blackboard (13): the 23 tail sites are two kinds — 12 predicate tails (pack) + 11 algebra tails (VL descent), per Codex on #312

Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01DCfrD5y19cvFc4AoyydXYv

Author: claude

.claude/blackboard.md

@@ -51,9 +51,15 @@ retire the 12 hand-rolled `if !tail.is_empty()` sites.
 > *Correction 2026-09-16 (counted, after Codex on lance-graph #1244):* "12"
 > is the number of branches literally named `tail`; the full set of
 > hand-rolled tail branches in `simd_masking_ops.rs` at `c746735` is
-> **23** — 12 `tail` + 7 `ta` + 3 `td` + 1 `ts` (`mask_shift_morton`) — and
-> `pack<const L>` would retire all of them. Whether the Morton-shift tail
-> fits the same helper is that follow-up's question.
+> **23** — 12 `tail` + 7 `ta` + 3 `td` + 1 `ts` (`mask_shift_morton`).
+> *Scope, corrected same day (Codex on #312):* they are TWO kinds. The 12
+> `tail` branches are predicate-packing tails (one `values` slice → output
+> words) — an un-gated `pack<const L>` retires exactly those. The 11
+> `ta`/`td`/`ts` branches are mask-ALGEBRA tails (multi-input, in-place,
+> and the accumulating Morton shift) — a packer cannot express them; they
+> are what the VL descent (`U64x4::ternlog` / `U64x2::ternlog` + a
+> `tail_descend` helper over the lane op) is for. Two follow-ups, one per
+> kind; neither retires the other's set.
 
 ### 2. `examples/ternlogq_sparse_reapply_probe.rs` — FULL-WIDTH sparse frontier. NO.