ADR-152: WiFi-Pose SOTA 2026 intake — WiFlow-STD benchmark, Rust integrations, ADR-153 802.11bf layer, efficiency frontier (#1008)

* feat(calibration): NodeGeometry transceiver-geometry recording (ADR-152 §2.1.1)

PerceptAlign-motivated geometry capture at enrollment: per-node optional
records (position, antenna orientation, inter-node distances, acquisition
method) — recorded when known, never required. Event-sourced via
EnrollmentEvent::GeometryRecorded (latest recording wins); persisted on
SpecialistBank with serde defaults so pre-ADR-152 bank JSON loads cleanly
(fixture-proven, and geometry-free banks serialize byte-shape-identical
to the old schema); threaded through MultiNodeMixture as data only — the
learned geometry embeddings and algorithmic fusion use are §2.1.2,
deliberately deferred until the ADR-151 P6 LoRA heads exist.

Geometry recorded from now on means banks captured today remain usable
for layout-conditioned training later — you can't retroactively add
geometry to data you didn't record.

8 new tests (3 geometry, 2 anchor, 2 bank, 1 multistatic) + full-loop
extension (2-node geometry, one tape-measured + one unknown, surviving
the bank JSON round-trip the runtime loads from). 50/50 calibration
(both feature configs) + 23 CLI tests green.

Co-Authored-By: RuFlo <ruv@ruv.net>

* feat(training): two-checkerboard camera↔room calibration for ADR-079 labels (ADR-152 §2.1.3)

Defends the camera-supervised pipeline against PerceptAlign's
"coordinate overfitting": MediaPipe keypoints were emitted in raw camera
coordinates with no shared frame and no transceiver-geometry metadata —
the exact label shape that memorizes deployment layout and collapses
cross-layout.

- scripts/calibrate-camera-room.py + calibration_lib.py: OpenCV
  two-checkerboard calibration → versioned bundle JSON (intrinsics,
  camera→room extrinsics, checkerboard spec, transceiver geometry,
  sha256 calibration_id). Intrinsics resolve from file > cache >
  multi-view computation > loud-warning 2-view fallback.
- collect-ground-truth.py --calibration <bundle>: every sample gains
  keypoints_room (unit bearing rays from the camera center in the room
  frame — documented projective alignment; raw image coords preserved
  so training chooses), camera_origin_room, calibration_id, and the
  transceiver geometry stamp. Without the flag, output is byte-identical
  to before (tested) + a one-line ADR-152 warning.

Design finding (recorded for ADR-152): a single planar checkerboard's
corner grid is centrosymmetric — the reversed corner ordering fits a
ghost camera pose with IDENTICAL reprojection error, so per-board flip
disambiguation is mathematically ill-posed. solve_two_board_extrinsics
solves the joint wall+floor set over all 4 flip combinations, where the
minimum is unique — an independent reason the TWO-checkerboard method is
required, beyond what PerceptAlign states.

15 headless pytest tests green (synthetic corners: extrinsics recovery
incl. ghost resolution, bundle round-trip + hash stability, ray
transforms w/ distortion + cross-resolution, no-calibration byte
identity).

Co-Authored-By: RuFlo <ruv@ruv.net>

* feat(benchmarks): WiFlow-STD reproduction harness + measurement (a) results (ADR-152 §2.2)

Shipped checkpoint REFUTED (0.08% PCK@20, wrong keypoint normalization);
6 reproducibility defects documented (broken imports, corrupted dataset
tail with float32-max garbage that NaN-poisons fp16 BatchNorm, unreachable
test phase). After repairs, retraining with upstream defaults reproduces
96.09% PCK@20 full-test / 96.61% corruption-free (published 97.25%) on
RTX 5080. Claims graded MEASURED-EQUIVALENT; 2.23M params + ~0.055 GFLOPs
verified. Third-party code/weights/data stay out of tree (gitignored).

Co-Authored-By: claude-flow <ruv@ruv.net>

* feat: ADR-152 Rust integrations + ADR-153 802.11bf protocol model

- calibration: GeometryEmbedding — 32-slot permutation-invariant NodeGeometry
  featurization for future LoRA-head conditioning (ADR-152 §2.1.2); derived
  SpecialistBank::geometry_embedding() accessor; 59 tests
- train: MaePretrainConfig + patchify/random-mask with UNSW measured recipe
  (80% masking, (30,3) patches; ADR-152 §2.3, arXiv 2511.18792); strict
  no-truncate/no-NaN policy; proptest properties
- train: WiFlowStdModel — tch-gated port of the verified ~96%-PCK@20
  WiFlow-STD architecture (ADR-152 §2.2 beyond-SOTA); ungated param formula
  pinned to 2,225,042; 15/17-keypoint support; 239 crate tests
- hardware: ieee80211bf forward-compatibility protocol model (ADR-153):
  SpecProfile gates, SensingCapabilities negotiation, required ConsentMode,
  session FSM, SensingTransport + SimTransport + OpportunisticCsiBridge;
  full acceptance checklist covered; 156+4 tests
- deps: ruvector bumps per ADR-152 §2.6 survey (mincut/solver 2.0.6,
  attention 2.1.0, gnn 2.2.0); vendor/ruvector synced to a083bd77f
- docs: ADR-153 accepted; ADR-152 §2.2 status, §2.4 amendment, §2.6 added

Workspace: 162 test suites green (--no-default-features); Python proof PASS.
Known pre-existing flake: homecore-api env_empty_falls_back_to_defaults
(unserialized env-var mutation) — untouched, follow-up.

Co-Authored-By: claude-flow <ruv@ruv.net>

* docs: CHANGELOG + CLAUDE.md entries for ADR-152 integrations and ADR-153

Co-Authored-By: claude-flow <ruv@ruv.net>

* fix(train): repair tch-backend bit-rot — gated path compiles and tests run again

Mechanical API refresh against current tch: Vec::from(Tensor) -> try_from
(+ explicit flatten), numel() usize cast, Rem/div ops -> remainder() /
divide_scalar_mode(floor) — the latter fixed a silent true-division bug in
heatmap argmax decoding; clamp(1.0, f64::MAX) -> clamp_min (torch 2.x scalar
overflow panic); petgraph EdgeRef import; missing EvalMetrics and
verify_checkpoint_dir APIs that tests documented. wiflow_std roundtrip test
uses safetensors (.pt _save_parameters roundtrip broken in torch 2.11
Windows). Gated: 349 passed (incl. all 20 wiflow_std); ungated: unchanged.
Known pre-existing: gaussian-heatmap convention mismatch (2 tests), proof
seed race under parallel threads — documented, deliberate follow-ups.

