Baseline operations. No unexpected events.
Training
PPO · daily env (1 step = 1 day)
12-week episodes · 200K outer steps
Weekly menu pricing, daily promo toggle
Price multiplier range [0.7, 1.3]
Pricing-review log (most recent suggestions)
Each entry is a Kairos suggestion paired with the GM's analyst note. Italicized text is generated by an LLM at pre-compute time — it reads the agent's proposal + the preceding week's stats and writes the rationale a manager would put in their notes.
Cumulative net margin: Kairos vs the do-nothing baseline
Both lines run the exact same simulated 12 weeks — same arrivals, same kitchen capacity, same random seed. Only the pricing strategy differs. The shaded green area is the margin gap created (or lost) by Kairos. Dashed vertical lines mark week boundaries. The dashed purple line extending past the current day is a 7-day-ahead forecast of Kairos's cumulative margin, built from a DOW-aware statistical projection on data observed so far (purple band = ±1 standard deviation). It uses no future ground truth.