MobiSys Supplementary
Use this when assembling MobiSys supplementary material. The supplement can support the paper,
but the 12-page double-column body must stand on its own — a reviewer may never open it.
Supplement structure
- Put full protocols, extra device sweeps, additional ablations, and per-device tables in a
clean appendix after the references (outside the 12-page cap).
- Put a demo video or media where the contribution is an interactive service or a
user-facing behavior a static figure cannot convey; MobiSys work often lives or dies on the
demo.
- Put executable assets in the artifact package rather than the appendix when the current
HotCRP form separates them (
mobisys-artifact-evaluation).
- Keep all supplementary files double-blind: no author names, institutions,
acknowledgements, device serial numbers, account handles in screenshots, internal codenames,
or repository owners in trace paths.
- Do not use the supplement to hide essential motivation, the core system description, the
headline device results, or the key design decisions.
- Verify archives and media open on a clean machine, render without your account, and contain
no credentials, caches, or hidden OS metadata.
Body / appendix / media split
| Content |
Where it goes |
Why |
| System design, headline latency + energy figures, accuracy-cost table |
12-page body |
The paper must be judgeable without the supplement |
| Full workload traces, per-device sweeps, extended ablations |
Appendix after references |
Supports the body under reviewer discretion |
| Interactive behavior, UI, real-time feel |
Demo video/media |
A CDF cannot show a user-facing service |
| Runnable code, scripts, model files |
Artifact package |
Evaluated by the AEC, not read as prose |
Appendix architecture
- Order appendix sections to mirror the body's experiment order; systems reviewers navigate by
figure and section number, not by page.
- Restate the setup before each extended result so the appendix reads standalone without
flipping back to the double-column body.
- Cross-reference every appendix table and figure from the body at least once; orphaned
supplement material is invisible under reviewer discretion.
- Keep the demo media short and self-explanatory; assume it is watched once, muted, at speed.
Vignette: splitting an on-device runtime paper
A submission presenting a thermal-aware inference runtime, plus device studies: the body keeps
the design, the latency-under-load curve, the energy-per-frame figure, and the accuracy-cost
table; the appendix holds the four-device sweep and the full workload traces; a 90-second demo
video shows the frame rate holding as the phone heats; the artifact carries the runtime and
the measurement scripts. Nothing decision-critical lives only in the appendix or the video,
because both are discretionary.
Output format
[Supplement status] Ready / Needs fixes / Not ready
[Body/appendix/media split] <what sits where, gaps flagged>
[Files] <appendix / demo video / traces / artifact>
[Anonymity checks] <passed / issues incl. screenshots and trace paths>
[Main-paper dependency] <what breaks if the supplement is ignored>
1---2name: mobisys-supplementary3description: Use when preparing MobiSys supplementary material — appendices, extra device sweeps, protocol details, demo video and media, and anonymized artifacts under the 12-page-body, double-blind, and reviewer-discretion constraints, including how to split a mobile-systems paper between body, references-and-appendix overflow, and HotCRP fields.4---56# MobiSys Supplementary78Use this when assembling MobiSys supplementary material. The supplement can support the paper,9but the 12-page double-column body must stand on its own — a reviewer may never open it.1011## Supplement structure1213- Put full protocols, extra device sweeps, additional ablations, and per-device tables in a14 clean appendix after the references (outside the 12-page cap).15- Put a **demo video or media** where the contribution is an interactive service or a16 user-facing behavior a static figure cannot convey; MobiSys work often lives or dies on the17 demo.18- Put executable assets in the artifact package rather than the appendix when the current19 HotCRP form separates them (`mobisys-artifact-evaluation`).20- Keep all supplementary files **double-blind**: no author names, institutions,21 acknowledgements, device serial numbers, account handles in screenshots, internal codenames,22 or repository owners in trace paths.23- Do not use the supplement to hide essential motivation, the core system description, the24 headline device results, or the key design decisions.25- Verify archives and media open on a clean machine, render without your account, and contain26 no credentials, caches, or hidden OS metadata.2728## Body / appendix / media split2930| Content | Where it goes | Why |31|---|---|---|32| System design, headline latency + energy figures, accuracy-cost table | 12-page body | The paper must be judgeable without the supplement |33| Full workload traces, per-device sweeps, extended ablations | Appendix after references | Supports the body under reviewer discretion |34| Interactive behavior, UI, real-time feel | Demo video/media | A CDF cannot show a user-facing service |35| Runnable code, scripts, model files | Artifact package | Evaluated by the AEC, not read as prose |3637## Appendix architecture3839- Order appendix sections to mirror the body's experiment order; systems reviewers navigate by40 figure and section number, not by page.41- Restate the setup before each extended result so the appendix reads standalone without42 flipping back to the double-column body.43- Cross-reference every appendix table and figure from the body at least once; orphaned44 supplement material is invisible under reviewer discretion.45- Keep the demo media short and self-explanatory; assume it is watched once, muted, at speed.4647## Vignette: splitting an on-device runtime paper4849A submission presenting a thermal-aware inference runtime, plus device studies: the body keeps50the design, the latency-under-load curve, the energy-per-frame figure, and the accuracy-cost51table; the appendix holds the four-device sweep and the full workload traces; a 90-second demo52video shows the frame rate holding as the phone heats; the artifact carries the runtime and53the measurement scripts. Nothing decision-critical lives only in the appendix or the video,54because both are discretionary.5556## Output format5758```text59[Supplement status] Ready / Needs fixes / Not ready60[Body/appendix/media split] <what sits where, gaps flagged>61[Files] <appendix / demo video / traces / artifact>62[Anonymity checks] <passed / issues incl. screenshots and trace paths>63[Main-paper dependency] <what breaks if the supplement is ignored>64```
brycewang-stanford