Non-Markovian Trajectory Modeling in Multi-Agent Biological Information Systems
Year: 2026
Type: article
Abstract: Correction notice Correction, October 2026 (v8). Version 7 reported four headline results, all on the synthetic cohort synthetic-nsclc-10k: a 22-day lead over clinical biomarkers a 41% compute reduction < 45 ms latency (62.5% faster) 100% uptime under a 400% surge It also reported an 84% reduction in clinical click friction. These results are withdrawn. A re-implementation and evaluation found the following: The original synthetic cohort's columns are statistically independent. It contains no time series, so the reported lead time could not be derived from it. The performance figures came from an initialization script whose outputs were fixed values, not measurements. The click-friction figure was modeled, not measured. The engine does not perform sequence screening, so the earlier statements about biosecurity sequence screening (OSTP/HHS/NTI) are also withdrawn. Version 8 replaces those claims with the results below. Version 7 remains accessible under its own DOI for the record. What version 8 reports The BSC Engine is a Gaussian-process change detector for serial circulating tumor DNA (ctDNA). It gives a calibrated probability that a patient's tumor signal has started rising again during immunotherapy. It runs on CPU, in Python. The developer self-assessment covers two runs of 10,000 simulated NSCLC patients from a longitudinal mechanistic simulator. Draws were every 14 days and scans every 56 days, and pass criteria were frozen before the runs. Measure Result Median lead over RECIST imaging progression 42 days (95% CI 42–56) Progressors flagged ≥ 22 days before imaging ~60% False alarms in non-progressors 3.5–4.5% Calibration (expected calibration error) 0.002–0.003 Latency, p99 per event < 0.4 ms Compute saved by gating, alerts unchanged 40–42% Limitations. All results are on simulated patients, at roughly technology readiness level (TRL) 3–4. The lead-time result depends on an assay noise profile (proportional CV curve, 16 tracked variants, 4% draw-level variation) that has not yet been confirmed by vendor validation data. With 10% draw-level variation, the median lead falls to 0 days. The engine has not been tested on real patients, has not been independently audited, and is not a medical device. Ownership and funding. The BSC Engine was developed by David Mincey and is owned by Scalar Logic Group, LLC, by assignment from its developer. Scalar Logic Group intends to sell the engine, so the author has a financial interest in how it is perceived. Funding: this work received no external funding. Code and data. The current engine source code, simulator and validation harness are not publicly released. Material published earlier stays under the license it was published with. Inquiries: [email protected].
Author: Mincey, David
Language: English
Subfield: Oncology
Field: Medicine
Domain: Health Sciences
Open Access status: green
OpenAlex ID: W7172393659
FWCI:
Cites:
Cited by:
Related to: