v0.2: wrap the M7 codegen campaign

CLAUDE.md

@@ -14,15 +14,17 @@ which of its decisions were superseded. Read PLAN.md before implementing anythin
 re-derive decisions it has already settled (compile-time direction, profile vocabulary, the
 three-leg test strategy, the pinned-eps cross-validation design).
 
-Current state: **v0.1 (M6 complete), M7a measurement harness landed.** Design/schedule/tables
-(M2), the streaming converter for float/Q15/Q31 with committed scipy reference vectors (M3/M4),
-the golden cross-validation against SampleRateTap at pinned eps (M5, test-only submodule), the
-bluetooth_bridge example + C ABI + executed demo notebook (M6), and the embedded CI matrix +
-instruction-count ratchet (M7a: Cortex-M33/M55 + Hexagon under QEMU, eight fixed workloads gated
-two-sided ±3% against `bench/baselines.json` — `scripts/icount.py`). Next: the M7 levers, one
-measured PR each, superblock codegen first — see PLAN.md section 7. Any change that moves a
-workload's count beyond ±3% must re-record baselines (`icount.py --update` per target) in the
-same PR; an *improvement* beyond tolerance fails the gate too, by design.
+Current state: **v0.2 — M7 codegen phase complete.** v0.1 (M0–M6): design/schedule/tables, the
+streaming converter for float/Q15/Q31 with committed scipy reference vectors, the golden
+cross-validation against SampleRateTap at pinned eps (test-only submodule), bluetooth_bridge +
+C ABI + executed notebook. M7 (v0.2): the embedded CI matrix + instruction-count ratchet
+(Cortex-M33/M55 + Hexagon under QEMU, eight workloads gated two-sided ±3% against
+`bench/baselines.json` — `scripts/icount.py`), then three measured codegen levers — superblock
+walk, committed trip counts, symmetry-halved tables — outputs bit-identical throughout; PLAN.md
+section 7 records each lever's numbers. Remaining levers are deferred until a consumer pulls
+them (they change the output contract). Any change that moves a workload's count beyond ±3%
+must re-record baselines (`icount.py --update` per target) in the same PR; an *improvement*
+beyond tolerance fails the gate too, by design.
 
 ## The charter constraints (load-bearing)
 

CMakeLists.txt

@@ -1,5 +1,5 @@
 cmake_minimum_required(VERSION 3.24)
-project(RatioTap VERSION 0.1.0 LANGUAGES CXX)
+project(RatioTap VERSION 0.2.0 LANGUAGES CXX)
 
 # ==============================================================================
 # RatioTap — synchronous 44.1 <-> 48 kHz sample rate conversion, as fast as

PLAN.md

@@ -273,11 +273,18 @@ executed (it measures the shipping C++, not a Python re-implementation).
     (Hexagon down_q31 +2.7%); Arm came out slightly ahead (M33 Q31 −2.5%).
 
 v0.1 ships at M6. Nothing in M7+ blocks it. **Status: M0–M6 complete —
-v0.1 shipped (2026-07-23). M7a measurement harness + M7b superblock
-codegen + M7c committed trip counts + M7d symmetry halving landed
-(2026-07-23/24); next lever candidates: multistage decomposition,
-minimum-phase economy variant, or the M33 float story if a consumer
-needs it.**
+v0.1 shipped (2026-07-23). M7 codegen phase complete — v0.2 (2026-07-24):
+M7a measurement harness, M7b superblock codegen, M7c committed trip
+counts, M7d symmetry halving, all measured, outputs bit-identical
+throughout. Cumulative vs the M7a baselines: M55 Q15 −59%/−60% and float
+−35%/−37%; M33 Q31 −26%/−27%, Q15 −15%/−16%; Hexagon Q15 −13%/−10%,
+Q31 −15%/−7%; table storage halved. The remaining levers (multistage
+decomposition, minimum-phase economy, IIR pre-filter, FFT offline path)
+all change the output contract or serve currently-unpressured needs
+(storage, latency) — deferred until a consumer pulls them: a latency
+need pulls minimum-phase; a storage need pulls multistage; an MCU float
+consumer pulls the accumulation-contract discussion (a DspTap decision,
+since double accumulation is the float golden model's identity).**
 
 ## 8. Acceptance criteria (v0.1)
 

README.md

@@ -13,18 +13,24 @@ on the Tap family's shared FIR substrate
 float/Q15/Q31 sample-format traits, measured dot-product kernels, row-sum
 quantization, measurement instruments).
 
-> **Status: v0.1 (milestone M6).** The converter is in for all three sample
-> formats: float (the golden model, pinned against committed scipy
-> reference vectors sample-for-sample), Q31 (tracks float within −147 dB;
-> measures 146 dB at 997 Hz — exceeding float, whose float32 I/O is its
-> own bound), and Q15 (format-limited: pair it with `economy`, which is
-> both cheaper *and* quieter than `transparent` at 16 bits). The golden
-> cross-validation against SampleRateTap's async engine at pinned
-> eps = L/M−1 agrees to −109 dB (down) / −99 dB (up) over every phase.
-> The `bluetooth_bridge` example, the C ABI (`tools/capi/`), and the
-> executed demo notebook (`notebooks/ratio_demo.ipynb`) complete v0.1.
+> **Status: v0.2 (M7 codegen campaign).** v0.1 shipped the converter for
+> all three sample formats: float (the golden model, pinned against
+> committed scipy reference vectors sample-for-sample), Q31 (tracks float
+> within −147 dB), and Q15 (format-limited: pair it with `economy`, which
+> is both cheaper *and* quieter than `transparent` at 16 bits), plus the
+> golden cross-validation against SampleRateTap's async engine (−109 dB
+> down / −99 dB up over every phase), the `bluetooth_bridge` example, the
+> C ABI, and the executed demo notebook. v0.2 is the measured optimization
+> campaign on top — four levers, each gated by the instruction-count
+> ratchet, outputs bit-identical throughout: the superblock walk,
+> committed compile-time trip counts, and symmetry-halved tables. Since
+> the campaign's baselines: **Q15 −59%/−60% and float −35%/−37% on
+> Cortex-M55, Q31 −26%/−27% on Cortex-M33, Q15 −13%/−10% on Hexagon —
+> with table storage halved** (economy Q15 up: 6.9 KiB). The remaining
+> PLAN §7 levers (multistage, minimum-phase, IIR, FFT) change the output
+> contract and stay deferred until a consumer needs them.
 > [PLAN.md](PLAN.md) is the authoritative roadmap (charter, architecture
-> decisions, milestones, acceptance criteria);
+> decisions, milestones, acceptance criteria, per-lever measurements);
 > [HANDOFF.md](HANDOFF.md) is the original design brief it grew from.
 
 ## Quick start

include/tap/ratio/ratio.h

@@ -35,7 +35,7 @@
 #include "tap/ratio/schedule.h"    // IWYU pragma: export
 
 #define TAP_RATIO_VERSION_MAJOR 0
-#define TAP_RATIO_VERSION_MINOR 1
+#define TAP_RATIO_VERSION_MINOR 2
 #define TAP_RATIO_VERSION_PATCH 0
 
 namespace tap::ratio {