Garmin + RaceBox video POC

01 / Inputs

Load the matching activity files

Display units

Ride files

Select multiple files or drop them here. Recording times suggest ride groups; uncertain files stay unassigned for review.

Drop files here Telemetry, video, GPS tracks, alignments, and other supporting files are inspected locally.

No files selected.

Currently loaded into analysis VideoNo video loaded Garmin TCXNo TCX loaded Garmin FITNo FIT loaded RaceBox CSVNo CSV loaded RaceBox VBONo VBO loaded AlignmentOptional saved anchors

Private VM storage · browser-side analysis

Analysis archive

Batch uploads create one archive run per reviewed ride group. The selected MP4 remains local for playback; its retained copy is stored separately.

Username or email Password

Automatically upload sources loaded into the current analysis run and save browser-generated results

Source files stay local until you sign in and choose an upload action.

Browse runs and their retained files below. File-modified time is not necessarily the ride recording time; device family is inferred from the file type, and exact device models are not currently recorded. Files can be up to 30 GiB and upload in resumable chunks with progress, rate, and ETA. After a page reload, reselect the same file (matching name, size, and modified time) to continue; the archive preserves a 5 GiB free-space reserve.

Load local files to inspect source coverage and telemetry.

Elevation comparison

Ascent by source

Paired comparison: load both activities to assess shared coverage.

Garmin TCX

Raw-sample ascent: —
1-second median ascent: —

Source not loaded.

RaceBox CSV

Raw-sample ascent: —
1-second median ascent: —

Source not loaded.

Paired difference requires both sources and independently verified timing.

Each source's ascent uses its own recorded coverage; the totals are not a synchronized comparison when their windows differ. The difference below uses only the verified shared interval, not the standalone totals above. Raw-sample ascent sums positive changes when distance advances (or is unavailable); filtered ascent uses median altitude in each 1-second bin. RaceBox distance is integrated from speed. Gaps over 2.5 seconds are omitted. The RaceBox − Garmin difference is withheld until independent timing residuals pass this POC's ≤100 ms gate. Neither grade estimate is ground truth; RaceBox altitude datum and absolute accuracy are unverified.

04 / Terrain mesh

LiDAR terrain and recorded position

Deferred to save resources

Selected local model Choose the exact file listed below; it is read locally and never uploaded.

data/fusion/racebox-lap-1-lidar-terrain-trail-200mm.stl 38,544,084 bytes · SHA-256 25a38a63b89aa2e4f79c0767164b07e715850f847c73066eabeb228f9a3f5ff6 EPSG:3071 local easting/northing on X/Y, centered at the footprint origin; Z is up from the 0 in base plane · inches · 7.87 × 7.87 in footprint · 3× terrain relief · 0.12 in base · 0.04 in raised trail

Position source Linked Garmin activity timeline Activity elapsed seconds No activity

Terrain is not loaded on startup to reduce memory and GPU use. Load it from the repository when needed.

No mesh loaded Measured bounds will appear after verification.

Load terrain to enable the verified Garmin terrain cursor.

Relative elevation and grade profiles · opt-in Show RaceBox ribbon and source profiles

Load terrain and the original RaceBox CSV/VBO to enable these profiles.

<= -6% -6 to <0% 0 to <2% 2 to <5% 5 to <8% >= 8%

02 / Video

Visual event selection

Waiting for video

Power — W

Heart rate — bpm

Speed: —mph
Distance: —mi
Ascent: —ft
Elapsed: —

Upcoming grade —

Now —

Choose a local video file to begin. Repository video is not loaded automatically.

SYNC UNVERIFIED · video telemetry is withheld until independent timing evidence passes the 100 ms gate and remains limited to the validated anchor span.

HUD telemetry channels

Toggle source-backed metrics for the HUD. Duplicate Ride HUD readings use dedicated widgets; gyroscope channels start unchecked.

Exports selected channels and layout, not telemetry or timing. Ride HUD summary/zone widgets are not rendered by the offline prototype; its selected channels use compact readout rails.

