web: surface personal training findings — cadence, fueling

- NatesNotes.md restructured into the personal-findings doc (cadence 170
  target / 161 natural, 80 g carb/h, 3/2 breathing, gait cues, injury log)
- load_activities exposes avg_cadence_spm (summaryDTO.averageRunCadence,
  falling back to Takeout's avgDoubleCadence — full coverage, 342/342 runs)
- load_fit_records: FIT running cadence is single-leg rev/min; now returns
  true steps/min ((cadence + fractional_cadence) * 2)
- Dashboard: cadence trend chart (per-run avg, 28-day mean, 170/161 lines)
  + Cad column in recent activities
- Activity Detail: per-second cadence trace with %time >= 170
- Race Plan: fueling section (carb g/h x est. finish time per race)
- one-time migration: 18 epoch-ms-as-text timestamps -> ISO (P2 item;
  backup at data/backups/garmin-20260612-082025.db)

Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>

NEXT_SESSION.md

@@ -38,6 +38,14 @@ architecture and conventions. Delete this file when the list is done.
   round-trip + sync button all run on garminconnect; verified headlessly via
   streamlit AppTest (login restore, button click, full sync, no exceptions).
   `openrun.ingest.auth` (garth) remains only for `--backend=garth`.
+- **Personal findings surfaced in UI (2026-06-12)**: NatesNotes.md is the
+  findings doc (cadence 170/161, 80 g carb/h, 3/2 breathing, gait cues).
+  Dashboard gained a cadence trend chart + table column; Activity Detail a
+  per-second cadence trace (FIT cadence is single-leg — load_fit_records now
+  doubles it); Race Plan a fueling calculator. Targets 170/161 are constants
+  in the pages — move to openrun.toml when config grows a slot. Still unsurfaced:
+  walking/time-on-feet (Dashboard filters type='running'; off-watch walking
+  only exists in daily_steps).
 - **New goal (user, 2026-06-12): shareable web UI** — let others run the
   openrun web app against their own Garmin account. Known gaps: per-user
   token store + DB (everything is cwd-relative single-user today), per-user
@@ -57,8 +65,11 @@ architecture and conventions. Delete this file when the list is done.
   `openrun-sync`, fall back to `../garmin-pgc/` python-garminconnect backend if
   garth hits Cloudflare 429), then populate `race_plan` through 2026-09-12
   using `banister_forecast` + `calibrate_tl_per_km`; log off-watch work.
-- **P2 — data quality**: fix the 18 mixed-format timestamps, utcnow warnings,
+- **P2 — data quality**: the 18 epoch-ms timestamps were migrated to ISO via
-  add `schema_version` + tiny migration runner before any public release.
+  one-time UPDATE on 2026-06-12 (they were stored as TEXT like
+  "1658951176000.0"; backup at data/backups/garmin-20260612-082025.db).
+  Still to do: normalize on ingest so Takeout re-ingest can't reintroduce
+  them, utcnow warnings, `schema_version` + tiny migration runner.
 - **P3 — merge sync fork**: fold `../garmin-pgc/` into openrun as
   `openrun-sync --backend=garminconnect` instead of a sibling project.
 - **P4 — roadmap items reordered for the ultra**: TrainingReadinessDTO +

NatesNotes.md

@@ -1,47 +1,78 @@
-# Nate Personal Notes
+# Nate — Personal Findings
 
-## Eisley
+Working notes from PT (Eisley), self-observation, and race experience.
+Last recap: 2026-06-12 (day before the 30K).
 
-### Exercises
+## Cadence & HR — the two numbers that matter
 
-### Gait / posture
+- **170 spm is ideal**; natural cadence is **161** — body falls back to it
-- knee forward
+  when not consciously cued.
-  - stepping too far forward
+- Holding ~170 has made HR consistently maintainable **< 145** (top of Z2 is
-  - lose momentum, hard strike, pulling
+  143) — at the cost of running slower. That trade is the point.
-  - instead, run on ice
+- Watch datafield: big HR + cadence only — the only fields needed in training.
-  - feet land underneath, push backward
+- Current block (week of Jun 5–12): mostly low-HR walking, deliberate.
-  - this has revealed weakness and asymmetry in feet/ankles/legs
-  - in particular, left foot is very difficult to load medial side during pushoff
 
-- 170 bpm is ideal
+## Breathing
-  - This has made HR consistently maintainable < 145
-  - I end up running slower to maintain HR
-  - datafield on watch showing big HR/Cadence - only important fields for training
 
-- feeling of winding/turning my feet sup on lateral, inf on medial
+- **3 strides in / 2 out.**
-  - aligns with medial MTP avoidance, good queue
+- Noticed the diaphragm very often lands in the same position on the same
+  foot — odd-count breathing (3/2) rotates which foot takes the exhale impact.
 
