docs: point at the pipeline repo; record the FoTA schema fix

chore: drop stale scripts/ copy; pipeline now lives on GitHub

fix: widen fota_unlabeled from 26 to 30 columns (schema parity across configs)

fix: widen fota_labeled from 26 to 30 columns (schema parity across configs)

SOURCES.md: clarify YCB-Sight uses the original GelSight sensor (not Mini) — primary reason for exclusion, with viral-copyleft license as secondary

README: full rewrite — 513 → 198 lines (~1-2 min read), all stale numbers refreshed (fota_labeled 29,494→26,394, threedcal 36,270→6,924, gelslam 89,612→114,019, real_tactile_mnist 56,723→30,956, etc); single composition table + 4-bullet pipeline summary + compact citation list; kept SOURCES.md as the deep-dive doc with stale-number fixes.

UniT: re-ingest with contact filter (11,340 → 387 kept); fix gel_variant to markered (gel has visible dot pattern). Update docs: remove TVL (was DIGIT not Mini), add to Investigated. Grand total: 852,915 frames.

remove tvl: uses DIGIT sensor, not GelSight Mini (verified at https://tactile-vlm.github.io: 'Tactile data are collected using a DIGIT sensor: a compact, open-source tactile sensor that provides observations in the form of RGB images.'). Parquet preserved locally for potential future digit-pretrain repo. Bumps real total from 745,958 to 536,163 frames.

SOURCES.md: remove duplicate Touch-and-Go bullet (stale 'Not yet ingested' line contradicted the new 'wrong GelSight model' explanation above it)

remove touchandgo: uses ORIGINAL GelSight (white illumination, 640x480 flat-channel R~G~B~137) not Mini. Verified via channel-mean signature mismatch with all other Mini sources (B>G>R with 30+ grey-level spread). Document reason in SOURCES.md 'Investigated but not included'. Bumps real total from 748,024 to 745,958 frames.

docs: add Touch-and-Go (## 12) + TVL (## 13) sections to SOURCES.md; add YAML configs + composition rows for both new sources; tagline updated to ~1.1M frames

add tvl via unified pipeline

add touchandgo via unified pipeline

v9: fix channel order. fota_unlabeled + unit swapped R<->B (were BGR-stored). Refresh sample grids + RGB-vs-BGR comparison + channel_order_diagnosis.json

docs: SOURCES.md add ## 10 UniT + ## 11 TacQuad sections, update all stale row counts (fota_unlabeled, RTM, sims, feelanyforce, etc), add Channel-order normalization section; README: update header tagline frame count

README: update NC extension blurb to reflect Sparsh-only / 66K frames

add RGB-vs-BGR comparison grids for each subset (side-by-side: top=RGB as stored, bottom=BGR R↔B swap)

v8 final: RTM video re-extract (30,956 peak-contact frames); add Balance metrics section (H̃=0.95 domain, 8,459 unique objects); refresh composition/pies charts + RTM sample grid; publish balance_report.json

assets cleanup: add samples_40 markered/markerless fota grids; delete 39 stale debug + samples_100 PNGs no longer referenced

sim rebalance from raw upstream: sim_starstruck → 166K (150K train + 16K test, stride=3); sim_tactile_mnist → 151K (102K train + 49K test, stride=2); refreshed sample grids

add UniT (11,340 frames) + TacQuad full (12,195 frames across 181 objects); deprecate tacquad_mini

refresh samples_40_fota_unlabeled.png (200K after loose dedupe)

refresh samples_40 grids (RTM+sims at I=15, others unchanged)

v7: I_min=12 (real) + I_min=10 (sim) reprocess of RTM and sims; add pixel_value_distribution chart; fix sim_tactile_mnist test routing; refresh all sample grids and composition charts.

Re-sample 4 grids with intensity≥12 contact filter (RTM/sim/feelanyforce)

Refresh sample grids for RTM / sim x2 / feelanyforce (seed=20260518)

Collapse fota_unlabeled to train-only (66,761 frames; was 516,523). Refresh assets/. Delete obsolete 8-shard and val-* layouts.

RTM v7 samples_40

RTM v7: K=0.3 per-frame Bernoulli + A=40, I=10 → 200K (hit budget)

cleanup: remove stale assets/samples_100_fota_labeled_markerless.png

cleanup: remove stale assets/samples_100_fota_labeled_markered.png

cleanup: remove stale assets/samples_100_sim_starstruck.png

cleanup: remove stale assets/samples_100_sim_tactile_mnist.png

cleanup: remove stale assets/samples_100_real_tactile_mnist.png

cleanup: remove stale assets/samples_100_tactile_tracking.png

cleanup: remove stale assets/samples_100_gelslam.png

cleanup: remove stale assets/samples_100_feelanyforce.png

cleanup: remove stale assets/samples_100_feats.png

cleanup: remove stale assets/samples_100_threedcal.png

cleanup: remove stale assets/samples_100_fota_unlabeled.png

cleanup: remove stale assets/samples_100_fota_labeled.png

README: update totals to ~700K (post-FoTA-dedupe)

Refresh summary_pies.png with post-dedupe FoTA counts

Refresh resolution_distribution.png with post-dedupe FoTA counts

Refresh composition.png with post-dedupe FoTA counts

Refresh samples_40_fota_unlabeled.png after dedupe (66.7K)

fota_unlabeled: dedupe (phash Hamming≤4, lookback 30) → 470K→67K (86% dedupe rate) + ready for 200K cap

Refresh threedcal_coverage.png: include tacquad_mini, log y-axis on resolution, current 1.12M counts

Refresh force_distribution.png: include tacquad_mini, log y-axis on resolution, current 1.12M counts

Refresh summary_pies.png: include tacquad_mini, log y-axis on resolution, current 1.12M counts