Power-zone reference · FTP (W) Default FTP is 250 W; adjust it to match your threshold. Signed-in settings are saved to your account. Age for max-HR estimate Max-HR formula

Estimated max HR Enter age

Heart-rate bands: Z1 50–60%, Z2 60–70%, Z3 70–80%, Z4 80–90%, Z5 90–100% of max HR. Values below 50% count as Z1 and above 100% as Z5. Age-based estimates are approximate.

How to work with overlay data

Load a video and telemetry from the same ride in 01 / Inputs. For another ride, add new video-to-Garmin anchors; do not reuse the current mapping unless the saved source identity matches. Video telemetry remains locked until the same event is paired independently in both timelines and residuals pass the 100 ms gate. A matcher candidate alone is not accepted.
Use the checkboxes above to select available HUD channels and choose a layout below. The Ride HUD shows source-backed values, power-zone time using the default 250 W FTP, heart-rate-zone time using the age-based max-HR estimate (220 − age by default or Tanaka), and the upcoming Garmin elevation profile. Enter an age above before heart-rate zones can be calculated. When selected and synchronized, the HUD also shows lean angle, color-zoned integer Garmin and RaceBox grades side by side, gear, cadence, satellite, breathing-rate, and heart-rate widgets. Display units are switched globally above. Gyroscope channels start unchecked. The comparison in 01 / Inputs compares raw and 1-second median ascent over the shared timestamp interval; gaps over 2.5 seconds are omitted. The Garmin and RaceBox grade comparison in 03 / Garmin timeline uses centered 10-second windows, but both grade and ascent results stay hidden until independent Garmin↔RaceBox timing evidence passes 100 ms. Neither device grade is treated as ground truth. RaceBox grade uses 1-second median altitude and distance integrated from speed. Both source pairings are provisional until independently verified. The separate Grade over time widget then plots each source's actual grade percentage—not their difference—over the preceding 60 seconds of verified coverage. The metric rows below show each metric's name, value, unit, and chart of observed samples without interpolation.
To add a field in code, parse it and document its evidence in web/src/telemetry.js, map its label and color in web/src/main.js, and add specialized visuals in web/src/hud-chart.js only when needed. Update tools/hud-server-render.js if it should also appear in the offline FFmpeg render, then add parser tests. Keep inferred units and unverified semantics explicit.

Display layout

Power-zone time, selected source-backed values, ride totals, and upcoming terrain.

Chart render path

The worker draws charts locally; video decode and HUD text stay in the browser.

Load the mapped 10-minute excerpt to enable the 5-second renderer comparison.

59.94 fps source frame step

Video media time 0:00.000

Fine adjust · video seconds

03 / Garmin timeline

Activity position and metrics

Waiting for TCX

Garmin activity time No activity

Fine adjust · elapsed seconds from activity start

Follow video only within a verified ≤100 ms alignment span

Garmin RaceBox Current position

Route motion

Ride track animation

Preview a glowing marker moving along a sample circuit or loaded GPS track.

Route Speed

Position 0%

Sample circuit loops; loaded GPS tracks play from start to finish at a steady visual pace.

RaceBox CSV cross-check Speed and Lap unavailable Waiting for timestamped RaceBox data

SYNC NOT VERIFIED · UTC/GPS proximity is diagnostic only; paired RaceBox values require independent residuals ≤100 ms and are limited to verified coverage.

Landmark label

SYNC UNVERIFIED · Pair the same visible event independently in video and Garmin; one anchor cannot verify timing.

Grade comparison

Garmin vs RaceBox

Independent estimates · neither source is ground truth

Garmin Grade —

RaceBox Grade —

RaceBox − Garmin —

Grade comparison is locked until Garmin↔RaceBox synchronization is independently verified to ≤100 ms for the displayed interval.

Metric display objects

One row per metric

Name · value · unit · 60 s source-sample chart.

