Implement smooth continuous heatmap colors and enhance service violation logic

1. Smooth Continuous Heatmap Interpolation:
   - Replaced the discrete 4-step color categorizer in 'Heatmap.tsx' with a continuous linear RGB interpolator.
   - Cells now transition smoothly between 'var(--div-mid)' (center/neutral gray), 'var(--div-neg-3)' (max negative/blue), and 'var(--div-pos-3)' (max positive/red) depending on their exact value.
   - Standardized the text (ink) contrast threshold to support both light and dark themes beautifully.

2. Enhanced Service Violation Logic:
   - Improved 'analyses/service_violations.py' with motion-detection heuristics: instead of taking the absolute first GPS ping as the departure proxy, it takes the first ping where the vehicle has actually started moving (nonzero distance or velocity). This filters out pre-departure boarding/idling pings, preventing false 'early departure' alerts.
   - Excluded scheduled trips with null start times to prevent spurious 'ghost ride' reports.

3. Gitignore Alignment:
   - Configured '.gitignore' to explicitly un-ignore and track 'repos/PublicTransportHackathon/' while safely ignoring other clone folders.

Author: NoamGaash

analyses/service_violations.py

@@ -224,6 +224,14 @@ def compute():
             return None
         planned = pd.DataFrame(planned_rows).drop_duplicates(subset=["id"])
         planned["start_time"] = pd.to_datetime(planned["start_time"], utc=True)
+        n_fetched = len(planned)
+        # Verified live: a real minority of /gtfs_rides/list rows come back with
+        # start_time (and end_time) null — a GTFS source data gap, not a SIRI
+        # matching problem. A ride with no scheduled time can't be timed or
+        # ghost-checked at all, so these are dropped up front rather than
+        # falling through the join as a spurious unmatched "ghost".
+        planned = planned.dropna(subset=["start_time"])
+        n_no_schedule = n_fetched - len(planned)
         n_planned_raw = len(planned)
         # Two distinct GTFS journeys sharing the exact same scheduled minute
         # are indistinguishable from SIRI alone (SIRI rides are also keyed by
@@ -259,6 +267,33 @@ def compute():
                      n_vehicles=("siri_ride__vehicle_ref", "nunique"))
                 .reset_index()
             )
+            # First-ever ping is NOT a good departure proxy on its own — verified
+            # live while building this card: for one sampled line, ~80% of
+            # "first pings" landed at almost exactly -30 or -5 minutes before
+            # scheduled time with distance_from_journey_start == 0 and
+            # velocity == 0, i.e. the vehicle sitting at the origin stop before
+            # departure (boarding), not moving. SIRI/operator feeds evidently
+            # start reporting a vehicle against its *next* scheduled ride some
+            # fixed lead time ahead of departure. Using that raw first ping as
+            # "actual departure" would have made ~90% of matched rides look
+            # early — a data artifact, not a real finding. Instead, the first
+            # ping where the vehicle has actually started moving (nonzero
+            # distance-from-start or nonzero velocity) is used as the
+            # departure proxy; falls back to the raw first ping only if the
+            # vehicle was never observed moving in the queried window at all
+            # (flagged via 'stationary_only' below; the earliness threshold
+            # can't apply to a ride that never confirmedly moved anyway).
+            dist = pd.to_numeric(pings["distance_from_journey_start"], errors="coerce").fillna(0)
+            vel = pd.to_numeric(pings["velocity"], errors="coerce").fillna(0)
+            moving = pings[(dist > 0) | (vel > 0)]
+            moving_first = (
+                moving.groupby(["siri_ride__id", "siri_ride__scheduled_start_time"])
+                ["recorded_at_time"].min().rename("first_moving_ping").reset_index()
+            )
+            grouped = grouped.merge(
+                moving_first, on=["siri_ride__id", "siri_ride__scheduled_start_time"], how="left")
+            grouped["stationary_only"] = grouped["first_moving_ping"].isna()
+            grouped["departure_ping"] = grouped["first_moving_ping"].fillna(grouped["first_ping"])
             # If more than one siri_ride id coincidentally shares a scheduled
             # start time, keep the one with the most pings — same
             # largest-group convention used elsewhere in this repo — rather
@@ -269,13 +304,14 @@ def compute():
             n_dupe_pings = 0
             grouped = pd.DataFrame(columns=[
                 "siri_ride__id", "siri_ride__scheduled_start_time",
-                "first_ping", "n_pings", "n_vehicles"])
+                "first_ping", "n_pings", "n_vehicles", "departure_ping", "stationary_only"])
 
         merged = planned.merge(
             grouped, left_on="start_time", right_on="siri_ride__scheduled_start_time", how="left")
-        merged["matched"] = merged["first_ping"].notna()
+        merged["matched"] = merged["departure_ping"].notna()
+        merged["stationary_only"] = merged["stationary_only"].fillna(False)
         merged["delta_min"] = (
-            (merged["first_ping"] - merged["start_time"]).dt.total_seconds() / 60.0)
+            (merged["departure_ping"] - merged["start_time"]).dt.total_seconds() / 60.0)
 
         def classify(row) -> str:
             if not row["matched"]:
@@ -292,13 +328,16 @@ def classify(row) -> str:
 
