[codex] add phase 5 evaluation baseline

docs/2026-05-07-session-handoff.md

@@ -0,0 +1,127 @@
+# 2026-05-07 Session Handoff
+
+## Current Focus
+
+Phase 5 is blocked on trustworthy audio-to-tab mapping before any more
+video-calibration or `lambda_vision` tuning.
+
+The key finding from today is that highres audio is strong at GuitarSet
+onset/pitch, but default pitch-to-string/fret decoding is weak. A learned
+GuitarSet train-split position prior improves Tab F1 substantially, but
+it should not become an unconditional production default yet.
+
+## What Changed Today
+
+- Added a raw GuitarSet audio-only evaluator:
+  - `tabvision/tabvision/eval/guitarset_audio.py`
+  - `tabvision/scripts/eval/guitarset_audio_eval.py`
+  - `tabvision/tests/unit/test_guitarset_audio_eval.py`
+- Added learned pitch-position prior support:
+  - `tabvision/tabvision/fusion/position_prior.py`
+  - `tabvision/tests/unit/test_position_prior.py`
+- Added a Modal L4 GPU runner for full highres eval:
+  - `tabvision/scripts/eval/guitarset_highres_modal.py`
+- Recorded the Phase 5 position-prior decision in `docs/DECISIONS.md`.
+
+The existing GuitarSet TFRecords were inspected and are insufficient for
+Tab F1 because they do not retain string/fret labels. Raw JAMS/WAV is the
+correct source for Tab F1.
+
+## Metric Snapshot
+
+Full GuitarSet validation split is player `05` (60 tracks).
+
+Modal L4 highres eval, full validation:
+
+| Run | Onset F1 | Pitch F1 | Tab F1 |
+| --- | ---: | ---: | ---: |
+| no position prior | `0.9218` | `0.9022` | `0.3878` |
+| GuitarSet train prior | `0.9218` | `0.9022` | `0.6104` |
+
+Delta from prior: `+22.26 pp` Tab F1, with onset/pitch unchanged.
+
+Per-track effect: 51/60 improved, 8/60 regressed, 1/60 unchanged. Mean
+track Tab F1 moved from `0.347` to `0.589`.
+
+Reports:
+
+- `tabvision-server/tools/outputs/guitarset_audio_eval-highres-validation-none-2026-05-07.md`
+- `tabvision-server/tools/outputs/guitarset_audio_eval-highres-validation-guitarset-train-2026-05-07.md`
+
+## Promotion Decision
+
+Do not make the GuitarSet train prior an unconditional production default
+yet.
+
+Recommended path:
+
+1. Promote the position prior as a versioned/configured production option.
+2. Create a checked-in prior artifact or generator so raw GuitarSet is not
+   required at runtime.
+3. Classify the 8 validation regressions, especially the SS/comp cases
+   where no-prior was already strong.
+4. Run the home-video Phase 5 benchmark with and without the prior.
+5. Make it the default only if the home-video benchmark has no regression
+   and the GuitarSet regressions are understood or accepted.
+
+## Modal State
+
+The Modal GPU path is now working.
+
+Useful commands:
+
+```bash
+tabvision-server/venv/bin/modal run tabvision/scripts/eval/guitarset_highres_modal.py --limit 1
+tabvision-server/venv/bin/modal run tabvision/scripts/eval/guitarset_highres_modal.py
+tabvision-server/venv/bin/modal run tabvision/scripts/eval/guitarset_highres_modal.py --position-prior none
+```
+
+The Modal volume `tabvision-guitarset` has been hydrated with 360 raw
+GuitarSet JAMS/WAV tracks from `taohu/guitarset`.
+
+## Verification At Handoff
+
+Last full fast suite:
+
+```bash
+cd tabvision
+.venv/bin/python -m pytest -q
+```
+
+Result: `249 passed, 9 skipped`.
+
+Focused checks after the eval/Modal changes:
+
+```bash
+cd tabvision
+.venv/bin/python -m pytest tests/unit/test_guitarset_audio_eval.py tests/unit/test_position_prior.py -q
+.venv/bin/python -m ruff check tabvision/eval/guitarset_audio.py tabvision/fusion/position_prior.py scripts/eval/guitarset_audio_eval.py scripts/eval/guitarset_highres_modal.py tests/unit/test_guitarset_audio_eval.py tests/unit/test_position_prior.py
+```
+
+Result: 11 focused tests passed; ruff passed.
+
+## Worktree Note
+
+The worktree is still dirty. Some dirty files predate this handoff. The
+most relevant new files from this session are:
+
+- `tabvision/tabvision/eval/guitarset_audio.py`
+- `tabvision/tabvision/fusion/position_prior.py`
+- `tabvision/scripts/eval/guitarset_audio_eval.py`
+- `tabvision/scripts/eval/guitarset_highres_modal.py`
+- `tabvision/tests/unit/test_guitarset_audio_eval.py`
+- `tabvision/tests/unit/test_position_prior.py`
+- `tabvision-server/tools/outputs/guitarset_audio_eval-highres-validation-none-2026-05-07.{md,csv}`
+- `tabvision-server/tools/outputs/guitarset_audio_eval-highres-validation-guitarset-train-2026-05-07.{md,csv}`
+
+There are also Phase 5 eval/fusion/video diagnostics and prior Phase 5
+implementation changes in the worktree. Do not assume every dirty file
+belongs to the GuitarSet work.
+
+## Next Best Step
+
+Build the production integration for the position prior as an explicit
+config option/artifact, then run home-video Phase 5 with prior on/off.
+
+Do not proceed to Phase 6 until Phase 5 is recorded and the user explicitly
+says `proceed`.

