docs(radar): add AN-029 sports app note and a vendor doc index

docs/radar/README.md

@@ -0,0 +1,52 @@
+# OPS243 vendor documentation
+
+Vendor PDFs for the OPS243-A Doppler radar, kept here so a build doesn't depend
+on OmniPreSense's site staying up. Check the source URL before assuming a copy
+is current — OmniPreSense revises these in place under new revision letters.
+
+| File | Rev | Date | Source |
+|---|---|---|---|
+| [`AN-010-AD_API_Interface.pdf`](AN-010-AD_API_Interface.pdf) | AD | Oct 2025 | [omnipresense.com](https://omnipresense.com/wp-content/uploads/2025/10/AN-010-AD_API_Interface.pdf) |
+| [`AN-027-A_Rolling Buffer.pdf`](<AN-027-A_Rolling Buffer.pdf>) | A | Jun 2025 | [omnipresense.com](https://omnipresense.com/wp-content/uploads/2025/06/AN-027-A_Rolling-Buffer-1.pdf) |
+| [`AN-029-A_OPS243-for-Sports.pdf`](AN-029-A_OPS243-for-Sports.pdf) | A | Jun 2026 | [omnipresense.com](https://omnipresense.com/wp-content/uploads/2026/06/AN-029-A_OPS243-for-Sports_260621.pdf) |
+| [`OmniPreSense_Sports_Ball_Detect_2507.pdf`](OmniPreSense_Sports_Ball_Detect_2507.pdf) | — | Jul 2025 | Vendor slide deck, not published on the site |
+| [`OPS-DS-003-01_OPS243-Datasheet.pdf`](OPS-DS-003-01_OPS243-Datasheet.pdf) | 0.1 | — | Not publicly linked; request from OmniPreSense customer service |
+
+## Which sports note to read
+
+**AN-029 rev A** supersedes the July 2025 `Sports_Ball_Detect` slide deck for
+everything except three items, so the deck is kept rather than deleted:
+
+- Saving the tuned configuration to persistent memory with `A!` — AN-029 never
+  mentions it, and `scripts/hardware-test/test_rolling_buffer_persist.py --setup`
+  depends on that step.
+- The 10° up-angle sensor recommendation for golf, dropped rather than
+  contradicted in AN-029.
+- The `R+` inbound-only direction filter, illustrated on the water polo slide.
+
+Two errors in AN-029 rev A worth knowing before copying settings out of it:
+
+- Page 7 says to set FFT size with `S>32`. That is wrong — the correct command
+  is `X=32`, as its own Table 6 states.
+- It gives the streaming report rate for 30 ksps / 128 samples / 4096 FFT as
+  "near 200 Hz (5 ms)", where the 2025 deck gives 56 Hz (18 ms) for identical
+  settings. ~200 Hz is the theoretical frame rate (128 / 30000 = 4.27 ms); 56 Hz
+  appears to include processing overhead. The docstrings in `ops243.py` still
+  quote the older 56 Hz figure. Rolling buffer mode doesn't depend on the
+  streaming report rate either way.
+
+## AN-029 Table 6 versus what we send
+
+Table 6 matches the documented golf configuration in `src/openflight/ops243.py` —
+30 ksps (`S=30`), 128-sample buffer (`S(`), 4096 FFT (`X=32`), peak detect (`K+`),
+two-object reporting (`O2`).
+
+It differs on the minimum-speed filter, which it sets to `R>10`:
+
+- `configure_for_rolling_buffer`, the production path, sets no `R>` filter at all.
+- `configure_for_speed_trigger` sets the stricter `R>20` plus `R-` for
+  outbound-only, and deliberately uses a 256-point FFT (`X=2`) rather than
+  Table 6's 4096, trading resolution for report rate.
+- `configure_for_swing_speed_training` defaults to `R>20`.
+
+AN-029 page 8 cites OpenFlight as reference code for an OPS243-A launch monitor.