Co-Authored-By: claude-flow <ruv@ruv.net>

* feat(train): WiFlow-STD PyTorch->tch weight import + numerical parity proof

export_to_safetensors.py maps the retrained checkpoint (295 tensors -> 248
mapped, param sum exactly 2,225,042; num_batches_tracked dropped) into a
tch-loadable safetensors plus a deterministic parity fixture. Gated #[ignore]
integration test loads it strictly and asserts forward-pass agreement:
max abs diff 1.192e-7 on the seed-42 fixture. dump_variable_names test makes
the tch name layout authoritative. Zero architecture discrepancies found.

Co-Authored-By: claude-flow <ruv@ruv.net>

* fix: workflow-review findings — BN gamma init, ThresholdParams serde, init docs

Concurrent validation workflow (2 review lanes + adversarial verification,
13 agents): 5 confirmed findings, 3 refuted. Fixes:
- wiflow_std: pin BatchNorm gamma to 1.0 (tch default draws Uniform(0,1) —
  silently halves activations in from-scratch training; loaded checkpoints
  unaffected, parity re-verified after the change)
- wiflow_std: document the conv-init divergences vs the reference's
  effective kaiming_normal(fan_out) re-init (from-scratch dynamics only)
- ieee80211bf: ThresholdParams deserialization validates via try_from so
  the <=100 invariant holds for untrusted payloads (+ rejection test)

Benchmarks (release, ruvzen): GeometryEmbedding 1.84us/call (542k/s),
MAE tokenization 7.38us/window (135k/s), 802.11bf FSM 8.9M events/s —
nothing suspicious.

Co-Authored-By: claude-flow <ruv@ruv.net>

* docs(adr): ADR-152 §2.1.4 gate resolved — PerceptAlign repo MIT, dataset on HF

Co-Authored-By: claude-flow <ruv@ruv.net>

* feat(benchmarks): edge optimization measured + measurement (b) blocked + 92.9% retraction

Edge optimization (ADR-152 optimize track): ONNX Runtime fp32 is the CPU
latency win (3.2 ms/window, ~3.4x faster than torch, parity 2.4e-7); ORT
dynamic int8 reaches 2.44 MB (paper's ~2.2 MB claim plausible only via
conv-capable toolchains; -0.16pt PCK@20, +18% MPJPE, 2x slower); torch
dynamic quant converts 0% of this conv-only model; fp16 halves storage free
but is slower on CPU.

Measurement (b) BLOCKED-ON-DATA: only 1,077 paired ESP32 windows exist
(stop rule <2k). Forensic recheck of the surviving April holdout RETRACTS
the ADR-079 '92.9% PCK@20' figure: constant-output model, absolute (not
torso) threshold, 69 near-static frames — mean predictor scores 100% under
that protocol; torso-PCK@20 is 19.1%. Corroborates PR #535. Stale citations
removed from user-guide, readme-details, ADR-152 §2.1.3; no-citation rule
extended to ADR-079 accuracy claims. Unblock: >=2k-window multi-pose paired
session + torso-PCK re-baseline.

Co-Authored-By: claude-flow <ruv@ruv.net>

* docs(user-guide): corrected camera-supervised collection tutorial

Step 0 CSI-rate check + session-length math (window yield = frames/20 —
the May session's 8x under-delivery was a ~12 Hz CSI rate, not an aligner
bug); two-checkerboard calibration step (ADR-152 §2.1.3); pose-variety and
confidence guidance; torso-normalized PCK + temporal-split + pred-variance
eval protocol (lessons from the 92.9% retraction); scale presets re-keyed
to realistic window counts.

Co-Authored-By: claude-flow <ruv@ruv.net>

* feat(benchmarks): static PTQ int8 (calibrated) results + overnight capture script

Conv-only static QDQ beats dynamic int8 on accuracy (PCK@20 96.61-96.63%
vs 96.52%, MPJPE +10% vs +18% over fp32) at ~equal size/latency; all-ops
QDQ strictly worse (int8 activations through attention glue). Entropy
calibration verified bit-identical to MinMax on this data. Deployment:
ONNX fp32 for speed (3.2ms), static conv-only QDQ for smallest (2.53MB).

Also: scripts/overnight-empty-capture.py — segmented UDP CSI recorder for
empty-room baselines (no glob collisions, detach-safe).

Co-Authored-By: claude-flow <ruv@ruv.net>

* feat(benchmarks): measurement (b) MEASURED — optimization transfer only, mean-pose baseline wins

WiFlow-STD fine-tuned on 2,046 fresh single-room ESP32 paired windows
(temporal 70/15/15, 70->540 adapter, K=17): pretrained-init 65% PCK@20 vs
scratch 0% (optimization transfer) but frozen-trunk ~0% (no feature
transfer), and NOTHING beats the mean-pose baseline (95.9% PCK@20 —
single subject, near-static normalized coords). Honesty gates held: pred
std 0.0113 (non-constant model) but mean-baseline dominance means no
citable CSI->pose capability from this data. ADR-152 open question 1
answered partially; definitive answer needs multi-subject/position data.

Two new aligner findings: heterogeneous csi_shape with silent zero-padding
(~20%), and extractCsiMatrix's transposed shape label (frame-major data,
[nSc, nFrames] label) — fixes pending.

Co-Authored-By: claude-flow <ruv@ruv.net>

* feat(benchmarks): efficiency sweep MEASURED — half model dominates full reference

Compact WiFlow-STD variants on the same data/split/protocol: half (843,834
params, 0.38x) strictly dominates the 2.23M reference (PCK@20 96.62 vs
96.61, PCK@50 99.47 vs 99.11, MPJPE 0.00898 vs 0.0094) — the published
architecture is over-parameterized for its own benchmark. quarter (338k)
96.05%; tiny (56,290 params, 1/39.5) holds 94.11% — a ~220KB fp32 edge
candidate. In-domain caveats recorded; cross-domain untested.

