NFR-P7 and NFR-P8 — half-measured, and P8 needs a decision #43

New Issue

2026-09-05T16:20:20Z

dtourolle commented

2026-09-05 16:20:20 +00:00

NFR-P7 and NFR-P8 are half-measured and untagged, and P8 needs a decision before more code helps.

NFR-P7 — export throughput

The bench's export row covers the encode half of the chain and no GPU render. So it can fail the requirement and cannot pass it. Either it measures the whole chain or the requirement is restated to be about encode.

Extend the row to cover the render, or restate NFR-P7.

NFR-P8 — memory ceiling

The memory row covers a process holding the catalog and nothing else — no toolkit, no adapter. That is the catalog layer's share of the 500 MB, not the figure the requirement is about.

And this one needs a decision, not more code. How much of the 500 MB belongs below the UI is unstated, and until somebody says, the metric can record but not judge.

docs/benchmarks.md also answers the question §4.1 raises about GPU memory — RSS cannot see device-local allocations at all — and recommends restating the requirement as two figures.

Decide the split, or adopt benchmarks.md's recommendation and restate NFR-P8 as two figures: host RSS and device-local.
Tag both once they measure what they claim to.

See docs/outstanding.md §8.

**NFR-P7 and NFR-P8 are half-measured and untagged**, and P8 needs a decision before more code helps. ## NFR-P7 — export throughput The bench's export row covers **the encode half of the chain and no GPU render**. So it can *fail* the requirement and cannot *pass* it. Either it measures the whole chain or the requirement is restated to be about encode. - [ ] Extend the row to cover the render, or restate NFR-P7. ## NFR-P8 — memory ceiling The memory row covers a process **holding the catalog and nothing else** — no toolkit, no adapter. That is the catalog layer's share of the 500 MB, not the figure the requirement is about. **And this one needs a decision, not more code.** How much of the 500 MB belongs below the UI is unstated, and until somebody says, the metric can record but not judge. `docs/benchmarks.md` also answers the question §4.1 raises about GPU memory — **RSS cannot see device-local allocations at all** — and recommends restating the requirement as two figures. - [ ] Decide the split, or adopt benchmarks.md's recommendation and restate NFR-P8 as two figures: host RSS and device-local. - [ ] Tag both once they measure what they claim to. See `docs/outstanding.md` §8.