SunShiftBuilt for the
whole journey.
A useful plan needs to follow you through every stop.
Here’s what we tested, what worked, and what remains unproven.
complete tested itineraries supported
One journey.
Continuous planning.
In our internal benchmark, SunShift supported all 30 synthetic itineraries: ten single flights, ten round trips, and ten multi-leg journeys.
The tested Dawnward version supported 11 complete itineraries. This is an itinerary-coverage result, not a score for clinical accuracy or jet-lag reduction.
Complete itineraries supported
A complete result needs the required itinerary behavior. Returning a schedule does not count if intermediate stays or changes of direction are unsupported.
| Journey type | SunShift | Dawnward |
|---|---|---|
| Single flights | 10 / 10 | 10 / 10 |
| Round trips | 10 / 10 | 0 / 10 |
| Multi-leg journeys | 10 / 10 | 1 / 10 |
| Complete itineraries | 30 / 30 | 11 / 30 |
The evaluated jetlag.ninja version is an offset-based scheduler without complete flight/date/location ownership; it supported 0 of these 30 as full itinerary plans. That does not measure the accuracy of its offset advice. Arcascope’s circadian library supplies scientific simulation models, not complete itineraries, so it is not assigned an itinerary score.
What SunShift passed
The integrated app adapter passed the independent published-instruction and full-stay-day context checks on all 30 primary scenarios. The corpus includes date-line crossings, daylight-saving changes, half-hour and quarter-hour zones, short stays, and overnight connections.
The improvements preserved accepted recommendations while suppressing rejected candidates and restoring context on quiet days. They improved publication reliability and visible coverage; modeled physiology was unchanged.
What we haven’t proven
This is an internal synthetic benchmark, not an independent clinical trial. It does not establish better physiological accuracy, faster adaptation, or fewer symptoms than another app.
Delays, cancellations, unknown ground transfers, individual physiology, and real-world adherence are outside this corpus. One separately reported reviewer regression is excluded from the 30-trip denominator.
Dawnward did better in conditional phase simulations.
For the ten single-flight cases, Dawnward produced lower simulated average phase error in all ten with Forger99 and nine with Hannay19, under the first-30-minutes light-exposure assumption.
| Model | SunShift | Dawnward |
|---|---|---|
| Forger99 | 2.185 h | 1.721 h |
| Hannay19 | 1.938 h | 1.514 h |
These are model-derived differences, not measured jet lag or clinical prediction errors. The planners used different sleep constraints, and results depend on assumed light exposure, adherence, and fallback routines. SunShift’s comfortable preset limits ordinary nightly changes to one hour and displacement to two hours. The comparison does not isolate algorithm quality under identical constraints.
A result you can inspect.
We froze the scenarios and tested native implementations, retaining raw outputs. Versions below describe the tested sources; they are not claims about later releases.
- SunShift
- sunshift-2026.09-itinerary-reconciliation-v1
- Dawnward
- ea2cfde5c6b9259e755c7b7d61338515dd2484db
- jetlag.ninja
- 2f0dfeb591b626d8dc4fa6476b502373e3e58335
- Arcascope circadian
- db6eb08b6c0b0586cdac7795bf10bbbbfd323c1a
Benchmark run: September 26, 2026. This page reports engineering coverage and conditional simulations separately. Neither establishes superior clinical accuracy.