|
1 | 1 | # VisualBuilderPlan-v6 — UI Wiring + Real Math-Effect for V5 Placeholders |
2 | 2 |
|
3 | | -**Status**: planned 2026-05-22 (epic id TBD) |
| 3 | +**Status**: planned 2026-05-22 (epic `cppmega-mlx-te6`) |
| 4 | + |
| 5 | +## bd ticket ID mapping (post-creation) |
| 6 | + |
| 7 | +| H# | bd id | Title | |
| 8 | +|-----|-----------------|--------------------------------------------------| |
| 9 | +| H01 | cppmega-mlx-fjb | Sharding UI → train | |
| 10 | +| H02 | cppmega-mlx-8p1 | fp8/mixed_precision toggles | |
| 11 | +| H03 | cppmega-mlx-ou1 | Cancel button + WS abort | |
| 12 | +| H04 | cppmega-mlx-c4f | train-warm-start checkbox | |
| 13 | +| H05 | cppmega-mlx-fr5 | checkpoint path inputs | |
| 14 | +| H06 | cppmega-mlx-ays | UI N=100 real-corpus walk | |
| 15 | +| H07 | cppmega-mlx-75m | DimensionsTab Apply dispatch | |
| 16 | +| H08 | cppmega-mlx-cem | train-probe-text textarea | |
| 17 | +| H09 | cppmega-mlx-sq3 | Save/Load roundtrip parity | |
| 18 | +| H10 | cppmega-mlx-v5i | SideChannelsTab proper | |
| 19 | +| H11 | cppmega-mlx-i04 | memory peak parity | |
| 20 | +| H12 | cppmega-mlx-vlh | WS reconnect recovery e2e | |
| 21 | +| H13 | cppmega-mlx-65v | other-stages extras UI walk | |
| 22 | +| H14 | cppmega-mlx-ep7 | AblationsTab full extras | |
| 23 | +| H15 | cppmega-mlx-8rb | real per-rank shard simulation | |
| 24 | +| H16 | cppmega-mlx-oku | real mlx dtype switching | |
| 25 | +| H17 | cppmega-mlx-gqu | real cross-doc attention mask | |
| 26 | +| H18 | cppmega-mlx-6ls | MoE forward hook (entropy + load balance) | |
| 27 | +| H19 | cppmega-mlx-56o | strict identical-loss-continuation | |
| 28 | +| H20 | cppmega-mlx-4o9 | fake-rank distributed smoke | |
| 29 | +| H21 | cppmega-mlx-bqy | inference-after-resume bounded drift | |
| 30 | +| H22 | cppmega-mlx-1ne | concurrent Train clicks | |
| 31 | +| H23 | cppmega-mlx-age | fp16 dtype option | |
| 32 | +| H24 | cppmega-mlx-sby | 100+ step UI walk w/ checkpoint | |
| 33 | +| H25 | cppmega-mlx-8px | honest categorisation audit | |
| 34 | + |
4 | 35 | **Driver**: V5 closure audit found that out of 25 closed tickets, only |
5 | 36 | ~11 are real math-effect. The remaining 14 split between: |
6 | 37 | - **"backend ready + UI not wired"** — 10 gaps where pytest passes but |
|
0 commit comments