Glossary & FAQ
Keep the keywords in one dictionary.
Terms follow engineering-doc and paper decisions; result mentions point to “pending public-data runs”. Threshold figures are default design targets, not site measurements.
§ 1
Glossary
Credibility & use
- COU
- Context of Use: scope, prediction targets, out-of-scope, assumptions, risks and validation requirements.
- Measurement model
- y_obs = g(x_sim) + ε; align to experimental columns with units, transforms, LOD and noise.
- Calibration
- Fit calibratable parameters on training conditions only (≤10/scenario); validation never enters the fit.
- UQ
- Uncertainty Quantification reporting coverage and sharpness (or WIS/CRPS) together.
- coverage
- Fraction of points covered by predictive intervals; train vs validation separately.
- sharpness
- Interval informativeness (e.g. relative width); blocks coverage gaming via huge intervals.
- L1 / L2
- Weak vs strong extrapolation; strict isolation.
- Verification-lite
- dt/grid sensitivity engineering floor; not full numerical verification.
Platform & replay
- ISSP
- Immune System Simulation Platform; public brand OUTLIVE.
- Scenario Pack
- Minimal reviewable unit: config, tasks, public data and COU.
- ABM
- Agent-based layer for discrete interactions.
- Parquet replay
- Shards + JSON index with provenance; reports cite time segments.
- replay_ref
- time_range + filters structure linking reports to replay.
- digitization_error_band
- Error band for digitized paper figures; must enter the audit trail.
Endpoints & immunology
- T_clear
- Clearance time when pathogen observables fall below detection threshold.
- AUC
- Area under curve for load or inflammation proxies.
- TGI
- Tumor Growth Inhibition, often TGI@dayX.
- IFN / NK / CTL
- Type-I interferon, natural killer, cytotoxic T (effector CD8) roles.
- Treg / MDSC / PD-1
- Suppression and checkpoint concepts for tumor L2 / interventions.
- DMTA
- Design–Make–Test–Analyze; v3 planned interfaces; no v1 hardware.
- Intervention strategy
- AI search output as intervention_script.yaml — not a molecular structure.
§ 2
Abbreviation sheet
| Abbr | Meaning | Default hint |
|---|---|---|
| COU | Context of Use | Required per scenario |
| UQ | Uncertainty Quantification | coverage + sharpness |
| V-lite | Verification-lite | Endpoint Δ default ≤20% |
| L1 / L2 | Extrapolation strength | Strict isolation |
| ABC / MCMC / LHS | UQ/sampling methods | v1 at least tier A |
| T_clear / AUC / TGI | Endpoints | By scenario |
| ABM | Agent-based modeling | One core layer |
| DMTA | Experiment loop | v3 planned |
Hints are document design targets, not measurements.
§ 3
FAQ
Is this a digital twin for individual patients?+
No. v1 does not promise individual-patient prediction or clinical decisions. COU covers research demos, teaching exhibits and early mechanism studies.
Can you give dosing or treatment advice?+
No. Strategy search yields mechanism-comparison scripts on calibrated+UQ models — not medical advice.
Why murine data before human data?+
Richer controllable series, lower heterogeneity, easier L2, more auditable observables. Human extrapolation is a v2+ discussion.
Do you have fit results or coverage numbers now?+
No measured results to show. Everything result-class awaits public-data runs; no fabricated DOIs.
What is L2 vs L1?+
L1: weak (inoculum/dose/timing). L2: strong mechanism changes (NK recruitment drop, IFN block, neutrophil defects, Treg/MDSC boost, …). Validation never calibrates.
Why not just widen intervals to raise coverage?+
UQ must report sharpness. Ultra-wide intervals that only inflate coverage are considered uninformative.
Is Verification-lite full numerical verification?+
No. It is a dt/grid sensitivity floor (default ≤20% endpoint change, COU-adjustable).
Where can replay open?+
Design goal: Web, Unity, Unreal on one protocol. Web player is planned on this site; Unity first, Unreal staged.
Popular vs research mode?+
Shared modules, graded outputs: popular for phenomena/story; research for pinned ranges, public data, intervals and reproducible configs.
Do you generate molecules?+
Not in v1. Only intervention strategy search; molecular/DMTA later.
Are default thresholds fixed?+
They are design defaults adjustable in COU — not measured pass/fail shown here.
Where was the paper submitted? Is the spine fixed?+
Status: in preparation. Default spine: murine Influenza-like + MC38-like + blockade/exhaustion L2 — replaceable.
Is digitizing paper figures reliable?+
Usable if scripts, hashes and digitization_error_band are retained, with cross-checks when needed.
How to collaborate?+
Email contact@outlive.top (TBD). Modes: scenario packs, data alignment, report/replay, exhibit narrative. See About.