Investigation tools · Timing

Review GPS route alignment

Compare GPS-derived time deltas, gyro candidates, spatial match ambiguity, and GPS position error across a ride. Hover or click a chart, or move the time slider, to inspect one local window in all methods. Reports are read in this browser; they are not uploaded or applied as clock corrections.

Loading the bundled full-ride example…

Loaded source timing

Temporal precision profile

Compare sample cadence, gaps, timestamp formatting, and ordering diagnostics for the telemetry loaded above.

Cadence is not clock accuracy

Load Garmin TCX/FIT or RaceBox CSV/VBO files to inspect their timestamp intervals.

Source	Valid / source rows	Coverage	Median interval	p95 interval	Interval MAD	Gaps >2.5× cadence median	Finest timestamp text step	Moved / equal / invalid rows

Gap duration is the total time between samples in flagged gaps. Timestamp text step reflects decimal formatting only, not device clock accuracy or measurement precision.

Source interval zoom

No input intervals available Scroll to zoom around the pointer; Shift+scroll to pan. This view has its own time range, and each stream starts at its first valid record.

Per-source sample intervals

Vertical whiskers show the min-max interval range within each horizontal display pixel. Gaps are marked, never joined.

Intervals above 2 seconds are shown as markers.

Large gap Equal timestamp Reordered row

Hover the chart to inspect raw record time, interval, and value.

Each source uses elapsed time from its own first valid record; aligned clock offsets are not inferred here. The route-vs-gyro chart below is candidate disagreement, not an independently measured timing residual. Independent anchor residuals are not present in the current matcher report.

No report loaded

Diagnostic only · corrections unapplied

Route offset range · p05–p95

Route vs gyro · p95 absolute

Close alternatives · gap ≤1 m

Position error · p95 window median

Show estimators Cubic Hermite Constant-velocity KF Experimental GPS-only EKF

Shared time zoom

Showing full range Scroll over a graph to zoom around the pointer; Shift+scroll to pan. Buttons zoom around the selected time. The route-comparison and EKF diagnostic graphs share this range; the source interval plot above has its own range.

GPS route time delta

Camera GPS time minus RaceBox GPS time. A stable offset should stay roughly flat.

GPS route vs gyro candidate difference

Camera GPS route candidate minus gyro start candidate by estimator; this is candidate disagreement, not an independent residual.

Competing route matches

How much worse a different time offset fits the GPS path. Small gaps mean ambiguity.

GPS position error

Median horizontal position error across matched fixes in each local window.

Linked window inspector

Compare all estimators at one point

0:00.0

Select an analysis time

Estimator	Route offset	GPS position error	Alternative route match	Route vs gyro	Matched fixes

How to interpret these plots

Route offset: changing local deltas can indicate poor route observability, GPS noise, or clock behavior; they are not automatically clock drift.
GPS route vs gyro candidate: the green band marks ±100 ms of agreement between two candidates. It is not an independently measured residual or error against true time; this report has no independent visual-anchor residuals.
Competing matches: the amber band marks a spatial score gap of 1 m or less to a separated offset. That guide is for investigation, not a pass/fail threshold.
Position error: high or unstable spatial residuals weaken the route match. Triangles mark values outside a chart's display range; exact values remain available in the linked-window table.
Windows overlap, so adjacent points are correlated and are not independent timing checks. Only independent visual or otherwise validated anchors can establish the absolute synchronization gate.

POC boundaries

Open validation gates

Video-to-Garmin values stay locked until independent visual evidence verifies the 100 ms limit, and are unavailable outside the validated anchor span.
RaceBox values aligned to Garmin stay locked until independent clock residuals verify the 100 ms limit, and remain unavailable outside verified coverage.
Respiration and rear pressure retain FIT gaps; unsupported lateral G is not shown as measured.
The terrain cursor uses verified original Lap=1 observations; RaceBox-to-Garmin timing is unavailable until independently verified. Final-video rendering is in scope but is not implemented or accepted yet.
Browser HEVC playback, orientation, color, dropped frames, and workstation performance still need local validation.

Ride telemetry alignment