From f2007326c45c3cac4d8147f18e56cbca6fdf8c79 Mon Sep 17 00:00:00 2001 From: Matysh Date: Wed, 9 Sep 2026 03:02:34 +0300 Subject: [PATCH] docs: record radar field evidence Issue: #485 User-Visible: no --- docs/specs/485-radar-presence-stage1.md | 8 ++++++++ 1 file changed, 8 insertions(+) diff --git a/docs/specs/485-radar-presence-stage1.md b/docs/specs/485-radar-presence-stage1.md index f6e7e122..322f040b 100644 --- a/docs/specs/485-radar-presence-stage1.md +++ b/docs/specs/485-radar-presence-stage1.md @@ -631,6 +631,14 @@ HA delivery, unexpected unknown coordinates, concave room, duplicate slots, disconnected client, scale change and private-source canaries — not only moving ideal points. Consented field CSV is supplementary, not a blocking dependency. +Field evidence received on 2026-09-08 was inspected locally and is not stored in +the repository. Its 68,367 observations covered all three LD2450 target slots +(47,350 / 18,344 / 2,673 rows). It confirms that `(x=0,y=0)` is the empty-slot +sentinel, while a zero on only one axis is a valid target coordinate and signed +X distinguishes the two sides of the radar. An approximately 1.4 m apparent +zone offset is recorded as Stage-2 calibration evidence; Stage 1 must not apply +an undocumented correction for it. + ## 10. Performance, release and rollback Baseline/candidate workload: 32 configured radars, eight slots each, four output