Where Should a Merged Ride Start & End?
1. The Scenario
2. Option 1: Neighbour Walk
3. Option 2: Clamp to Own Span
4. The Trade-off
15:30
15:45
16:00
16:15
16:30
16:45
17:00
Garmin FIT
1 continuous file
Continuous Garmin FIT · 89m 16s (Power + HR + Cadence across entire session)
Garmin slice 1
Garmin slice 2
Garmin slice 3
↓ fill channels ↓
↓ fill channels ↓
↓ fill channels ↓
Zwift Rides
3 separate rides
4m 39s
6m 46s
(1 W rest)
7m 24s
(74 W cool)
Zwift Ride 1 · 23m
Zwift Ride 2 · 23m
Zwift Ride 3 · 22m 19s
Merged
Ride 2
Ride 2 expands forward →
Merged
Ride 3
Ride 3: 22m 19s
SAME 6m 46s COUNTED TWICE!
Plus 2 bigger prod problems:
• Late / deleted Ride 2 freezes
wrong window in sources.tras
• Forgotten Edge at 0 W → 15.4h ride!
Result
Exact spans
Ride 1 · 23m 20s
+ Garmin channels
Ride 2 · 22m 41s
+ Garmin channels
Ride 3 · 22m 19s
+ Garmin channels
drop
dropped
dropped
✓ Zero double-counting · ✓ Zero neighbour lookups or .tras rewrites · ✓ Immune to 15h dead tails
Matches the exact start & end times athletes already see on their calendar today — while adding missing channels inside
1. Extra Warm-up / Gap Minutes Are Dropped
Garmin recorded 12 min of warm-up before a 40-min Zwift race:
12m warm-up
Garmin FIT (52 min total)
Clamped Zwift Race · stays 40 min (warm-up dropped)
2. Late-Start / Dead-Battery Recovery Is Dropped
Own device started 8 min late (or battery died) while Device 2 kept going:
8m lost
Second device recorded full 45 min ride
Own file (37 min) — clamped ride stays 37 min
How Often Does This Happen in Prod?
6-hour production sample · 13,895 merged rides
Identical duration (0s change)
92.7%
Tiny 1–60s button-press shift
5.3%
INSIDE THE REMAINING 1.75% THAT GROW > 1 MIN:
~0.90%
(~4× more common)
Junk extensions avoided
HR watches & 0 W dead tails (up to +14.5h)
~0.24%
(~1 in 420 rides)
Genuine recoveries lost
~0.20% indoor · ~0.04% outdoor
Team call: Keep clamping simple, or extend with guards + neighbour sync?
Real anonymized production ride (445450494) & 6h Preview telemetry
Step 1 of 4