Co-Authored-By: claude-flow <ruv@ruv.net>

* feat(train): compact WiFlow-STD presets in Rust + tiny edge artifact (ADR-152)

WiFlowStdConfig gains half()/quarter()/tiny() mirroring the overnight sweep
exactly: TcnGroupsMode (Fixed/Gcd/Depthwise), input_pw_groups, derived
stride schedule and decoder-mid (all default to upstream behavior; legacy
serde JSON unaffected). Param formulas pin to trained ground truth first
try: 843,834 / 338,600 / 56,290; default 2,225,042 pin and 1.192e-7 parity
unchanged. 248 tests green.

Tiny edge artifact (tiny_edge_bench.py): ONNX fp32 = 295 KB, 0.66 ms/win
(~1,500/s CPU), 94.11% PCK@20 (matches sweep clean-test exactly; parity
1.49e-7). Static int8 is a bad trade at this scale (-1.43pt, +19% MPJPE,
-16% size, slower) — recorded as negative result. Export note: width-16
breaks AdaptiveAvgPool((15,1)) TorchScript export; replaced by exact
mean+matmul equivalent, proven by parity.

Co-Authored-By: claude-flow <ruv@ruv.net>

* fix: resolve all 10 confirmed code-review findings (7-angle review, 20/20 verified)

wiflow_std: min_feature_width (default 15) replaces the keypoints->stride
coupling — for_keypoints(17) now provably builds the trained [2,2,2,2]
graph and pools 15->17, matching the validated Python protocol (pinned by
tests); param_count() total on invalid configs; random_mask returns Result
and rejects non-finite/out-of-range ratios; trainer checkpoints switched
to safetensors (.pt VarStore roundtrip broken on Windows torch 2.11).

ieee80211bf: SBP proxy now re-triggers instances and relays reports via
Action::RelaySbpReport -> SensingFrame::SbpReport (clients consume via
their existing path); missed_instances reset on success = consecutive
semantics; SessionTable gains a guarded SBP entry point + unknown-id drop
counter; initiator-role sessions reject inbound setup/SBP requests
(RejectedNotSupported) closing the idle hijack; StartSetup/StartSbp
outside Idle return InvalidStateForCommand; SBP validation unified
through evaluate_setup with a 1:1 SetupStatus->SbpStatus mapping.
events.rs split out to honor the 500-line cap.

calibration/cli: enrollment geometry now actually reaches trained banks —
both production call sites attach .with_geometry; --geometry flag on
train-room and POST /enroll/geometry + train-body geometry on
calibrate-serve give production a recording surface; geometry-free banks
log the ADR-152 §2.1.2 note.

benchmarks: corruption masks committed as ground truth (unregenerable
after in-place cleaning; verified bit-identical regeneration from the
pristine copy) + generate_corruption_masks.py producer; _bench_common.py
dedups the 5x-copied shim/evaluate/seed/remap (post-refactor PCK@20
re-verified equal to the last digit); remote scripts get the mmap patch;
tiny_edge --calib validated multiple-of-64; onnx_bench --help no longer
executes (and overwrote) the export — artifact restored byte-exact.

Workspace: 2,963 tests passed, 0 failed; Python proof PASS.

Co-Authored-By: claude-flow <ruv@ruv.net>

* ci: build workspace tests without debuginfo — runner disk exhaustion

The combined 38-crate debug target exceeds the GitHub runner's disk
('final link failed: No space left on device'); the same tree measured
151GB locally with full debuginfo. CARGO_PROFILE_{DEV,TEST}_DEBUG=0
shrinks the target ~5-10x; debuginfo serves no purpose in CI test runs.

