# Original contributions

This file separates code written in this repository from published equations, external software, and ideas that have not been shown to improve the model. It does not claim that a method is new to the literature. No literature review broad enough to support that claim was done.

## Independently developed equations or algorithms

The valve-event restart in `src/hdt/physiology/independent/algorithms.py` is a standard half-step refinement at a detected discontinuity. The detection rule and the depth limit were written here. The comparison rejected it as the default because 1 ms RK4 was already within 0.005 percent of a 0.25 ms reference on ejected volume. It is not described as an original numerical method.

The mechanism account of the 3.76 L/min output and the 371 mL tidal volume is an analysis of the existing `independent-v1` equations. It did not add an equation.

## Original implementations of published equations

These are implementations written in this repository. The relationships come from the cited papers.

- Time-varying elastance, `P = E(t) (V - V0)` with `V0 = 0`, from Suga and Sagawa, Circ Res 1974. The half-cosine activation and the 35 percent and 15 percent active fractions are modeling choices stated in `docs/independent-engine.md`.
- Diode valves and resistive compartment exchange. The open-valve resistance, sized as 5 mmHg at 500 mL/s, is an assumption.
- Organ flow fractions from Jones and Rowland-Yeo, CPT Pharmacometrics Syst Pharmacol 2013, used to size resistances at a design cardiac output of 5 L/min (Guyton). The 5 L/min value sizes `R`. It is not an imposed output.
- Baroreflex slope 8.1 ms/mmHg from the AltitudeOmics sea-level sequence method, with a 3 s lag that is an assumption.
- Ventilatory slope 2.60 L/min/mmHg and the associated intercept treatment from Irsigler, Am Rev Respir Dis 1976. The 15 s chemoreflex lag is an assumption.
- Hill oxygen content with `n = 2.7`, `P50 = 26.8 mmHg`, Hüfner 1.34 mL/g, hemoglobin 150 g/L, and a linear carbon-dioxide content. The carbon-dioxide line is a standard linearization, not a new dissociation curve.
- Klotz et al., Am J Physiol Heart Circ Physiol 2006, normal-human coefficients `P = 1.6e-11 mmHg * V^6`. Implemented as an optional left-ventricular diastolic law. Not the default. The comparison gain was about 4 percent of cardiac output.
- Ursino, Am J Physiol Heart Circ Physiol 2000, local oxygen conductance with gain 30 and time constant 10 s. Implemented as an optional muscle-resistance update driven by mixed-venous content. Not the default. The comparison gain was about 0.27 L/min.

## Modified established numerical methods

Fixed-step classical RK4 is the default integrator. The stored flow samples use the RK4 stage weights so that the recorded interval-mean flow matches the volume update. That is a correction of the sample, not a new stepper. The valve-event option is the half-step refinement above. It was measured and was not selected.

## Integrated external software

Pulse 4.4.0 remains an optional reference build. The independent engine does not import it. The Prompt 1B report sets `pulse_invoked` false. Jones and Rowland-Yeo tissue volumes remain the drug-model table. No new external solver was linked for these comparisons.

## External scientific datasets

No new human dataset was fitted. The numbers cited above are published parameters or published orders of magnitude. The 4–8 L/min cardiac-output interval, the 400–600 mL tidal-volume interval, and a cardiac-output slope of about 5–6 L of blood per L of oxygen are comparison ranges. They were not used as targets.

## Experimental ideas not yet demonstrated

- A central-command or muscle-pump exercise model with gains taken from Magosso and Ursino, Med Biol Eng Comput 2002. The numeric gains were not in the source used for this work, so they were not coded.
- A muscle-venous oxygen concentration, as distinct from the mixed-venous pool used by the Ursino trial.
- A diastolic curve or stressed-volume assignment identified for this circuit. The Klotz trial did not justify replacing the linear law.
- A recovery long enough for the 1 L carbon-dioxide store to finish returning to the 36 s rest. At 120 s the return had started and had not finished.
- Replacing 1 ms RK4. The refinement study did not justify it.

`independent-v1` is the regression configuration. Later equation changes have to be a named version rather than an edit of these defaults.

## organ-v1

Added after the Prompt 1B checkpoint. This is a named layer, not an edit of the cardiovascular defaults.

