docs: fix how-it-works inaccuracies, drop choose-a-kernel page, sweep stale claims

CHANGELOG.md

@@ -4,6 +4,16 @@ All notable changes to this project will be documented here.
 Format follows [Keep a Changelog](https://keepachangelog.com/en/1.1.0/) and the
 project adheres to [Semantic Versioning](https://semver.org/spec/v2.0.0.html).
 
+## [Unreleased]
+
+### Removed
+
+- The interactive kernel picker page (`docs/choose-a-kernel.html`). It predates
+  the API consolidation: with `maxsim` dispatching on layout and the patchers /
+  native PyLate & colpali-engine backends covering the framework paths, a
+  decision tree over many entry points no longer reflects the library. The
+  README "Choose a kernel" section and links are gone with it.
+
 ## [0.4.2] - 2026-06-09
 
 ### Added

README.md

@@ -16,7 +16,6 @@
 ---
 
 [[How it works]](https://hcompai.github.io/late-interaction-kernels/how-it-works.html)
-[[Kernel picker]](https://hcompai.github.io/late-interaction-kernels/choose-a-kernel.html)
 [[Benchmarks]](docs/benchmarks.md)
 [[Design]](docs/design.md)
 [[Changelog]](CHANGELOG.md)
@@ -178,10 +177,12 @@ is asserted at `atol=1e-2` before timing.
 
 |             | Rerank /<br>inference | PyLate<br>cached-contrastive | PLAID rerank<br>vs `fast_plaid` | Fused D-head<br>(training) | FP8 vs bf16<br>(Hopper) | LateOn-Code-edge<br>e2e |
 | ----------- | --------------------- | ---------------------------- | ------------------------------- | -------------------------- | ----------------------- | ----------------------- |
-| **Speedup** | 1.7-16×               | 5.0-6.9×                     | 8-23× full<br>18-51× partial    | 1.5-4.5×                   | 1.1-1.3×                | 1.00-1.06×              |
+| **Speedup** | 1.7-16×               | 5.0-6.9×                     | 8-23× full<br>18-51× partial    | 0.94-4.5×                  | 1.1-1.3×                | 1.00-1.06×              |
 
 Rerank is vs both the eager fp32-accumulator path *and* `torch.compile`;
-PLAID rerank includes top-k; the fused D-head win grows with `Nd · Ld`;
+PLAID rerank includes top-k; the fused D-head win grows with `Nd · Ld`
+(the two smallest LateOn shapes are 0.94-0.95×, i.e. slightly slower;
+≥1.4× from `Nd=128, Ld=1024` up — every ColBERT/ColPali-scale shape);
 FP8 is at `Ld ≥ 256`. Full tables and reproduction commands live in
 [`docs/benchmarks.md`](docs/benchmarks.md); for how the bench scripts
 themselves are organised — CLI conventions (`--only`, `--variants`),
@@ -191,16 +192,18 @@ per-script summaries, and how to run one bench, the whole sweep, or a
 
 ### Memory
 
-The naive einsum allocates the full `[Nq · Nd · Lq · Ld]` similarity tensor as
-fp32 scratch before `max(-1)`. The fused kernel never writes it: document
-tiles stream through SRAM and only `[Nq, Nd]` scores come back, plus a
-`[Nq · Nd, Lq]` int32 argmax buffer when training.
+The naive einsum materialises the full `[Nq · Nd · Lq · Ld]` similarity
+tensor in fp32 before `max(-1)`; its column reports the measured allocator
+peak, which runs above the similarity tensor alone because the fp32-cast
+operand copies coexist with it. The fused kernel never writes any of that:
+document tiles stream through SRAM and only `[Nq, Nd]` scores come back,
+plus a `[Nq · Nd, Lq]` int32 argmax buffer when training.
 
 | shape                                     | naive scratch | fused fwd | fused fwd + bwd |
 | ----------------------------------------- | ------------- | --------- | --------------- |
 | `Nq=1, Nd=1k, Lq=32, Ld=300`              | 183 MB        | 4 KB      | 128 KB          |
 | `Nq=1, Nd=1k, Lq=128, Ld=1024` (ColPali)  | 1.0 GB        | 4 KB      | 512 KB          |
-| `Nq=16, Nd=32, Lq=32, Ld=8192`            | 2.1 GB        | 64 KB     | 64 KB           |
+| `Nq=16, Nd=32, Lq=32, Ld=8192`            | 2.1 GB        | 2 KB      | 64 KB           |
 
 The ColPali row assumes a short text query expanded to `Lq = 128`
 (ColBERT-style query augmentation) against a `Ld ≈ 1024`-patch page.
@@ -216,27 +219,6 @@ buffer, no atomics, deterministic) for roughly half the backward peak, e.g. a
 `B256 × 16-neg` ColPali step from 4.3 GB to 2.2 GB. Full tables in
 [`docs/benchmarks.md`](docs/benchmarks.md#memory).
 
-## Choose a kernel
-
-<table>
-<tr>
-<td width="55%" valign="middle">
-
-Not sure which entry point fits your stack? The docs site ships an interactive decision tree that narrows the public API down to the right function in four questions (stack · phase · layout · workload):
-
-**👉 [hcompai.github.io/late-interaction-kernels/choose-a-kernel.html](https://hcompai.github.io/late-interaction-kernels/choose-a-kernel.html#choose-a-kernel)**
-
-</td>
-<td width="45%" valign="middle" align="center">
-
-<a href="https://hcompai.github.io/late-interaction-kernels/choose-a-kernel.html#choose-a-kernel">
-  <img src="assets/kernel_picker_widget_preview.webp" alt="Pick a kernel · interactive decision tree" width="420">
-</a>
-
-</td>
-</tr>
-</table>
-
 ## API
 
 | Symbol                                                | What it does                                                          |

docs/benchmarks.md

@@ -662,17 +662,20 @@ step time, same VRAM, same numerics, one `patch_pylate()` call.
 
 ## Memory
 
-Naive allocates `Nq · Nd · Lq · Ld · 4` bytes of fp32 scratch. Fused
-keeps the `Nq · Nd` result plus — only if autograd is on — a
-`Nq · Nd · Lq` int32 argmax buffer.
+The naive column is the measured allocator peak. Its floor is the
+`Nq · Nd · Lq · Ld · 4`-byte fp32 similarity tensor, but the measured
+peak runs above that formula because the fp32 casts and broadcasts of
+the operands coexist with the tensor at the peak. Fused keeps the
+`Nq · Nd` result plus — only if autograd is on — a `Nq · Nd · Lq` int32
+argmax buffer.
 
 
 | scenario                          | naive scratch | fused fwd | fused fwd + argmax |
 | --------------------------------- | ------------- | --------- | ------------------ |
 | `Nq=1, Nd=1000, Lq=32, Ld=300`    | 183 MB        | 4 KB      | 128 KB             |
 | `Nq=128, Nd=128, Lq=32, Ld=300`   | 621 MB        | 64 KB     | 2 MB               |
 | `Nq=1, Nd=1000, Lq=128, Ld=1024`  | 1.0 GB        | 4 KB      | 512 KB             |
-| `Nq=16, Nd=32, Lq=32, Ld=8192`    | 2.1 GB        | 64 KB     | 64 KB              |
+| `Nq=16, Nd=32, Lq=32, Ld=8192`    | 2.1 GB        | 2 KB      | 64 KB              |
 
 
 This is what unlocks large in-batch negatives: at the same HBM budget