Co-Authored-By: claude-flow <ruv@ruv.net>

benchmarks/wiflow-std/RESULTS.md

@@ -0,0 +1,486 @@
+# WiFlow-STD (DY2434) Benchmark Results — ADR-152 §2.2
+
+Upstream: <https://github.com/DY2434/WiFlow-WiFi-Pose-Estimation-with-Spatio-Temporal-Decoupling>
+pinned at `06899d29` (2026-04-05), Apache-2.0. Dataset: Kaggle `kaka2434/wiflow-dataset`
+(12.8 GB archive → 15.5 GB extracted; 360,000 windows of 540×20 CSI + 15-keypoint 2D labels).
+
+Published claims (README "Setting 1"): PCK@20 97.25%, PCK@30 98.63%, PCK@40 99.16%,
+PCK@50 99.48%, MPJPE 0.007 m, 2.23M params, 0.07 GFLOPs.
+
+## Measurement (a): their model on their data
+
+### Artifact verification (MEASURED, 2026-06-10, this repo `eval_repro.py`)
+
+| Check | Result |
+|---|---|
+| Parameter count | **2,225,042 (2.23M) — matches claim** |
+| FLOPs (torch profiler, batch 1) | ~0.055 GFLOPs — consistent with 0.07B claim |
+| CPU latency (Windows box, torch 2.12 CPU) | 13.2 ms/window @ batch 1 (76/s); 2.48 ms/sample @ batch 64 (403/s) |
+| Checkpoint load | `weights_only=True` (no pickle code execution) |
+
+### Released checkpoint does NOT reproduce the claims — REFUTED as shipped
+
+Running the released `best_pose_model.pth` through the released code on the released
+dataset with the released split procedure (seed-42 file-level 70/15/15; 54,000 test
+samples) yields:
+
+| Metric | Published | Measured (shipped checkpoint) |
+|---|---|---|
+| PCK@20 | 97.25% | **0.08%** |
+| PCK@30 | 98.63% | 0.78% |
+| PCK@40 | 99.16% | 5.53% |
+| PCK@50 | 99.48% | 15.42% |
+| MPJPE | 0.007 | **NaN** (dataset contains NaN CSI windows) |
+
+Raw output: `results/repro_a.json`.
+
+Diagnostics (on 2,000 NaN-free windows from the first files of the dataset, i.e.
+mostly would-be *training* data — so this is not a split mismatch):
+
+- Predictions correlate with targets (Pearson r ≈ 0.76) — the checkpoint is a trained
+  model, but in a **different keypoint normalization/order** than the released data.
+- Best-case post-hoc global per-axis affine correction: PCK@20 ≈ 20%.
+- Best-case per-keypoint affine correction (15×2 fitted transforms — generous
+  cheating): PCK@20 ≈ 72%, still far below 97.25%.
+- Pred↔target keypoint correspondence matrix is degenerate (multiple predicted
+  keypoints best-match the same target joint) — keypoint convention mismatch.
+
+### Reproducibility defects in the released artifacts
+
+1. `models/__init__.py` imports `TemporalConvNet`, which `models/tcn.py` does not
+   define — **the published code does not import/run as-is**.
+2. The released root checkpoint uses pre-rename module names (`att.*`, `final_conv.*`)
+   vs the published code (`attention.*`, `decoder.*`) — same shapes/param count, but
+   confirms the checkpoint predates the published code.
+3. The second shipped checkpoint (`cross_dataset_test/WiFlow/best_pose_model.pth`) is
+   a **different architecture** (342-channel input = MM-Fi layout, 3 TCN layers,
+   3-channel/3D decoder) — not usable on their own dataset.
+4. `run.py` ignores `--data_dir` and hardcodes `../preprocessed_csi_data`.
+5. The released dataset's final 13 files (indices 487–499; 9,072 windows, 2.52%)
+   are corrupted: NaN values plus garbage amplitudes up to 3.4e38 (float32 max) in
+   data that is otherwise [0,1]-normalized. Upstream code has no NaN/inf handling;
+   training as published on this download diverges — the first corrupted batch
+   overflows fp16 autocast and permanently poisons BatchNorm running statistics
+   (GradScaler step-skipping does not protect BN). The authors' training curves
+   show normal convergence, so their local data evidently differed from the
+   Kaggle upload. Window masks: `results/nan_windows_mask.npy`,
+   `results/big_windows_mask.npy`.
+
+### Reproducing the corruption masks
+
+The two mask files (9,070 NaN/Inf windows, 9,072 with |amplitude| > 1.5;
+union 9,072, all in dataset files 487–499) are **committed ground truth**
+(gitignore-negated, ~352 KB each). They can only be regenerated from a
+**pristine** Kaggle download: `remote/clean_v2.py` repairs the dataset by
+zeroing the corrupted windows in place, after which the corruption evidence
+is gone and a rescan returns all-False. `generate_corruption_masks.py`
+re-derives them (chunked scan, criteria: any non-finite value OR
+max |finite| > 1.5 per 540×20 window) and refuses to write all-False masks,
+which indicate a cleaned copy. Verified 2026-06-11: a regeneration from the
+local pristine download is bit-identical to the committed masks.
+
+### Retraining result (MEASURED, 2026-06-10): claims APPROXIMATELY REPRODUCED
+
+Since the shipped checkpoint is unusable, measurement (a) fell back to retraining
+with upstream code + defaults (seed 42, batch 64, early-stopped at epoch 41 of 50,
+best epoch 36, ~75 s/epoch) on ruvultra (RTX 5080). Deviations, all forced and
+documented: one-line fix for defect (1); torch 2.x+cu128 instead of pinned 2.3.1
+(Blackwell sm_120 unsupported); the 9,072 corrupted windows (defect 5) zeroed
+entirely — without this the published pipeline produces NaN from epoch 1 (observed).
+Scripts mirrored in `remote/`; raw metrics in `results/eval_retrained.json`.
+
+| Metric | Published | Retrained (full test, 54,000) | Retrained (corruption-free, 52,560) |
+|---|---|---|---|
+| PCK@20 | 97.25% | **96.09%** | **96.61%** |
+| PCK@30 | 98.63% | 97.89% | 98.23% |
+| PCK@40 | 99.16% | 98.58% | 98.79% |
+| PCK@50 | 99.48% | 98.99% | 99.11% |
+| MPJPE | 0.007 | 0.0098 | 0.0094 |
+
+Within ~0.6–1.2 PCK points of every published figure (single run, corrupted train
+windows zeroed, different torch/GPU). **Verdict: the accuracy claims are credible
+and approximately reproducible — but only after repairing the released dataset and
+code.** Val best: PCK@20 96.99%, MPJPE 0.0086 (epoch 36).
+
+One more defect found during the run:
+
