feat(ride): compute moving time from GPX with backtest

Add ride.MovingTime, a speed-threshold algorithm over track timestamps.
Add a -backtest CLI command comparing computed values against Strava's
reference across all stored rides, with a percentile distribution summary.

Baseline: mean err 3.9%, p50 1.6, p75 4.6, p90 8.4, p95 12.1, p99 44.9.

1 files changed, 117 insertions, 0 deletions

diff --git a/cli/cli.go b/cli/cli.go
index eb96356..d95f06f 100644
--- a/cli/cli.go
+++ b/cli/cli.go
@@ -4,12 +4,18 @@ import (
 	"context"
 	"database/sql"
 	"flag"
+	"fmt"
 	"log/slog"
+	"math"
+	"os"
+	"sort"
+	"text/tabwriter"
 	"time"
 
 	"github.com/martinlehoux/biking_home/config"
 	"github.com/martinlehoux/biking_home/mountain_pass"
 	"github.com/martinlehoux/biking_home/osmpass"
+	"github.com/martinlehoux/biking_home/ride"
 	"github.com/martinlehoux/biking_home/rides"
 	"github.com/martinlehoux/biking_home/web"
 	"github.com/martinlehoux/kagamigo/kcore"
@@ -23,6 +29,7 @@ var (
 	extractOSM  = flag.String("extract-osm", "", "extract mountain passes from an OSM PBF file into the database")
 	enrich      = flag.Bool("enrich", false, "backfill mountain pass coordinates from OSM data")
 	backfill    = flag.Bool("backfill", false, "recompute all stored ride values")
+	backtest    = flag.Bool("backtest", false, "run data benchmarks comparing computed values against stored reference values")
 )
 
 func Run(db *sql.DB, configPath string, appConfig config.Config) {
@@ -46,6 +53,8 @@ func Run(db *sql.DB, configPath string, appConfig config.Config) {
 		count, err := rides.Backfill(db)
 		kcore.Expect(err, "failed to backfill ride values")
 		slog.Info("Backfilled ride values", "rides", count)
+	case *backtest:
+		kcore.Expect(runBacktest(db), "failed to run backtest")
 	default:
 		runServer(db, configPath, appConfig)
 	}
@@ -56,3 +65,111 @@ func runServer(db *sql.DB, configPath string, appConfig config.Config) {
 	slog.Info("Starting web server", "address", appConfig.Server.PublicURL)
 	kcore.Expect(server.ListenAndServe(appConfig.Server.Address), "web server stopped")
 }
+
+type backtestSample struct {
+	externalID  string
+	name        string
+	computed    time.Duration
+	reference   time.Duration
+	relErrorPct float64
+}
+
+func runBacktest(db *sql.DB) error {
+	items, err := rides.List(db)
+	if err != nil {
+		return err
+	}
+	var samples []backtestSample
+	for i, item := range items {
+		if i > 0 && i%50 == 0 {
+			slog.Info("Backtest progress", "processed", i, "total", len(items))
+		}
+		if item.GPXPath == "" {
+			continue
+		}
+		parsed, err := ride.ParseFile(ride.GPXRideParser{}, item.GPXPath)
+		if err != nil {
+			slog.Warn("Failed to parse ride GPX", "ride", item.ExternalID, "file", item.GPXPath, "error", err)
+			continue
+		}
+		reference := time.Duration(item.MovingTimeS) * time.Second
+		if reference <= 0 {
+			continue
+		}
+		computed := ride.MovingTime(parsed)
+		samples = append(samples, backtestSample{
+			externalID:  item.ExternalID,
+			name:        item.Name,
+			computed:    computed,
+			reference:   reference,
+			relErrorPct: (float64(computed) - float64(reference)) / float64(reference) * 100,
+		})
+	}
+	if len(samples) == 0 {
+		slog.Info("No rides with GPX to backtest", "rides", len(items))
+		return nil
+	}
+	sort.Slice(samples, func(i, j int) bool {
+		return math.Abs(samples[i].relErrorPct) > math.Abs(samples[j].relErrorPct)
+	})
+	w := tabwriter.NewWriter(os.Stdout, 0, 0, 2, ' ', 0)
+	fmt.Fprintln(w, "computed\treference\terr %\tride")
+	limit := 10
+	if len(samples) < limit {
+		limit = len(samples)
+	}
+	for _, s := range samples[:limit] {
+		fmt.Fprintf(w, "%s\t%s\t%+.1f\t%s (%s)\n", s.computed.Round(time.Second), s.reference.Round(time.Second), s.relErrorPct, s.name, s.externalID)
+	}
+	w.Flush()
+
+	absErrors := make([]float64, 0, len(samples))
+	for _, s := range samples {
+		absErrors = append(absErrors, math.Abs(s.relErrorPct))
+	}
+	percentiles := []float64{50, 75, 90, 95, 99}
+	distribution := errorDistribution(absErrors, percentiles)
+	table := tabwriter.NewWriter(os.Stdout, 0, 0, 2, ' ', 0)
+	fmt.Fprintln(table, "percentile\tabs err %")
+	for _, p := range percentiles {
+		fmt.Fprintf(table, "p%g\t%.1f\n", p, distribution[p])
+	}
+	table.Flush()
+	slog.Info("Moving time backtest", "rides", len(samples), "mean_err_pct", meanAbsErr(absErrors), "stderr_pct", stdErrAbsErr(absErrors))
+	return nil
+}
+
+func meanAbsErr(absErrors []float64) float64 {
+	var sum float64
+	for _, err := range absErrors {
+		sum += err
+	}
+	return math.Round(sum/float64(len(absErrors))*10) / 10
+}
+
+func stdErrAbsErr(absErrors []float64) float64 {
+	mean := meanAbsErr(absErrors)
+	var sumSquares float64
+	for _, err := range absErrors {
+		sumSquares += (err - mean) * (err - mean)
+	}
+	variance := sumSquares / float64(len(absErrors)-1)
+	return math.Round(math.Sqrt(variance/float64(len(absErrors)))*10) / 10
+}
+
+func errorDistribution(absErrors []float64, percentiles []float64) map[float64]float64 {
+	sorted := append([]float64(nil), absErrors...)
+	sort.Float64s(sorted)
+	distribution := map[float64]float64{}
+	for _, p := range percentiles {
+		index := int(math.Ceil(p/100*float64(len(sorted))) - 1)
+		if index < 0 {
+			index = 0
+		}
+		if index >= len(sorted) {
+			index = len(sorted) - 1
+		}
+		distribution[p] = math.Round(sorted[index]*10) / 10
+	}
+	return distribution
+}