-## Anatomy Observations
+## Nutrition
-- Natural cadence is 161. Body falls back to this if I don't think about it
-- Feeling of legs out of socket, or tailbone out of place, or sacrum twisted(?)
-- asymettery in motion of each leg
-- tightness and weakness where collarbones join
-  - feel of trying to pull that area down seems right
-- likewise in shoulders, feel out of socket or twisted/tight
 
-## Other PT Findings
+- **I have undernourished in races.**
+- Goal: **80 g carbs/hour** during long efforts.
+  - At 30K (~3.5 h?) that's ~280 g total — plan bottles/gels ahead, don't wing it.
 
-### Connor Harris
+## Gait / form cues (Eisley)
-- Big toe ball exercise
-  - side faces wall 
-  - forearm against wall, opposite leg slightly lateral
-  - raise wall-side leg marcher
-  - lift body w/far leg, focus on pressing into big toe ball
-  - imagine trying to lift toes from ground
 
-## Race Logistics
+- Problem: knee forward / stepping too far out front → lose momentum, hard
-- I have undernourished 
+  strike, pulling instead of pushing.
-  - 80g carb/hr goal
+- Cue: **"run on ice"** — feet land underneath, push backward.
-- 3 strides in / 2 out breathing
+  - This revealed weakness and asymmetry in feet/ankles/legs; in particular
-  - noticed diaphragm very often in same place on same foot
+    the **left foot struggles to load the medial side during pushoff**.
-- 
+- Cue: feeling of winding/turning the feet — sup on lateral, inf on medial —
+  aligns with the medial-MTP avoidance; good cue.
+- Ice-walk stride note from injury rehab (2026-03-08): shorter, less
+  out-in-front steps; cue of rotating musculature from inner to outer
+  (pull meat back from inside inner ankle bone, push out behind outer
+  ankle bone) felt good.
+
+## Anatomy observations
+
+- Asymmetry in the motion of each leg.
+- Feeling of legs out of socket / tailbone out of place / sacrum twisted(?).
+- Tightness + weakness where the collarbones join — the feel of pulling that
+  area down seems right; likewise shoulders feel out of socket or twisted.
+
+## Injury history
+
+- **2026-03-07, right leg**: right inner leg felt "short", tending to limp,
+  unable to pull the right leg back far; legs heavy, cardio fine.
+  Rehabbed via the ice-walk stride work above (see ../RunningLogs.md).
+
+## PT exercises (Connor Harris, YouTube)
+
+- **Big toe ball exercise**: side faces wall; forearm against wall, opposite
+  leg slightly lateral; raise wall-side leg (marcher); lift body with the far
+  leg, focus on pressing into the big-toe ball; imagine lifting the toes
+  from the ground.
+
+## 50M training strategy
+
+- Races: 30K **2026-06-13**, 50K **2026-07-25**, 50M **2026-09-12**.
+- **Minimize Z3 junk miles** — polarize: Z2 volume + deliberate quality.
+- Low-HR walking counts as training (time on feet) even though the dashboard
+  currently ignores non-running activities.
+
+## Where this lives in the data / web UI
+
+| Finding | Surfaced today? |
+| --- | --- |
+| HR < 145 / zone discipline | ✅ Time-in-zone (Activity Detail), polarized split (Dashboard) |
+| Minimize Z3 junk | ✅ Dashboard "Polarized split (last 12 weeks)" |
+| Aerobic base quality | ✅ Pa:HR decoupling (Activity Detail) |
+| **Cadence 170 vs 161** | ❌ not shown anywhere, though per-second cadence is in every FIT |
+| **80 g carb/hr fueling** | ❌ nothing — Race Plan page could compute totals from projected duration |
+| **Walking / time-on-feet** | ❌ Dashboard filters to `type="running"` only |
+| 3/2 breathing, form cues | n/a — not watch-measurable; lives in this file |

