Add drop-test protocol, Edison synthesis, and first-data analysis

.gitignore

@@ -41,3 +41,6 @@ Thumbs.db
 
 # Edison upload bundles (regenerated by submit scripts)
 edison-trajectories/*/bundle/
+
+__pycache__/
+*.pyc

CLAUDE.md

@@ -0,0 +1,41 @@
+# Repo instructions for Claude
+
+## Drop-test analysis: standing conventions
+
+### T-drift watch (standing instruction, PR #86 · 2026-08-24)
+
+In **every** drop-test analysis — in this PR or any other — screen each
+session's within-session transmissibility series (T = TOP/CH5, CFC-180,
+stabilized drops) for drift, report the numbers, and **notify prominently**
+if a session breaks the historical envelope. Requested by @me-madsen after
+the `r2d2c2` event; the round-2 batch is analyzed in a separate PR, so this
+must not be forgotten between threads.
+
+- Historical 1/2 in PU mat envelope (17 sessions, ~1,300 stabilized drops,
+  08-10 → 08-24): |OLS slope| ≤ 0.051 %/drop, |first-5 vs last-5| ≤ 2.2 %.
+  Per-session table: `data/drop-tests/r2d2-checkin/figures/t_drift_history.json`;
+  figure: `.../figures/04_t_drift_history.png`.
+- Flag thresholds: |slope| > 0.06 %/drop (slope_p < 0.01) or |end-to-end| > 2.5 %.
+- Reference event: `r2d2c2` (08-24) stepped +3.5 % (1.023 → ~1.059 plateau,
+  from ~drop 7, plateaued ~drop 14). Output-side — top-vertex peak rose while
+  the CH5 input fell — lead suspect wax/mount coupling re-seat. `r2d2c1`,
+  minutes earlier on the same mat, was dead flat.
+- When flagged: the session's T mean is drift-contaminated. Do not hand it to
+  the BO — use a plateau-only average or ask for a mount re-seat + re-run.
+  Attribute the drift (output-side = mount; common-mode = rig/mat, cancels in T).
+- The campaign pipeline (`scripts/analysis/drop_test_campaign_analysis.py`,
+  `t_drift_watch()`) automates this: `t_drift_watch` block in
+  `campaign_metrics.json`, `t_drift_flag`/slope/e2e columns in
+  `campaign_summary.csv`, console warnings. Reuse it (or its convention) in
+  any new analysis script, and state the watch outcome — flagged or clean —
+  in every findings writeup.
+
+### Other conventions (pointers, not duplicates)
+
+- Per-capture pipeline + capture settings: `analyze_capture` in
+  `scripts/analysis/drop_test_abc123_blind_analysis.py`; use the **tail
+  baseline** for sessions from 08-17 on (mat contact precedes the 2 ms
+  pre-trigger window).
+- Δv rig-health bars, warm-up discard, and dataset index:
+  `docs/drop-test-speed-decay-analysis.md`, `data/drop-tests/README.md`.
+- T3 specimen IDs are case-insensitive; treat as lowercase.

data/drop-tests/60in-5felts-validation/README.md

@@ -0,0 +1,59 @@
+# 60 in / 5 felt validation — derived analysis outputs
+
+Derived dataset (no raw data of its own) for the validation campaigns of the
+felt-sheet sweep's recommended operating point (**60 in drop height + 5 felt
+sheets**), posted by @ctrhjk on PR #86:
+
+| specimen | raw data | captures | session |
+|---|---|--:|---|
+| `7xadt6` | [`../7xadt6 _60in_5felts folder/`](../7xadt6%20_60in_5felts%20folder/) — `Marcus_{1..4}.zip` | 100 | "7xadt6 60in+5felts", 2026-07-20 19:51–20:59 |
+| `9GMQYQ` | [`../9GMQYQ_60in_5felts/`](../9GMQYQ_60in_5felts/) — `jin_{1..4}.zip` | 101 | "9GMQYQ 60in+5felts", 2026-07-20 21:12–22:32 |
+
+The raw TP4 exports (200 ms / 125 kHz) stay inside the committed zips; the
+analysis script reads them directly.
+
+**Stack composition:** per @me-madsen (PR #86), every "5 felt" session —
+these two included — physically used **4 felt sheets + 1 cardboard** (the
+lab owns only four felt sheets). Composition was constant across sessions,
+so the analysis conclusions are unaffected; the names keep the label the
+data was posted under.
+
+## Setup / channel map
+
+Rig unchanged from the felt-sheet sweep and `500drops-nobot` runs; bottom
+tri-axis removed. Per @ctrhjk's channel note on PR #86:
+
+| channel | station | full scale |
+|---|---|--:|
+| CH2 | top-vertex key-seat tri-axis, **X** ("TOP" output) | 14,492.8 G |
+| CH3 | top-vertex key-seat tri-axis, **Y** | 14,992.5 G |
+| CH4 | top-vertex key-seat tri-axis, **Z** | 13,624.0 G |
+| CH5 | base-plate single-axis (input, **trigger @ 300 G**) | 9,442.9 G |
+
+## Analysis
+
+Script: `scripts/analysis/drop_test_60in_5felts_analysis.py` (emits
+`figures/60in_5felts_metrics.json`). Writeup:
+[`docs/drop-test-60in-5felts-analysis.md`](../../../docs/drop-test-60in-5felts-analysis.md)
+— capture health (201/201 clean), the burn-in scan + stabilized-phase OLS
+drift regressions, the felt-wear saturation story (CH5 raw spike 2.1 →
+6.5 kG over the evening), specimen discrimination, and the keep-60-in/5-felt
++ manage-felt-as-consumable recommendation.
+
+| figure | content |
+|---|---|
+| `figures/01_full_series.png` | per-capture raw CH5 / TOP peaks + FS/3 line, both specimens |
+| `figures/02_stabilized_ols.png` | stabilized-phase OLS fits: TOP and T = TOP/CH5 per specimen |
+| `figures/03_saturation.png` | CH5 / CH4 raw peak as % of full scale across the evening |
+| `figures/04_specimen_comparison.png` | stabilized TOP and T boxplots, Welch p, Cohen's d |
+
+## Slow-mo video
+
+[`video/`](video/) — @ctrhjk's slow-motion videos of both campaigns
+(Sony RX100 IV @ 960 fps, 2026-07-20; committed as
+`{7xadt6,9GMQYQ}_slomo.mp4` by @sgbaird) plus the full frame-by-frame
+kinematics pass (`drop_test_60in_5felts_video_analysis.py`): pulse ≤ 2
+capture frames ≈ 1–2 ms (corroborates the DAQ's ~1.6 ms), anti-rebound
+brake catch ~86–89 ms after impact (no secondary hits), top-vertex
+snap-back ~0.7× impact speed, specimens intact throughout. Writeup: §7 of
+the analysis doc.