// Command aioe-sim characterizes the Adaptive Intelligence Optimization Engine updater under known
// synthetic outcome regimes and prints a report for the AEDT model card. It is a PURE calibration
// harness: no database, no client data, no network. It proves the deterministic update converges
// sensibly, keeps its clamps, and never runs away. The weights it produces are a validation
// artifact, never a shipped model: they only recover the correlations the synthetic regime encodes.
//
// Usage: aioe-sim [-cycles 40] [-events 200]
package main

import (
	"flag"
	"fmt"

	"github.com/missionworkforce/missionmatch/internal/aioe"
)

func main() {
	cycles := flag.Int("cycles", 40, "number of update cycles to run per regime")
	events := flag.Int("events", 200, "synthetic outcomes generated per cycle")
	flag.Parse()

	p := aioe.DefaultParams()
	fmt.Printf("AIOE updater characterization (pure simulation, no client data)\n")
	fmt.Printf("params: eta=%s beta=%s maxStep=%s floor=%s ceiling=%s minEvents=%d windowDays=%d\n\n",
		p.Eta, p.Beta, p.MaxStep, p.Floor, p.Ceiling, p.MinEvents, p.WindowDays)

	allSound := true
	for _, r := range aioe.StandardRegimes() {
		s := aioe.Simulate(r, *cycles, *events, p)
		fmt.Print(s.Report())
		fmt.Println()
		if !s.SumAlways1 || !s.ClampsHeld {
			allSound = false
		}
	}
	if allSound {
		fmt.Println("RESULT: every regime kept the sum-to-one invariant and the clamps. The updater is bounded and stable.")
	} else {
		fmt.Println("RESULT: an invariant was violated. Investigate before promoting any profile.")
	}
}