src/openrun/model.py

@@ -37,7 +37,11 @@ def load_activities(conn: sqlite3.Connection, *, type: str | None = None) -> pd.
                distance_m, duration_s, moving_duration_s,
                avg_speed_mps, max_speed_mps, avg_hr, max_hr, calories,
                elevation_gain_m, elevation_loss_m,
-               training_load, aerobic_te, anaerobic_te, vo2_max
+               training_load, aerobic_te, anaerobic_te, vo2_max,
+               COALESCE(
+                   json_extract(raw, '$.summaryDTO.averageRunCadence'),
+                   json_extract(raw, '$.avgDoubleCadence')
+               ) AS avg_cadence_spm
         FROM activities
     """
     if type:
@@ -760,7 +764,16 @@ def load_fit_records(conn: sqlite3.Connection, activity_id: int) -> pd.DataFrame
     out["heart_rate"] = df.get("heart_rate")
     out["speed_mps"] = df.get("enhanced_speed", df.get("speed"))
     out["distance_m"] = df.get("distance")
-    out["cadence_spm"] = df.get("cadence")
+    # FIT "cadence" for running is single-leg rev/min (~80s); true steps/min
+    # is (cadence + fractional_cadence) * 2.
+    cad = df.get("cadence")
+    if cad is not None:
+        cad = cad.astype("float64")
+        frac = df.get("fractional_cadence")
+        if frac is not None:
+            cad = cad + frac.astype("float64").fillna(0.0)
+        cad = cad * 2.0
+    out["cadence_spm"] = cad
     out["altitude_m"] = df.get("enhanced_altitude", df.get("altitude"))
     out["power_w"] = df.get("power")
     out["vertical_oscillation_mm"] = df.get("vertical_oscillation")

src/openrun/web/pages/1_Dashboard.py

@@ -172,7 +172,49 @@ with right:
 
 
 # ---------------------------------------------------------------------------
-# Row 4 — Recent activities
+# Row 4 — Cadence trend
+# ---------------------------------------------------------------------------
+
+# Training cue (see NatesNotes.md): hold ~170 spm; body falls back to ~161
+# when uncued. Per-run average cadence vs those two reference lines.
+CADENCE_TARGET_SPM = 170
+CADENCE_NATURAL_SPM = 161
+
+st.subheader("Cadence")
+cad_runs = (
+    runs.dropna(subset=["start_time_local", "avg_cadence_spm"])
+    if "avg_cadence_spm" in runs
+    else pd.DataFrame()
+)
+if cad_runs.empty:
+    empty_state("No cadence data yet — it arrives with synced activity details.")
+else:
+    cad = cad_runs.tail(60).copy()  # runs are sorted ascending by start time
+    latest = cad["avg_cadence_spm"].iloc[-1]
+    st.caption(
+        f"Latest run: **{latest:.0f} spm** · target **{CADENCE_TARGET_SPM}** · "
+        f"natural fallback **{CADENCE_NATURAL_SPM}** (last {len(cad)} runs)"
+    )
+    fig3, ax3 = plt.subplots(figsize=(12, 2.8))
+    ax3.scatter(cad["start_time_local"], cad["avg_cadence_spm"],
+                s=16, color="#264653", alpha=0.65, label="run avg")
+    roll = (
+        cad.set_index("start_time_local")["avg_cadence_spm"]
+        .rolling("28D").mean()
+    )
+    ax3.plot(roll.index, roll.values, color="#2a9d8f", lw=2, label="28-day mean")
+    ax3.axhline(CADENCE_TARGET_SPM, color="#2a9d8f", ls="--", lw=1, label=f"target {CADENCE_TARGET_SPM}")
+    ax3.axhline(CADENCE_NATURAL_SPM, color="#e76f51", ls=":", lw=1, label=f"natural {CADENCE_NATURAL_SPM}")
+    ax3.set_ylabel("avg cadence (spm)")
+    ax3.legend(fontsize=8, loc="lower right", ncol=4)
+    for s in ("top", "right"):
+        ax3.spines[s].set_visible(False)
+    fig3.tight_layout()
+    st.pyplot(fig3, clear_figure=True)
+
+
+# ---------------------------------------------------------------------------
+# Row 5 — Recent activities
 # ---------------------------------------------------------------------------
 
 st.subheader("Recent activities")
@@ -189,15 +231,17 @@ else:
         lambda p: f"{int(p)}:{int((p - int(p)) * 60):02d}/km" if pd.notna(p) else "—"
     )
     recent_runs["distance"] = recent_runs["distance_km"].round(2).astype(str) + " km"
+    recent_runs["cadence"] = recent_runs.get("avg_cadence_spm", pd.Series(dtype=float)).round(0)
     display = recent_runs[[
         "start_time_local", "activity_name", "distance", "pace",
-        "avg_hr", "training_load", "vo2_max",
+        "avg_hr", "cadence", "training_load", "vo2_max",
     ]].rename(columns={
         "start_time_local": "When",
         "activity_name": "Name",
         "distance": "Distance",
         "pace": "Pace",
         "avg_hr": "Avg HR",
+        "cadence": "Cad",
         "training_load": "TL",
         "vo2_max": "VO₂",
     })

