Livemore Bio 0.1.0 release values

Fidelity advisory (10 November 2026)

Our rule. Every part of the twin must work as the real human body does, calculated only from real biology. Nothing it calculates counts as a result until living human tissue or real people confirm it.

Most of 0.1.0 does not meet that rule yet. We checked the parts of 0.1.0 that calculate biology against it, and we are withdrawing nearly all of them.

What this means for you now.

• 0.1.0 still shows these values. Please do not use them.
• Treat any numbers you have already exported, shared or published from them as withdrawn.
• In 0.1.1, the software will show a reason instead of a number wherever a value is withdrawn.

Withdrawn: every value 0.1.0 calculates about biology, except the four items below.

This covers, for example:

• Glucose and insulin: the twin's glucose and insulin, meals, foods and lifestyle, and the glucose tolerance test and its "effect" verdicts.
• Heart values shown by the twin: every heart-cell value the 0.1.0 twin and console show (APD90, voltage and calcium), the hERG-block results, and the "torsades risk" warning. They are paced at the twin's rule-of-thumb heart rate and include the glucose link and the block setting.
• Drugs: drug and metformin effects.
• Disease and genetics: disease progression and staging, the diabetes-prevention benchmark, and genotype and stress or sleep effects.
• Heart rhythm: heart rate and heart-rate variability.
• The virtual LIFE Patch. 0.1.0 has no physical LIFE Patch, so every LIFE Patch reading in 0.1.0 is synthetic, calculated from the withdrawn glucose engine.
• Generated people: virtual and synthetic people and their cohorts, including cohorts of "Reference-grounded research persons" (for example the "NHANES older diagnosed diabetes" and "DPP high metabolic risk" strata).
• The exogenous-insulin program in Cendos ("paired within-individual arms"). It runs the Hovorka model on a sampled virtual type 1 population, so its glucose responses and its counts of time below 3.9 mmol/L are withdrawn.
• Other calculations: the botanical exposure programs, the lobeline protocol, discovery scores, and the "combined" disease score and its ranking.

Many of these are typed-in or rule-of-thumb values, not mechanisms of the body.

Kept.

1. The published O'Hara-Rudy CiPA heart cell, as a calculation labelled "prediction, untested". It is a paced cell (one beat per second, as in a laboratory preparation) with no drug-block setting. Its equations are integrated correctly. Around them, 0.1.0 has the errors listed below.
2. The published Hovorka glucose-insulin model of type 1 diabetes (Hovorka 2004), with its single published parameter set, as a calculation labelled "prediction, untested", after the fixes below. It applies only to type 1 diabetes and is not dosing guidance.
3. The Human-GEM energy calculation, labelled as the capacity of a generic human cell network, not a value for you. It is calculated correctly.
4. Each ChEMBL and Open Targets record, labelled as that source's own association, after the fixes below.

Not affected.

• Measurements you recorded yourself with a real instrument, and files you imported, remain measurements.
• Cohorts made with "Create frozen reference cohort" under "Joint public-source records" are real NHANES public survey records and remain so.
• Updating to 0.1.1 deletes nothing you stored. Stored records that hold withdrawn values will show a reason instead of the numbers.

Known calculation errors in 0.1.0, fixed or withdrawn in 0.1.1.

Visible in the app or its local API:

• Heart-cell APD90. It can be wrong when a beat lasts nearly as long as the interval between beats (strong potassium-channel block at the twin's highest heart rate), or when a beat fails to recover.
• The reported heartbeat is not yet settled. Its APD90 can differ from the settled value by up to 16 ms.
• Look-ups:
  • they can report a partly failed look-up as complete and keep it cached;
  • they count records reported only as "greater than" a value as potent (for example ">1000 nM", meaning the half-maximal effect was not reached up to that concentration);
  • the "in genome-scale model (Recon3D)" count always shows 0.
• The diabetes-prevention benchmark (local API only). Its result depends on the random seed: 27–42% across seeds, and 36% at the API's default settings, against the trial's 28.9%. Its one scaling value was tuned to that trial figure.
• Out-of-range inputs. Some are accepted instead of refused. For example, a negative meal size gives negative glucose. Some inputs make the local Livemore server stop responding while it calculates.
• The Hovorka model reports time below 3.9 mmol/L without the body's counter-regulation, and clamps liver glucose output. Both show in the withdrawn exogenous-insulin program.

Only when the engines are called directly from Python:

• The Hovorka model:
  • it can miss a short insulin infusion (a few minutes or less) that starts partway through a run;
  • it ignores a bolus given exactly at the run's end;
  • its basal "steady state" is not converged (for example 5.96 instead of 5.05 mmol/L at 0.4 U/h);
  • by default it starts from zero insulin.
• Full potassium-channel block in the heart cell gives a wrong APD90.
• A heart-cell parameter override can carry over into later runs on the same object.
• The default glucose engine can skip a meal in runs longer than a day.
• The Cendos lab-records module (LIMS) stores calculated readouts as experiment evidence. 0.1.1 will show a reason instead of readouts calculated by withdrawn engines, and will label any others as calculations, not experiments.

Nothing produced by 0.1.0 is clinical evidence. Every value the twin calculates is, at best, a prediction until living human tissue or real people confirm it.

0.1.1 fixes or withdraws each problem above and corrects every label that overstates what an output is.

The same text is published with the distribution and in the GitHub release.

Release values

Version 0.1.0 is the stable maintenance release. Before you run the installer, check that what you downloaded matches these values. The installation guide shows how.

The same values are published here and on GitHub, at the Livemore-Biosciences release. If the two disagree, or a download does not match, do not install. Contact Livemore Biosciences instead.

Email us

After installation, livemore --version prints LivemoreBio 0.1.0. This release is for research use with synthetic and public data. It is not for clinical use.