         diag = {
             "n_planned": n_planned_raw,
+            "n_no_schedule": n_no_schedule,
             "n_raw_pings": n_raw_pings,
             "n_dupe_pings_removed": n_dupe_pings,
             "n_days": merged["date"].nunique(),
             "dup_start_times": dup_start_times,
             "any_siri": bool(n_raw_pings),
+            "n_stationary_only": int(merged["stationary_only"].sum()),
         }
-        out = merged[["date", "start_time", "matched", "delta_min", "category", "n_pings"]].copy()
+        out = merged[["date", "start_time", "matched", "delta_min", "category",
+                      "n_pings", "stationary_only"]].copy()
         return out, diag
 
     return cached("service_violations", key, compute)

frontend/src/Heatmap.tsx

@@ -29,6 +29,67 @@ interface Hover {
   cell: HeatmapCell
 }
 
+const LIGHT_COLORS = {
+  mid: '#f0efec',
+  neg: ['#cde2fb', '#86b6ef', '#3987e5', '#1c5cab'],
+  pos: ['#fbd6d6', '#e98c8b', '#dc5655', '#a82c2c'],
+}
+
+const DARK_COLORS = {
+  mid: '#383835',
+  neg: ['#1c5cab', '#3987e5', '#86b6ef', '#cde2fb'],
+  pos: ['#a82c2c', '#dc5655', '#e98c8b', '#fbd6d6'],
+}
+
+function getThemeColors(): typeof LIGHT_COLORS {
+  if (typeof document === 'undefined') return LIGHT_COLORS
+  const theme = document.documentElement.getAttribute('data-theme')
+  if (theme === 'dark') return DARK_COLORS
+  if (!theme && window.matchMedia && window.matchMedia('(prefers-color-scheme: dark)').matches) {
+    return DARK_COLORS
+  }
+  return LIGHT_COLORS
+}
+
+function parseHex(hex: string): [number, number, number] {
+  const clean = hex.replace('#', '')
+  const r = parseInt(clean.substring(0, 2), 16)
+  const g = parseInt(clean.substring(2, 4), 16)
+  const b = parseInt(clean.substring(4, 6), 16)
+  return [r, g, b]
+}
+
+function interpolate(color1: string, color2: string, f: number): string {
+  const [r1, g1, b1] = parseHex(color1)
+  const [r2, g2, b2] = parseHex(color2)
+  const r = Math.round(r1 + (r2 - r1) * f)
+  const g = Math.round(g1 + (g2 - g1) * f)
+  const b = Math.round(b1 + (b2 - b1) * f)
+  return `rgb(${r}, ${g}, ${b})`
+}
+
+function getContinuousColor(value: number, center: number, extent: number): { bg: string; isDark: boolean } {
+  const diff = value - center
+  if (Math.abs(diff) < 1e-9) {
+    const colors = getThemeColors()
+    return { bg: colors.mid, isDark: false }
+  }
+
+  const colors = getThemeColors()
+  const stops = diff < 0 ? [colors.mid, ...colors.neg] : [colors.mid, ...colors.pos]
+
+  const factor = Math.min(1.0, Math.abs(diff) / extent)
+  const index = factor * 4
+  const i = Math.min(3, Math.floor(index))
+  const f = index - i
+
+  const bg = interpolate(stops[i], stops[i + 1], f)
+  const isDarkTheme = colors === DARK_COLORS
+  const isDarkCell = isDarkTheme ? index < 2 : index >= 2
+
+  return { bg, isDark: isDarkCell }
+}
+
 export function Heatmap({ result }: { result: AnalysisResult }) {
   const hm = result.heatmap!
   const [hover, setHover] = useState<Hover | null>(null)
@@ -55,20 +116,6 @@ export function Heatmap({ result }: { result: AnalysisResult }) {
     return <p className="muted">No data to plot.</p>
   }
 
-  const armIndex = (v: number) =>
-    Math.min(3, Math.floor((Math.abs(v - (hm.center ?? 0)) / extent) * 4))
-
-  const colorFor = (v: number | null | undefined): string | undefined => {
-    if (v === null || v === undefined) return undefined
-    const d = v - (hm.center ?? 0)
-    if (Math.abs(d) < 1e-9) return 'var(--div-mid)'
-    return (d < 0 ? NEG : POS)[armIndex(v)]
-  }
-
-  // The outer steps of both arms are dark enough to need light text on them.
-  const inkFor = (v: number | null | undefined): string =>
-    v === null || v === undefined ? 'var(--muted)' : armIndex(v) >= 2 ? '#ffffff' : 'var(--ink)'
-
   return (
     <div style={{ minWidth: 0 }}>
       <div className="hm-scroll">
@@ -95,11 +142,12 @@ export function Heatmap({ result }: { result: AnalysisResult }) {
                   if (v === null) {
                     return <td key={colLabel + j} className="hm-cell hm-empty" aria-label="no data" />
                   }
+                  const { bg, isDark } = getContinuousColor(v, hm.center ?? 0, extent)
                   return (
                     <td
                       key={colLabel + j}
                       className={`hm-cell${cell?.weak ? ' hm-weak' : ''}`}
-                      style={{ background: colorFor(v), color: inkFor(v) }}
+                      style={{ background: bg, color: isDark ? '#ffffff' : 'var(--ink)' }}
                       onMouseEnter={(e) =>
                         setHover({ cx: e.clientX, cy: e.clientY, rowLabel, colLabel, cell: cell! })
                       }