+6. `train.py` calls `plot_training_history`, which is not defined anywhere — the
+   built-in post-training test evaluation is unreachable as published (crashes
+   with NameError after training completes).
+
+## ADR-152 §2.2 citation rule
+
+Evidence grade for the WiFlow-STD accuracy claims after measurement (a):
+**MEASURED-EQUIVALENT (96.1–96.6% PCK@20 reproduced by retraining; shipped
+checkpoint REFUTED; dataset/code require repairs)**. RuView docs may cite
+"~96% PCK@20 (our reproduction)" — still **not comparable** to our 17-keypoint
+ESP32 numbers (different hardware, 5 subjects, in-domain random split,
+15 keypoints).
+
+## Edge optimization (measured)
+
+ADR-152 "optimize beyond SOTA" track, 2026-06-10, this Windows box (Windows 11,
+16 torch threads, torch 2.12.0+cpu, onnxruntime 1.26.0). Subject: the retrained
+checkpoint `results/retrained_best_pose_model.pth` (2,225,042 fp32 params).
+Scripts: `quantize_bench.py`, `onnx_bench.py`, `eval_ort_accuracy.py`.
+Raw numbers: `results/edge_optimization.json`.
+
+Accuracy is on a **10,000-window seed-42 random subset** of the corruption-free
+test split (same seed-42 file-level 70/15/15 split as `eval_repro.py`; 54,000
+test windows, 1,440 corrupted excluded via `results/nan_windows_mask.npy` |
+`results/big_windows_mask.npy`, leaving 52,560; subset drawn with
+`np.random.default_rng(42)`). The fp32 subset PCK@20 (96.68%) matches the full
+clean-test figure (96.61%), so the subset is representative.
+
+Latency is CPU ms/window, median of repeated runs, 3 interleaved repetitions
+per variant (medians below; run-to-run spread on this box is large, roughly
+±20-40% at batch 1 — reps are in the JSON).
+
+| Variant | Disk size | Batch 1 (ms/win) | Batch 64 (ms/win) | PCK@20 | PCK@50 | MPJPE |
+|---|---|---|---|---|---|---|
+| torch fp32 (baseline) | 9.07 MB | 11.0 | 2.27 | 96.68% | 99.15% | 0.00936 |
+| torch fp16 (`.half()`) | **4.58 MB** | 24.3 | 2.42 | 96.68% | 99.15% | 0.00946 |
+| torch int8 dynamic | 9.07 MB (unchanged) | 15.6 | 2.06 | 96.68% (identical) | 99.15% | 0.00936 |
+| ONNX fp32 (onnxruntime) | 8.97 MB | **3.2** | **2.0** | 96.68% | 99.15% | 0.00936 |
+| ONNX int8 (ORT dynamic, supplementary) | **2.44 MB** | 6.5 | 5.8 | 96.52% | 99.15% | 0.01108 |
+
+Findings:
+
+- **torch dynamic INT8 quantizes nothing on this model.** The architecture has
+  **zero `nn.Linear` layers** — it is entirely Conv1d (21) + Conv2d (22) +
+  BatchNorm. `torch.ao.quantization.quantize_dynamic` (requested over
+  `{Linear, Conv1d, Conv2d}`) converted **0 modules / 0.0% of params**: dynamic
+  quantization only has kernels for Linear/RNN-family modules and silently
+  skips convolutions. The "int8" model is bit-identical to fp32 (same outputs,
+  same 9.07 MB). Conv quantization would require static (PTQ) quantization
+  with calibration — out of scope here; the ORT dynamic path below is the
+  honest int8 datapoint.
+- **fp16 halves size for free accuracy-wise** (PCK@20 −0.005 pt, MPJPE
+  +0.0001) but is *slower* on CPU at batch 1 (~2.2×) — torch CPU fp16 conv
+  kernels are emulated. fp16 is a storage/transport format here, not a CPU
+  runtime win.
+- **ONNX Runtime is the real batch-1 latency win: ~3.4× faster than torch**
+  (3.2 vs 11.0 ms/window) at identical accuracy (parity 2.4e-7).
+
+### Verdict on the paper's "~2.2 MB int8" claim
+
+**Plausible but not free, and unreachable by the obvious PyTorch route.**
+2,225,042 params × 1 byte ≈ 2.2 MB assumes *every* parameter quantizes.
+PyTorch dynamic quantization — the one-liner most readers would reach for —
+yields **9.07 MB (0% quantized)** because the model has no Linear layers.
+ONNX Runtime dynamic quantization, which does have int8 conv weight support,
+gets **2.44 MB** (close to the claim; the overhead is BatchNorm params/buffers
+and quantization scales kept in fp32) at a measurable accuracy cost:
+PCK@20 96.68 → 96.52% (−0.16 pt) and MPJPE 0.00936 → 0.01108 (+18%), and
+~2× slower inference than ONNX fp32 (ConvInteger kernels). The paper does not
+state a method or an int8 accuracy; treat "2.2 MB" as a weight-arithmetic
+estimate, achievable in practice only via conv-capable quantization toolchains
+and with a small accuracy penalty.
+
+### ONNX export status
+
+**Works.** Exported via the TorchScript exporter (`dynamo=False`), opset 17,
+with a dynamic batch axis — `results/retrained_fp32_dynamic.onnx` (8.97 MB),
+verified to run at batch 1/2/64. The axial attention's
+`view(N*W, C, H)` reshape traced correctly (sizes recorded as graph ops, not
+baked constants). The dynamo exporter also captures the graph but crashed on
+this box writing a ✅ to a cp1252 console (cosmetic Windows encoding issue, not
+a model blocker). Parity vs torch on the stored fixture
+(`results/parity_fixture.npz`, batch 2, seed 42): **max abs diff 2.4e-7 —
+PASS** (< 1e-4). ORT-quantized int8 model: `results/retrained_int8_ort_dynamic.onnx`.
+
+### Static PTQ (calibrated) — follow-up
+
+Follow-up to the dynamic-int8 row above (2026-06-10, same box, onnxruntime
+1.26.0): ONNX Runtime **static** post-training quantization
+(`quantize_static`, QDQ format, per-channel int8 weights + int8 activations)
+of the same fp32 export, calibrated on **corruption-free TRAINING-split
+windows only** (seed-42 file-level split, same masks; 1,000 windows for
+MinMax, 512 for the histogram calibrators; never test windows). Scopes:
+"conv-only" (`op_types_to_quantize=["Conv"]` — the attention path exports as
+Einsum/Softmax, which ORT never quantizes anyway, so "all-ops" additionally
+quantizes the elementwise Mul/Sigmoid/Add/AveragePool glue). Accuracy on the
+identical 10k-window seed-42 corruption-free test subset; latency median of
+3 interleaved reps (fp32/dynamic re-benched in-session as references).
+Script: `static_ptq_bench.py`; raw: `results/edge_optimization.json`
