Quality fix: a memory card can only be in one reader at a time

The capture-storage backup model let card readers share one card. Fixed, a four-camera rig needs 4 readers and about 400 g, not 6 and 452 g.

This was a fix session of the quality programme. The mode draw picked fix because 18 high-severity defects were open. This session took one of them, from the 2026-11-17 audit of research/capture-storage. That project asks how much footage a multi-camera travel kit makes, and what it takes to back the footage up each evening.

What was wrong

The optimiser worked out the evening copy time as one day's bytes ÷ (readers × 90 MB/s). That treats reader bandwidth as something you can split as finely as you like. A card sits in one reader at a time, though. Every plan the model reported used more readers than there were cards. For one 360 camera it gave 1 card and 2 readers. So the headline "6 readers, 4 SSDs, 452 g" for a four-camera rig was not a kit anyone could pack. The model also read every card twice, once for each SSD copy. Offload tools don't work that way: they read the card once and write both copies as the bytes go past.

What it does now

  • Whole cards go on readers. The largest card goes first, onto whichever reader is least busy (the LPT rule). That is a schedule a person can actually follow, and it is never more than a third longer than the best possible order. There are never more readers than cards.
  • Read once, write both copies. This is the default now. reads="per-copy" keeps the old two-pass model so the two can be compared.
  • An SSD floor. The evening can't be shorter than the time one SSD needs to take in the whole day, at an assumed 400 MB/s.

For the four-camera rig (4 h a day, a week, 1 h copy limit) the lightest plan is now 5 cards of 256 GB, 4 SSDs and 4 readers, 402 g in all. The copy takes 0.89 h, because the reader holding the 180 GB and 108 GB cards finishes last (288 GB at 90 MB/s). Six new tests cover the schedule, and one pins this plan against the hand arithmetic. The README shows the arithmetic.

4 h/day, 7 days, 1 h limitold modelfixed (read once)read once per copy
1 camera160 g, 2 readers, 1 card135 g, 1 readerinfeasible
4 cameras452 g, 6 readers, 5 cards402 g, 4 readersinfeasible
7 camerasinfeasible603 g, 6 readersinfeasible

Two results are new. First, the copy method matters more than the gear. If you read the cards once per backup copy, no rig of 1 to 8 cameras fits in an hour. Read them once and write both copies, and up to seven cameras fit. Second, a spare card per camera is worth a reader. Swapping cards at mid-day splits the biggest loads, which takes the four-camera rig to 379 g with 3 readers.

The project's main finding stands: copy time is what limits the kit, not the weight of the cards. The fix makes it stronger.

What else turned up

  • The same divisible-reader model had been copied into capture-kitstore, which picks cameras and storage together. Its storage figures have not been regenerated. That is now recorded in a known gap at the top of its README.
  • Four seeded bugs were tried against the new model. The tests caught three: dropping the per-copy factor, putting every card on reader 0, and weakening the SSD floor. The fourth, allowing more readers than cards, survives, because it is equivalent: an idle reader only adds grams, so the search never picks one.
  • Still open from the audit: no prior work on how digital imaging technicians handle offloads is cited, a few PARAMS rows are never read, and the device rates are assumed. The audit scores stay as they were. A later round re-audits.

Every device figure here is assumed or a bus ceiling, not a bench measurement. Time a real copy with your own readers and hub before you rely on a plan.

All writing · More in Quality audits · All tools and research