataset, labels, sampling method, split protocol, leakage controls, software version and per-class error analysis.
What remains unproven by this page
- Customer discovery completeness or classifier accuracy.
- A typical scan time or improvement over manual review.
- Independent reproduction of a product benchmark.
- The availability of automated scoring or any planned platform module.
More in The Theos Method