# Waiting Room Timing Simulator

## PURPOSE
The video's central advice is timing: wait in the "waiting room," estimate "6 to 12
months before a startup is genuinely ready" and "12 to 18 months for the labs to catch
up." This SimPy + NumPy prototype turns that advice into a discrete-event simulation so a
business can test three strategies over 24 months — BUILD NOW (a custom AIOS), WAIT THEN
GO-ALL-IN on a startup platform, or WAIT for a lab platform — and see which loses least.

## WHY THIS MATTERS
It makes the transcript's hidden cost explicit. The video warns that custom builds need
"maintenance every time there's a model update, an API change, or a new platform feature
that makes half of it redundant" and that "at least one of those things is happening
every month." A simulation that fires those disruption events is the honest way to price
"waiting isn't free" against "you'll have paid tens of thousands for a prototype."

## HOW TO BUILD
Use SimPy for the event clock and NumPy (seeded) for random model-release / API-change /
feature-ship events and for Monte Carlo draws of the adoption windows the video quotes.
Key snippet:

```python
def model_release(env, api, mean_gap_days=30):
    while True:
        yield env.timeout(max(1.0, np.random.exponential(mean_gap_days)))
        api["events"] += 1            # a release invalidates custom code
        api["maintenance"] += api["rework_cost"]
```

Run each strategy across many seeds, average total cost (build + maintenance + seat +
switching), and report the cheapest strategy per scenario. Parameterise the startup vs
lab arrival windows directly from the transcript's 6–12 / 12–18 month ranges.

## WHEN TO USE IN PRODUCTION
Use it to sanity-check a build-vs-buy decision before committing budget, or to model the
"switching cost" the video calls "a huge time and money drain." It is a decision-support
model, not a forecast: do NOT present its numbers as predictions of real vendor
timelines, and do not use it where you already know compliance rules force a specific
platform (for example a clinic needing a BAA).

## HOW IT HELPS STUDENTS
You learn discrete-event simulation, generators as processes, seeded reproducibility,
and Monte Carlo aggregation with NumPy — a rare, employable skill set. You also learn to
translate a podcast's qualitative claims into explicit, testable parameters and to
explain model assumptions, which is exactly what production ML/ops work demands.

## STARTER CODE
See `starter/main.py` — a seeded SimPy environment with a release-event process, three
strategy runners, and a Monte Carlo comparison across the video's timing windows.
