From ef4344f0f9ba505137f0ed3fc92aae6e63ab598b Mon Sep 17 00:00:00 2001
From: ruv <ruv@ruv.net>
Date: Sun, 31 May 2026 02:04:37 -0400
Subject: [PATCH] =?UTF-8?q?docs(adr-150):=20LoRA=20calibration=20data=20re?=
 =?UTF-8?q?quirement=20=E2=80=94=20completes=20calibration=20spec?=
MIME-Version: 1.0
Content-Type: text/plain; charset=UTF-8
Content-Transfer-Encoding: 8bit

11KB adapter needs ~100-200 labeled samples/room for ~72% (knee ~50->70%);
below ~20 it hurts. Evidence-complete calibration-service spec: base +
~100-200 samples -> 11KB LoRA -> ~72% cross-subject. Encoder goal now
precisely posed: cut the sample requirement / lift the per-budget ceiling.

Co-Authored-By: claude-flow <ruv@ruv.net>
---
 docs/adr/ADR-150-rf-foundation-encoder.md | 7 +++++++
 1 file changed, 7 insertions(+)

diff --git a/docs/adr/ADR-150-rf-foundation-encoder.md b/docs/adr/ADR-150-rf-foundation-encoder.md
index 8bee40b1..1ce47921 100644
--- a/docs/adr/ADR-150-rf-foundation-encoder.md
+++ b/docs/adr/ADR-150-rf-foundation-encoder.md
@@ -196,6 +196,13 @@ encoder/adapter split is validated empirically — a frozen shared trunk + tiny
 deployable path, and the foundation-encoder objective should be "make this adapter even smaller /
 need fewer calibration frames."
 
+**Calibration data requirement (measured, 3 seeds):** the 11 KB LoRA needs **~100–200 labeled
+samples/room** to reach ~72% (knee at ~50 → 70%); below ~20 samples it can't fit and may *hurt*
+(5 samples → 61% < zero-shot 64%). So the evidence-complete **calibration-service spec** is:
+ship shared base → collect **~100–200 labeled samples on-site** → fit a **~11 KB LoRA** →
+**~72% cross-subject** (SOTA-level). The encoder's research goal is now precisely posed: push that
+~100–200-sample requirement down and/or lift the >72% ceiling per fixed calibration budget.
+
 ## 4. Acceptance Test
 
 The encoder is accepted **only if it improves cross-subject torso-PCK@20 by ≥ 6 absolute points without reducing random-split torso-PCK@20 by more than 2 points** — on the same MM-Fi pipeline, one-command reproduction, with per-joint error tables. Results land as AetherArena witness rows (ADR-149), nothing published until reviewed.