src/openrun/web/pages/3_Race_Plan.py

@@ -12,6 +12,7 @@ Workflow:
 
 from __future__ import annotations
 
+import re
 from datetime import date, timedelta
 
 import matplotlib.pyplot as plt
@@ -265,3 +266,47 @@ else:
             if race_rows:
                 st.subheader("Race-day projection")
                 st.dataframe(pd.DataFrame(race_rows), hide_index=True, width="stretch")
+
+
+# ---------------------------------------------------------------------------
+# Fueling plan
+# ---------------------------------------------------------------------------
+
+st.subheader("Fueling")
+st.caption(
+    "Race-experience finding (NatesNotes.md): historically undernourished — "
+    "plan carbs ahead, don't wing it. Default target **80 g/h**."
+)
+
+
+def _race_km(label: str) -> float | None:
+    """Pull a distance out of a race label like 'wk 4 — 30K' or '50 MILE'."""
+    m = re.search(r"(\d+(?:\.\d+)?)\s*(mile|mi\b|k\b)", label, re.IGNORECASE)
+    if not m:
+        return None
+    val = float(m.group(1))
+    return val * 1.60934 if m.group(2).lower().startswith("m") else val
+
+
+if cfg.races:
+    carb_rate = st.number_input(
+        "Carb target (g/h)", min_value=30, max_value=120, value=80, step=5,
+        help="60–90 g/h is the trained-gut range for ultras; 80 is the plan.",
+    )
+    fuel_cols = st.columns(len(cfg.races))
+    for i, (label, d) in enumerate(cfg.races):
+        km = _race_km(label)
+        # Conservative trail-ultra default pace (7.5 km/h); edit per race.
+        default_h = round((km / 7.5) * 2) / 2 if km else 4.0
+        with fuel_cols[i]:
+            hours = st.number_input(
+                f"{label} ({d}) — est. finish (h)",
+                min_value=1.0, max_value=24.0, value=float(default_h), step=0.5,
+                key=f"fuel_hours_{i}",
+            )
+            total = carb_rate * hours
+            st.metric(label, f"{total:.0f} g carbs",
+                      f"≈ {total / 25:.0f} × 25 g gels over {hours:.1f} h",
+                      delta_color="off")
+else:
+    st.info("Add races to openrun.toml to plan fueling.")

src/openrun/web/pages/8_Activity_Detail.py

@@ -187,6 +187,32 @@ if fit_linked:
         else:
             ax = plot_fit_decoupling(records, segments=segments)
             st.pyplot(ax.figure, clear_figure=True)
+
+            # Cadence trace — training cue is ~170 spm, natural fallback ~161
+            # (see NatesNotes.md). 30 s rolling mean to cut sensor noise.
+            cad = records.dropna(subset=["cadence_spm"])
+            cad = cad[cad["cadence_spm"] > 0]  # drop standing/paused samples
+            if not cad.empty:
+                st.subheader("Cadence (per-second from FIT)")
+                smooth = (
+                    cad.set_index("elapsed_s")["cadence_spm"]
+                    .rolling(30, min_periods=5, center=True).mean()
+                )
+                fig, axc = plt.subplots(figsize=(8, 2.4))
+                axc.plot(smooth.index / 60, smooth.values, color="#264653", lw=1.2)
+                axc.axhline(170, color="#2a9d8f", ls="--", lw=1, label="target 170")
+                axc.axhline(161, color="#e76f51", ls=":", lw=1, label="natural 161")
+                axc.set_xlabel("elapsed (min)")
+                axc.set_ylabel("spm")
+                axc.legend(fontsize=8, loc="lower right")
+                for s in ("top", "right"):
+                    axc.spines[s].set_visible(False)
+                fig.tight_layout()
+                st.pyplot(fig, clear_figure=True)
+                st.caption(
+                    f"Moving average cadence: **{cad['cadence_spm'].mean():.0f} spm** · "
+                    f"time ≥170: **{(cad['cadence_spm'] >= 170).mean() * 100:.0f}%**"
+                )
     except FileNotFoundError as exc:
         st.error(str(exc))
 else: