diff --git a/README.md b/README.md
index 71fb70c..99540b6 100644
--- a/README.md
+++ b/README.md
@@ -40,7 +40,19 @@ sensors/finger/
 ## Prerequisites
 
 - Python 3.10+
-- Blender version pinned in `blender.lock.json` (install separately)
+- Blender version pinned in `blender.lock.json` (install separately; see below)
+
+Install the CLI (editable):
+
+```powershell
+cd D:\orisys_ws\orisys-sensor-design
+python -m pip install -e .
+```
+
+The base install is stdlib + PyYAML. Feature extras are `dev`, `preview`, `optics`,
+`viz` and `probe`; see `requirements.txt` for the flat list, which is what the full
+test suite needs. Blender is only used by the in-scene commands, so `validate`,
+`export`, `texture` and `manifest` run without it.
 
 No Git LFS is required — binaries are committed as ordinary files.
 
@@ -66,13 +78,6 @@ Convenience launcher (auto-finds Blender for this process, then exports):
 
 ## Commands
 
-Install the CLI (editable):
-
-```powershell
-cd D:\orisys_ws\orisys-sensor-design
-python -m pip install -e .
-```
-
 Four commands for v1:
 
 ```powershell
@@ -516,54 +521,20 @@ per-cell pattern does not.
 
 ### The noise floor was wrong, and it changed the answer
 
-**Read [`docs/probe-findings.md`](docs/probe-findings.md) for the conclusions and
-recommendations.** The short version is below.
+**Read [`docs/probe-findings.md`](docs/probe-findings.md) for the conclusions, the tables and
+the numbers; [method](docs/probe-method.md) for why.** The short version:
 
-**This section replaces an earlier claim that nx=31 is the best pattern and that "the
-marker cell must match the estimator's effective window". That claim is retracted.** It
-was driven by the noise floor, and the floor was not a valid noise model.
+An earlier claim here — that nx=31 is the best pattern and that "the marker cell must match
+the estimator's effective window" — is **retracted**. It was driven by `floor_px`, which is a
+render-seed antialiasing artefact, not sensor noise: its trend reverses under additive noise,
+and it understates the real idle floor (0.2267 px p95 |flow| on `data/dd02-0703/190.mp4`) by
+roughly 2–6x, understating the coarse patterns most.
 
-`floor_px` is measured as `reference` vs a re-render at **seed + 1** — Cycles
-antialiasing jitter. On this data it differs by ~5 grey levels RMS on half the pixels and
-scales with edge contrast. It is not sensor noise, and three checks show it cannot rank
-patterns:
-
-1. **Its trend reverses** under additive noise. p95 |flow| over the ROI, same DIS config:
-
-   | pattern | render-seed | injected σ=2 | σ=4 |
-   |---|---|---|---|
-   | nx=80 | 0.0292 | **0.0161** | 0.0278 |
-   | nx=31 | 0.0376 | **0.0084** | 0.0182 |
-   | nx=56 | 0.0134 | 0.0102 | 0.0203 |
-
-2. **It understates the real floor.** Measured on idle dd02 footage
-   (`data/dd02-0703/190.mp4`, 1280x720) with the same estimator:
-   **0.2267 px** p95 |flow| over the ROI, dominant marker period **25.5 px**, i.e.
-   **0.0089 floor per period**. Short baseline (consecutive frames) equals long baseline
-   (vs frame 0) — ratio 0.98 — so this is frame-independent white noise, not drift, and
-   the clip is genuinely idle. Every simulated floor per cell is well below it, and the
-   **coarse patterns are understated most** (nx=56 by ~6x, nx=31 by ~4x, nx=80 by ~2x) —
-   exactly the bias that made nx=31 and nx=56 look best.
-
-3. **Signal is not noise-invariant either**, so gain must be quoted under a stated model.
-   At +0.10 mm on the x axis, retention is `gain(σ)/gain(clean)`:
-
-   | pattern | σ=4 | σ=16 |
-   |---|---|---|
-   | nx=80 dm1 | **99%** | **84%** |
-   | nx=56 dm1 | 96% | 71% |
-   | nx=31 dm1 | 72% | 59% |
-
-**What the calibrated model supports:** at the render scale, `radius_mm=0.5`, `dis_mode=1`,
-nx=80 has the **highest clean response on both axes** (x 9.10 vs 4.74 px/mm; y 11.69 vs
-10.90) and the **best noise retention** (84–99% vs 59–73%), so on this evidence the
-shipped 80x80 is competitive or better, not worse. What is *not* yet re-ranked: the
-process (250x250) scale and the `radius_mm=1.5` block, where the score ordering differs.
-
-Every report now carries `noise_models` (per-axis gain, floor and `response_retention` per
-σ, measured under one model) and `noise_calibration` (the real idle numbers above), and
-`settings_response.png` has a retention panel so a pattern that looks strong but collapses
-under noise is visible.
+Under the calibrated model, the shipped **80x80 has the highest clean response on both axes
+and the best noise retention** (84–99% vs 59–73%), so it is competitive or better, not worse.
+Not yet re-ranked: the process (250x250) scale and the `radius_mm=1.5` block, where the score
+ordering differs. Every report carries `noise_models` and `noise_calibration`, and
+`settings_response.png` has a retention panel.
 
 ### What the bench measures
 
diff --git a/requirements.txt b/requirements.txt
new file mode 100644
index 0000000..43c6e08
--- /dev/null
+++ b/requirements.txt
@@ -0,0 +1,18 @@
+#   python -m pip install -r requirements.txt
+#   python -m pip install -e .            # the `orisys-design` entry point
+#
+# Flat instead of `-e .[dev,preview,optics,viz,probe]`, which needs network.
+
+PyYAML>=6.0
+pytest>=7.0
+numpy>=1.24
+PyQt5>=5.15
+PyOpenGL>=3.1
+matplotlib>=3.7
+opencv-python>=4.8
+
+# tests/test_uv_map.py imports `orisys.unroll.geometry` with no importorskip guard, so the
+# suite also needs the sibling repo: pip install -e ../orisys-sdk
+#
+# Blender is not a pip dependency: install the version in blender.lock.json, on PATH or via
+# ORISYS_BLENDER. Only the in-scene commands need it.