- Original implementations of published anchors: Guyton GFR 125 mL/min and a constant filtration fraction implied by the Jones kidney flow at 5 L/min; Wilkinson and Shand well-stirred clearance through the existing PBPK law; Ursino 2000 muscle gain 30 and time constant 10 s applied, as a rejected trial, to muscle venous content rather than mixed-venous content.
- Modified established numerical methods: donor-limited flux for urine; the closed form of a linear two-compartment transit; RK4 at 60 s for the glucose state; matrix exponential retained for drug.
- Not implemented, and not claimed: central command, renal autoregulation, ADH, a glycogen store, and a validated glucose or exercise response.
- Tested and not adopted: step-doubling control of the organ exchange interval, and the same controller on the glucose state. Both are standard local error tests. The exchange controller matched a fine fixed step and cost more simulated time. The glucose controller was less efficient than RK4 at 60 s.

## wholebody-v1

A named coupling layer. It does not edit independent-v1 defaults.

- Original implementation of a shared clock: interval-mean organ flows from the 1 ms heart drive the existing renal, glucose, transit, and well-stirred PBPK steps. The muscle sample is the existing finite-volume solver with its own tracer.
- Parametric ellipsoids sized to the Jones volumes. Positions are construction choices. BodyParts3D is not an input.
- Established methods, compared and not renamed: a 1 s frozen exchange, predictor-corrector averaging, step doubling, polar meshing, a projected cube, Murray's cube law, and a Poiseuille inversion onto the existing resistances.
- Adopted from those comparisons: the 1 s exchange while conductance is fixed, polar surfaces, and resistance-matched synthetic radii.
- Rejected or left in place: step doubling, predictor-corrector, the projected cube, Murray radii as a match to this circuit, and adaptive cell counts for the muscle sample.
- Not claimed: clinical anatomy, central command, or a cardiac output that rises with a threefold gas-exchange target.
- The glucose feedback closures and the 0.20 muscle share of resting oxygen uptake are assumptions, stated in `docs/INDEPENDENT_ORGAN_PHYSIOLOGY.md`.

## Portfolio classification

The portfolio release uses Pulse 4.4.0 as the physiology runtime when a scenario requests it, and BodyParts3D as the default view. The items below are the local engineering record. Standard RK4, matrix exponentials, donor limiting, and finite volumes were not invented here. No entry claims literature novelty.

| Component | What was built | Established method | Local change | Evidence | Novelty |
| --- | --- | --- | --- | --- | --- |
| PBPK | Jones teaching-compound state and `advance_linear` | Well-stirred clearance, matrix exponential | One owner for drug, so Pulse clearance is not added again | Dose residual at roundoff on recorded runs | Implementation, not a new law |
| Tissue | Finite-volume sample with arterial inflow and venous return | Conservative finite volume | Exchange with the systemic pools | Outlet error already ~1e-9 at 4 cells; refinement not adopted | Implementation |
| Renal | Donor-limited water and sodium step | Explicit Euler plus a store limit | Refuses a negative donor | Adopted; long Euler can go negative | Standard limiter, local accounting |
| Transit | Closed-form stomach and intestine | Linear two-compartment | Exact step instead of Euler | Euler at 3600 s goes negative; closed form adopted | Standard solution, local use |
| Scheduler | Integer-microsecond lattice and held physiology | Zero-order hold | Pulse is advanced once per scenario, then held for the drug step | Documented in `docs/METHODS.md` | Integration, not continuous inner-loop Pulse |
| independent-v1 | Heart, lung, baroreflex, CO2 controller | Suga–Sagawa, Irsigler slope | Original code, parameters not retuned | Resting output ~3.8 L/min; experimental backend | Implementation |
| course-correction-1 | Plasma share into the systemic vein | Fixed volume partition | Opt-in; 200 mL removal moves output, pressure, and GFR | `docs/COURSE_CORRECTION.md` | Not the portfolio default |
| Held-flow split | 4 s heart, 16 s frozen organs | Multirate freeze | Urine error 5.1 percent | Rejected | Not adopted |
| Anatomy fixture | Jones-volume ellipsoids | Parametric surfaces | Not derived from BodyParts3D | Volume fixture only | Not the displayed body |

The longer narrative is `docs/PORTFOLIO_CASE_STUDY.md`.

## course-correction-1

A named opt-in exchange. It does not edit `independent-v1` or the wholebody-v1 default.

- The plasma share of intake minus urine is added to systemic venous volume between cardiac steps. The share is the existing Jones plasma volume divided by the existing 15 L extracellular volume. Red-cell volume is fixed for that water exchange.
- A 200 mL whole-blood removal is a scenario input. It is not a target used to tune elastance.
- The 4 s / 16 s frozen-flow split was implemented and rejected: urine error 5.1 percent on a 60 s test, despite a lower wall time.
- Not implemented: central command, a Starling interstitial refill, a retune of resting output or tidal volume, and an independent organ surface checked against measurements.
- The licensed BodyParts3D view is the workbench default again. The ellipsoids are labeled a volume fixture.