Health economic models live in Excel, and they get complicated fast: many tabs, many formulas, thousands of cells. Following a single parameter from where it is defined to where it is actually used is where a reviewer loses the thread. Model QCR takes the workbook as it is.
You upload the workbook and give it a few pointers: the names of the treatment arms, and the cells where total cost and total QALYs land for each one.
You also say which sheets to leave out. Simulation output sheets hold thousands of iterations rather than model logic, so the tool pre-selects them by name and you confirm the rest.
Model QCR reads every formula, traces each one back to the input cells it depends on, and builds a map of how the model is constructed.
The map is a diagram drawn from what the formulas do.
Before it judges anything, it proves the map. The configured results and every cell they depend on are reproduced from the workbook's own formulas and matched against the workbook's own saved numbers.
When it cannot prove that rebuild, it issues no report and tells you why.
The structural checks are fixed and deterministic, and each one reads as a plain English sentence.
A standard battery of behavioural tests runs on every model. Which of them can run is decided by the model's own structure, and nothing about their selection, execution or result involves a language model.
Tests written for the model in particular follow. In the film, one equalises the treatment effects between the two arms and predicts that their QALYs should converge, and they converge exactly.
In the film, nothing failed: 17 structural checks, 30 behavioural tests and 4 outcome comparisons pass.
Four things are still flagged for review. Two health states are declared in both arms and never entered at any cycle, so they carry no cost and no QALYs in either arm.
The report says what to check and why it matters: either those pathways are unreachable as built, or the simplification is deliberate and should be documented before anyone relies on those outputs.
All of it travels in one report you can hand on. The report also describes the model: population, health states, treatment arms, costs, utilities, and each parameter with its cell.
You can ask about the report. Every answer is grounded in the recorded evidence and cites where it came from, and the conversation cannot change the model or re-run anything.
A review takes about five minutes, unattended. Version one does not run probabilistic or deterministic sensitivity analysis, and does not execute macros.