+(`onnx_static_ptq`).
+
+| Variant | Disk size | Batch 1 (ms/win) | Batch 64 (ms/win) | PCK@20 | PCK@50 | MPJPE |
+|---|---|---|---|---|---|---|
+| ONNX fp32 (reference) | 8.97 MB | 2.5 | 1.9 | 96.68% | 99.15% | 0.00936 |
+| ORT dynamic int8 (baseline) | **2.44 MB** | 5.7 | 4.6 | 96.52% | 99.15% | 0.01108 |
+| static QDQ **Percentile(99.99) conv-only** | 2.53 MB | 5.3 | 4.7 | 96.61% | 99.16% | **0.01031** |
+| static QDQ MinMax conv-only | 2.53 MB | 5.2 | 3.3 | **96.63%** | 99.19% | 0.01084 |
+| static QDQ Entropy conv-only | 2.53 MB | 5.2 | 3.1 | 96.60% | 99.19% | 0.01078 |
+| static QDQ MinMax all-ops | 2.60 MB | 6.5 | 3.9 | 95.45% | 99.14% | 0.01486 |
+| static QDQ Entropy all-ops | 2.60 MB | 5.7 | 4.1 | 95.30% | 99.13% | 0.01510 |
+| static QDQ Percentile all-ops | 2.60 MB | 5.3 | 4.3 | 96.39% | 99.17% | 0.01218 |
+
+**Verdict: static PTQ (conv-only) is the new best int8 point on accuracy —
+but only modestly, and it does not fix int8's latency penalty.**
+
+- **Accuracy: beats dynamic.** All three conv-only calibrations land at
+  PCK@20 96.60–96.63% (vs dynamic 96.52%, fp32 96.68% — recovers ~⅔ of the
+  dynamic gap) and MPJPE 0.0103–0.0108 (vs dynamic 0.01108). Best MPJPE:
+  Percentile conv-only, +10% over fp32 instead of dynamic's +18%.
+- **Size: slightly worse.** 2.53 MB vs 2.44 MB (+3.6%) — QDQ nodes and
+  per-channel scales cost a little; BatchNorm stays fp32 in both (the 12 BNs
+  follow Slice/Einsum/Reshape, never Conv, so they cannot be folded).
+- **Latency: a wash vs dynamic, still ~2× slower than ONNX fp32 at batch 1.**
+  Batch-1 medians 5.2–5.3 vs dynamic 5.7 ms/win in-session — within this
+  box's ±20–40% noise. Batch 64 leans static (3.1–3.3 for MinMax/Entropy
+  conv-only vs 4.6), same caveat.
+- **All-ops QDQ is strictly worse**: up to −1.4 pt PCK@20 and +60% MPJPE for
+  zero size/latency benefit — int8 activations through the elementwise glue
+  around the attention blocks is where the damage is. Conv-only is the right
+  scope.
+- Negative result worth recording: **Entropy calibration is a no-op here** —
+  on an identical calibration set it selects full-range thresholds
+  bit-identical to MinMax (all 247 scales equal; verified on a 64-window
+  smoke set). Also, ORT 1.26's `CalibMaxIntermediateOutputs` raises a
+  spurious "No data is collected" when the batch count divides the chunk
+  size (worked around in the script).
+
+Deployment guidance: need speed → ONNX fp32 (3.2 ms b1). Need int8 weights
+for size → static QDQ conv-only (Percentile or MinMax,
+`results/retrained_int8_static_percentile_conv.onnx`), which strictly
+dominates dynamic int8 on accuracy at ~equal latency and +0.09 MB.
+
+## Efficiency sweep (MEASURED, overnight 2026-06-10/11)
+
+ADR-152 beyond-SOTA track: compact purpose-built variants of the WiFlow-STD
+architecture, trained from scratch on the same cleaned dataset, identical
+seed-42 file-level split, loss and protocol as the measurement-(a) reference
+(fp32, batch 64, ≤50 epochs, patience 5; RTX 5080, ~22–29 min/variant).
+Variant transforms are pure channel/group/stride scalings of an
+architecture-exact parameterized model (validated: reproduces 2,225,042 params
+at the reference config). Scripts: `remote/sweep/`; raw:
+`results/efficiency_sweep.jsonl`; checkpoints `results/{half,quarter,tiny}_best.pth`
+(gitignored).
+
+| Variant | Params | vs 2.23M | Clean-test PCK@20 | PCK@50 | MPJPE | Best epoch |
+|---|---|---|---|---|---|---|
+| full (reference, meas. a) | 2,225,042 | 1× | 96.61% | 99.11% | 0.0094 | 36 |
+| **half** | **843,834** | **0.38×** | **96.62%** | **99.47%** | **0.00898** | 23 |
+| quarter | 338,600 | 0.15× | 96.05% | 99.43% | 0.00928 | 50 |
+| tiny | 56,290 | 0.025× | 94.11% | 99.36% | 0.0125 | 47 |
+
+Findings:
+
+- **The half model (843k params) strictly dominates the full reference** on
+  this dataset — equal PCK@20, better PCK@50 and MPJPE, converges in fewer
+  epochs. The published 2.23M architecture is over-parameterized for its own
+  benchmark.
+- **tiny (56k params, 1/39.5) holds 94.11% PCK@20** — a ~220 KB fp32 /
+  ~60 KB int8-class model in reach of severely constrained edge targets,
+  at −2.5 pt from the full reference.
+- Caveats: in-domain (5-subject random-file split) like every number on this
+  dataset; single run per variant; corruption-free test subset (52,560).
+  Cross-domain behavior of compact variants is untested — ADR-150's evidence
+  says capacity *hurts* cross-subject, so the compact end may generalize no
+  worse, but that is a hypothesis, not a measurement.
+
+### Compact-variant edge artifacts (MEASURED, 2026-06-11)
+
+Edge pipeline for the **tiny** checkpoint (56,290 params), same machinery and
+protocol as the full-model edge rows above (this Windows box, torch
+2.12.0+cpu, onnxruntime 1.26.0; dynamic-batch opset-17 TorchScript export;
+static QDQ **Percentile(99.99) conv-only** int8 calibrated on **512**
+corruption-free TRAIN-split windows; accuracy on the identical 10k-window
+seed-42 clean test subset; latency = median ms/window over 3 interleaved
+reps, with the full-model fp32/int8 sessions interleaved as same-session
+references). Script: `tiny_edge_bench.py`; raw:
+`results/edge_optimization.json` (`tiny_variant`). Torch-vs-ORT parity on the
+stored fixture input: **max abs diff 1.5e-7 — PASS** (< 1e-4). The tiny fp32
+subset PCK@20 (94.11%) matches the full clean-test sweep figure (94.11%)
+exactly, so the subset remains representative.
+
+Two forced deviations, both recorded in the JSON:
+
+1. **Adaptive-pool export rewrite.** tiny's derived stride schedule
+   `[2,1,1,1]` leaves feature width 16, and the TorchScript exporter rejects
+   `AdaptiveAvgPool2d((15,1))` when 15 is not a factor of the input height
+   (the full model never hit this — its width was exactly 15). Since the
