Repository navigation
Evidence and Limits
hffgzktbcc-blip edited this page Nov 10, 2026
·
1 revision
| Status | Meaning here |
|---|---|
| Implemented | The path is present in source. Its tests or run evidence define the exercised inputs. |
| Demonstrated | A particular environment or scenario has a recorded result. |
| Experimental | An opt-in or limited path exists. A separate run is needed to establish a device or title outcome. |
| Not established | The reviewed material does not establish the broader claim. |
| Capability | Evidence and scope |
|---|---|
| Own-storage flow and companion | Implemented. The repository has the storage contract, reference server, relay, queue, and Deck runner. Hosted CI passed the automated suite and compared the checked-in Decky zip against a rebuild at the reviewed revision. That run uses browser and software checks on Ubuntu. Contract · Hosted run 40287235671 |
| Hardware launch timings | Historical demonstrations. The README attributes the numbers below to predecessor code on one Steam Deck and a cloud WebDAV store. Reported measurements |
| Prefill and early launch |
Implemented with experimental controls. The exact pre-mount fill branch requires DECKSTREAM_PREMOUNT_FILL=1 and a usable profile/cache namespace. Early launch additionally requires DECKSTREAM_EARLY_LAUNCH=1 and a qualifying measured demand curve. The code has a full-fill fallback inside that opt-in branch. Device benefit depends on the title and connection. Stream runner · Demand gate test
|
| Hedged range requests |
Experimental. DECKSTREAM_HEDGE=1 enables the hedge path. The README describes replay-rig benefit and records no Deck measurement for it. Hedge proxy · README
|
| General game compatibility, current end-to-end device timing, or NAS performance | Not established by the reviewed evidence. Automated checks and predecessor timings have different scopes. The README explicitly leaves NAS performance unbenchmarked. README |
The README describes one October 2026 home Wi-Fi setup at about 23 MB/s. Its results use different finish points.
| Case | Reported observation |
|---|---|
| Cyberpunk 2077, cold and empty cache | 12.88 GB fetched in 9 min 31 s. Title menu at 17 min 42 s. |
| Cyberpunk 2077, warm | 9 s from Play until Proton started. |
| Sifu, cold and profiled | 5.35 GB fetched in about 4 min. In control at about 7 min. |
| Baldur's Gate 3, first launch without a profile | About 25 min, with 76,000 small reads. The README attributes the bottleneck to server latency. |
Compare only runs with the same milestone, cache state, and network conditions. Testing and Reproduction