You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
One MINOR fix.
The §13 incremental-shipping table said Stage 1.5 reads Stage 1 files
"Yes (wire-format-identical)" — bit-identical at the bytes level but a
Stage 1.5 reader using EDEN-`S` to dequantize codes written under
`S = 1` produces incorrect values (the codes index a centroid grid
scaled for S=1; dividing by a different S at decode time yields a
different vector than the encoder intended).
Resolution: commit explicitly to "Stage 1 is preview, no backward-compat
commitment; Stage 1.5 is the first stable wire format." Pre-Stage-1.5
files must be re-encoded. Documented in §6 (Stage 1 refinement) and §13
(Incremental shipping table). The alternative — adding a per-file
`s_variant` metadata field — was considered but rejected for permanent
metadata cost on a short-lived experimental phase.
Signed-off-by: Will Manning <will@willmanning.io>
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Signed-off-by: Will Manning <will@willmanning.io>
| 1 (MSE) | Preview only (no compat) | N/A (first preview) | Single block, biased only, `S = 1`. Wire format is provisional. |
1698
+
| 1.5 (EDEN-`S`) |**First stable release**| No — Stage 1 was preview; Stage 1 files must be re-encoded | Layers EDEN-`S` on top; no metadata or storage shape change. Pins the wire format. |
0 commit comments