+   pool over a fixed-size map is a fixed linear operator, the export wrapper
+   replaces it with `mean(-1)` (W axis, a factor) + a constant averaging
+   matmul using PyTorch's exact bin rule; the parity check (vs the original
+   torch model with the real pool) proves exactness.
+2. **Calibration count 512, not "~500"**: ORT 1.26's histogram collector
+   `np.asarray()`'s the per-batch maxima, so the calibration count must be a
+   multiple of the 64-window calibration batch or the ragged last batch
+   crashes it (the earlier static-PTQ run dodged this by using exactly 512).
+
+| Variant | Disk size | Batch 1 (ms/win) | Batch 64 (ms/win) | PCK@20 | PCK@50 | MPJPE |
+|---|---|---|---|---|---|---|
+| full ONNX fp32 (same-session ref) | 8.97 MB | 2.27 | 1.42 | 96.68% | 99.15% | 0.00936 |
+| full static QDQ Percentile conv-only (same-session ref) | 2.53 MB | 5.53 | 3.82 | 96.61% | 99.16% | 0.01031 |
+| **tiny ONNX fp32** | **0.295 MB** | **0.66** | **0.24** | **94.11%** | 99.37% | 0.01253 |
+| tiny static QDQ Percentile conv-only | 0.248 MB | 0.85 | 1.03 | 92.68% | 99.33% | 0.01491 |
+
+(tiny torch `.pth` checkpoint for reference: 0.34 MB on disk; 56,290 fp32
+params ≈ 225 KB of weights.)
+
+Findings:
+
+- **The smallest deployable WiFlow-class model is the tiny ONNX fp32
+  artifact: ~295 KB on disk, 0.66 ms/window batch-1 CPU (~1,500 windows/s),
+  94.1% PCK@20** — 30× smaller and ~3.4× faster (in-session) than the full
+  ONNX fp32 model for −2.6 pt PCK@20.
+- **int8 is a bad trade at this scale.** Static QDQ conv-only — the recipe
+  that cost the full model only 0.07 pt — costs tiny **−1.43 pt** PCK@20
+  (94.11 → 92.68%) and +19% MPJPE, saves only 47 KB (−16%; QDQ scales and
+  the fp32 BN/attention glue are proportionally larger in a small graph),
+  and is *slower* than tiny fp32 (0.85 vs 0.66 ms b1; 1.03 vs 0.24 ms b64 —
+  QDQ kernel overhead dominates when the convs are this small). A 56k-param
+  model has little redundancy left to absorb weight+activation rounding.
+- Deployment guidance, compact edition: ship tiny as **ONNX fp32** — at
+  295 KB the int8 size saving solves no real constraint and costs accuracy
+  and speed. If ~250 KB vs ~295 KB ever matters, weight-only quantization
+  would be the thing to try next, not QDQ.
+
+## Measurement (b): BLOCKED-ON-DATA (attempted 2026-06-10)
+
+The fine-tune-on-ESP32 measurement stopped at dataset characterization, per the
+pre-registered stop rule (<2,000 paired windows). Findings (MEASURED):
+
+- **Only one trainable paired dataset exists**: `ruvultra:~/work/cog-pose-train/paired.jsonl`
+  — 1,077 windows (one subject, one room, one 29.9-min session, single node;
+  CSI [56, 20]; 17 COCO keypoints, MediaPipe confidence mean 0.44 — only 264
+  windows pass ADR-079's own conf>0.5 training filter). Prior measured attempts
+  on this exact set: 0–3% torso-PCK@20 (temporal splits, three independent
+  pipelines). Fine-tuning a 2.23M-param model on ~860 train windows would
+  measure memorization, not transfer.
+- **The April session behind the old "92.9% PCK@20" claim is lost** (345
+  samples, 35 subcarriers; raw CSI gone from ruvzen/ruvultra/cognitum-v0; only
+  a 69-sample predictions+GT holdout survives at `models/wiflow-real/eval-holdout.jsonl`).
+- **Forensic recheck of that holdout RETRACTS the 92.9% figure**: the trainer's
+  `pck()` used an absolute 0.2 image-unit threshold (not torso-normalized) and
+  the model output a **constant pose** (pred std 0.0000 across 69 near-static
+  frames; a mean predictor scores 100% under the same protocol). The
+  torso-normalized PCK@20 on the same holdout is 19.1%. This corroborates the
+  2026-05-11 audit retraction (CHANGELOG, PR #535); stale doc citations were
+  removed 2026-06-10 (user-guide, readme-details, ADR-152 §2.1.3). The §2.2
+  no-citation rule now applies to ADR-079 accuracy claims.
+
+Unblock criteria: a paired collection session of ≥2k windows (≈35+ min at the
+observed stride; multi-pose, conf>0.5, ideally with the §2.1.3 two-checkerboard
+calibration), plus a re-baselined our-pipeline number under torso-PCK@20 on the
+same split. WiFlow-STD assets stand ready on ruvultra (`~/wiflow-std-bench/`).
+Also worth investigating: ADR-079's protocol predicts ~9k windows per 30 min;
+the May session under-delivered ~8× (aligner drop rate?).
+
+## Measurement (b) (MEASURED 2026-06-10/11)
+
+The data baseline unblocked: the 2026-06-10 22:10–22:40 collection session produced
+**2,046 paired windows** (`ruvultra:~/wiflow-std-bench/paired-20260610.jsonl`; ONE
+subject, ONE room, ONE ESP32 node, varied poses: walk/raise/squat/kick/wave/turn/
+jump/sit; aligner `scripts/align-ground-truth.js`, non-overlapping 20-frame windows
+~0.42 s; 17 COCO keypoints in normalized [0,1] camera coords; MediaPipe confidence
+mean 0.802, min 0.692 — all windows pass the conf>0.5 filter). The −4 h timestamp
+bug and the empty-frame confidence-dilution aligner findings are recorded
+separately; results only here. Trained on ruvultra (RTX 5080, torch 2.11+cu128,
+fp32, batch 32, GPU shared with the efficiency sweep). Scripts mirrored in
+`remote/measb/`; raw metrics + full training curves in `results/measurement_b.json`.
+
+### Two new aligner/dataset findings (forced deviations, MEASURED)
+
+1. **`csi_shape` is heterogeneous, not [70, 20]**: 1,347× [70,20], 284× [134,20],
+   243× [26,20], 130× [12,20], 42× [20,20]. The ESP32 stream emits mixed frame
+   types and `extractCsiMatrix` stamps each window's subcarrier count from
+   `window[0].subcarriers`, zero-padding/truncating the other frames — even
+   native-70 windows contain ~20.4% internally zero-padded short frames
+   (subcarriers 40–69 all-zero). Handling: the primary suite ("all 2,046")
+   linearly resamples every frame's subcarrier axis to 70 bins (identity for
+   native-70 frames) so the pre-registered n and split sizes hold; a secondary
+   suite restricts to the 1,347 native [70,20] windows as a homogeneity check.
+2. **Aligner layout bug**: `extractCsiMatrix` fills `matrix[f * nSc + s]`
+   (frame-major) but declares `shape: [nSc, nFrames]` — the stored shape label is
+   transposed relative to the data. Confirmed by coherent per-frame zero-tails;
+   corrected on load (`reshape(nFrames, nSc).T`).
+
+### Protocol (pre-registered, followed)
+
+Temporal split, no shuffling across time: first 70% train (1,432), next 15% val
+(307), last 15% test (307); seed 42 elsewhere. Model: learned 1×1 Conv1d 70→540
+adapter prepended to the upstream WiFlow-STD trunk; K=17 via the parameter-free
+adaptive pool (`AdaptiveAvgPool2d((17,1))` — pretrained weights load strict for
+any K). CSI normalized by the TRAIN-split p99 amplitude (129.7 all / 130.9
+native-70), clipped to [0,1]. Three runs, ≤60 epochs, early-stop patience 8 on
+val MPJPE, AdamW (adapter lr 1e-4; pretrained trunk lr 1e-5, 10× lower; scratch
+all 1e-4), fp32. Pretrained init = the measurement-(a) **retrained** checkpoint
+(`upstream/test/best_pose_model.pth`, ~96% PCK@20 on WiFlow data; the
+`att.`/`final_conv.` key remap from `eval_repro.py` applied defensively — a no-op,
+that checkpoint already uses post-rename keys). Frozen-trunk run: trunk
+`requires_grad=False` **and** held in `.eval()` so BatchNorm running stats cannot
+drift — a pure transfer probe; only the 70→540 adapter (38,340 params) trains.
+
+PCK is torso-normalized with **torso = ‖l_shoulder(5) − l_hip(11)‖** (upstream
+`calculate_pck` math — per-frame norm clamped at 0.01, mean over keypoints ×
+frames — but upstream's `NECK_IDX/PELVIS_IDX = 2, 12` is a 15-keypoint
+convention; on 17-kp COCO those indices are right_eye/right_hip, so the indices
+were replaced, not the math). MPJPE is in normalized image units (not meters).
+
+### Results — primary suite, all 2,046 windows (test = last 307)
+
+| Run | PCK@10 | PCK@20 | PCK@30 | PCK@40 | PCK@50 | MPJPE | pred std | best ep |
+|---|---|---|---|---|---|---|---|---|
+| **mean-pose baseline** (honesty bar) | **73.1%** | **95.9%** | **98.7%** | 99.3% | 99.3% | **0.0148** | 0 (by constr.) | — |
+| (i) pretrained-init, full fine-tune | 26.0% | 65.0% | 88.0% | 96.4% | 98.9% | 0.0313 | 0.0113 | 58/60 |
+| (ii) scratch | 0.0% | 0.0% | 0.0% | 0.0% | 0.0% | 0.2554 | 0.0002 | 4 (stop @13) |
+| (iii) frozen-trunk (adapter only) | 0.0% | 0.0% | 0.2% | 3.2% | 14.4% | 0.1260 | 0.0073 | 59/60 |
+
+Secondary suite (native [70,20] windows only, n=1,347, test=202) reproduces the
+same ordering: mean-baseline 96.0% / pretrained 67.1% / scratch 0.0% /
+frozen-trunk 0.0% PCK@20 (MPJPE 0.0153 / 0.0318 / 0.2236 / 0.1343) — the
+subcarrier-resampling choice does not change any conclusion.
+
+### Interpretation
+
+- **Did pretraining-transfer happen? Partially — as optimization transfer, not
+  feature transfer, and not past the honesty bar.**
+  - *Pretrained vs scratch*: dramatic (65.0% vs 0.0% PCK@20). The pretrained init
+    is the only configuration that trains at all under the pre-registered budget.
+  - *Frozen-trunk*: near-zero (0.0% PCK@20, 14.4% @50). WiFlow-STD's frozen
+    features do **not** transfer to our ESP32 domain through a linear subcarrier
+    adapter — the pretrained benefit is a well-conditioned initialization (incl.
+    calibrated BN/output scales), not reusable CSI→pose features.
+  - *Everything vs mean-pose baseline*: **no run beats it.** A constant
+    train-mean pose scores 95.9% torso-PCK@20 / 0.0148 MPJPE on this test split,
+    because a single subject in one camera frame barely moves in normalized
+    coordinates. The fine-tuned model is a real, non-constant model
+    (pred std 0.0113 > 0 — passes the constant-pose detector that retracted the
+    old 92.9% figure) but its deviations from the mean hurt: it fits train-period
+    temporal dynamics that do not generalize across the temporal split.
+- **Verdict for ADR-152 §2.2(b): fine-tuning WiFlow-STD on this dataset does not
+  demonstrate CSI→pose signal beyond the mean pose.** Until a model beats the
+  mean-pose baseline on a temporal split, no PCK number from this line may be
+  cited as pose-estimation capability.
+
+### Caveats (honest, pre-registered)
+
+- Single subject, single room, single session (30 min), single ESP32 node —
+  in-domain temporal split only; nothing here speaks to cross-room or
+  cross-subject generalization.
+- 2k windows vs the 360k-window WiFlow-STD corpus — **NOT comparable** to the
+  ~96% in-domain measurement-(a) number, and the published 97.25% even less so.
+- The scratch run's total collapse (it cannot even reach the mean pose; its
+  output BatchNorm/SiLU head must learn output scale from random init at lr 1e-4)
+  is an optimization outcome under the fixed budget, not proof the architecture
+  cannot learn from scratch — the pretrained-vs-scratch gap partially reflects
+  this conditioning advantage.
+- Mixed-subcarrier frames (finding 1) mean even the "clean" windows carry ~20%
+  zero-padded frames; collection-side frame-type filtering should precede the
+  next session.
+- Mean-baseline PCK is inflated by low pose variance relative to torso size
+  (~0.2–0.3 image units); PCK@10 (73.1%) shows the same ceiling effect at a
+  stricter threshold — the bar is the bar, but a livelier dataset would lower it.
+
+## Pending
+
+- (b) fine-tune on our ESP32 17-keypoint eval set — **MEASURED 2026-06-10/11**,
+  see above: no run beats the mean-pose baseline; pretraining transfers as
+  optimization aid only.
+- (c) our internal WiFlow on their dataset (15-keypoint subset mapping) — also
+  affected: there is currently no validated internal pose model to compare
+  (the 92.9% artifact is retracted; the MM-Fi SOTA models in ADR-150 §3 are a
+  different input domain).