docs/DECISIONS.md

@@ -380,3 +380,45 @@ to validate; "the hand is around frets 3-6" is a stronger, more stable
 visual prior for resolving audio's same-pitch string/fret ambiguity. Keeping
 the signal as a prior lets audio and playability override it when the visual
 evidence is weak or wrong.
+
+---
+
+## 2026-05-07 — Phase 5 GuitarSet pitch-to-tab bottleneck
+
+**Phase:** 5 (audio-to-tab mapping)
+**Decision tree:** GuitarSet audio-only diagnostic — if pitch F1 is good
+but Tab F1 is bad, fix string/fret candidate selection before tuning video
+calibration or `lambda_vision`.
+**Branch taken:** **Add an optional learned pitch-position prior.** Raw
+GuitarSet JAMS provide held-out string/fret labels; the evaluator now learns
+`P(string,fret | pitch)` from train players and attaches it via the existing
+2D `AudioEvent.fret_prior` path before audio-only Viterbi decode.
+**Evidence:** On full validation, oracle gold-onset/gold-pitch events scored
+only `0.4335` Tab F1 with the default decoder, proving the mapping is bad even
+when audio extraction is perfect. On the first 10 validation tracks, about
+two-thirds of same-pitch events landed on the wrong adjacent string/fret. Most
+errors were low-fret equivalents such as G-string notes decoded on B or B-string
+notes decoded on high E. A GuitarSet train-split prior raised oracle
+full-validation Tab F1 to `0.6802`. On the 3-track highres smoke, Tab F1 moved
+from `0.3356` to `0.7260` while onset F1 (`0.9692`) and pitch F1 (`0.9555`)
+stayed unchanged.
+**Reasoning:** This confirms the immediate bottleneck is pitch-to-position
+ambiguity, not highres onset/pitch extraction and not Phase 5 vision weighting.
+The prior is optional for now; it does not change public fusion APIs or the
+default production decode until a full validation run and home-video check
+justify promoting it.
+
+**Follow-up evidence:** A Modal L4 full-validation highres run completed on
+2026-05-07. With no position prior: onset F1 `0.9218`, pitch F1 `0.9022`,
+Tab F1 `0.3878`. With the GuitarSet train-split prior: onset F1 `0.9218`,
+pitch F1 `0.9022`, Tab F1 `0.6104` (`+22.26 pp`). Per-track, 51/60 improved,
+8/60 regressed, and 1/60 was unchanged. Mean track Tab F1 moved from `0.347`
+to `0.589`.
+**Promotion decision:** **Do not make this an unconditional production default
+yet.** Promote the prior next as a versioned/configured production option, then
+make it the default only after (a) a checked-in prior artifact is available
+without requiring raw GuitarSet at runtime, (b) same-pitch regressions are
+classified and reduced or accepted, and (c) the home-video Phase 5 benchmark
+shows no regression. The full GuitarSet result is strong enough to justify the
+production integration path, but the 8 regressed validation clips make a silent
+global default premature.

tabvision-server/tools/outputs/guitarset_audio_eval-highres-validation-2026-05-07.csv

@@ -0,0 +1,4 @@
+track_id,backend,gold_notes,audio_events,decoded_events,onset_f1,pitch_f1,tab_f1,tab_tp,tab_fp,tab_fn
+05_BN1-129-Eb_comp,highres,148,150,150,0.979866,0.973154,0.221477,33,117,115
+05_BN1-129-Eb_solo,highres,44,46,46,0.933333,0.933333,0.111111,5,41,39
+05_BN1-147-Gb_comp,highres,96,100,100,0.969388,0.938776,0.612245,60,40,36

tabvision-server/tools/outputs/guitarset_audio_eval-highres-validation-2026-05-07.md

@@ -0,0 +1,22 @@
+# GuitarSet Audio Eval (highres)
+
+Split: **validation**
+Tracks: **3**
+Gold notes: **288**
+Audio events: **296**
+
+## Aggregate
+
+| Metric | Mean F1 | Micro P | Micro R | Micro F1 |
+| --- | ---: | ---: | ---: | ---: |
+| Onset | 0.961 | 0.956 | 0.983 | 0.969 |
+| Pitch | 0.948 | 0.943 | 0.969 | 0.955 |
+| Tab | 0.315 | 0.331 | 0.340 | 0.336 |
+
+## Per Track
+
+| Track | Gold | Audio | Decoded | Onset F1 | Pitch F1 | Tab F1 |
+| --- | ---: | ---: | ---: | ---: | ---: | ---: |
+| `05_BN1-129-Eb_comp` | 148 | 150 | 150 | 0.980 | 0.973 | 0.221 |
+| `05_BN1-129-Eb_solo` | 44 | 46 | 46 | 0.933 | 0.933 | 0.111 |
+| `05_BN1-147-Gb_comp` | 96 | 100 | 100 | 0.969 | 0.939 | 0.612 |

tabvision-server/tools/outputs/guitarset_audio_eval-highres-validation-guitarset-train-2026-05-07.csv

@@ -0,0 +1,61 @@
+track_id,backend,gold_notes,audio_events,decoded_events,onset_f1,pitch_f1,tab_f1,tab_tp,tab_fp,tab_fn
+05_BN1-129-Eb_comp,highres,148,150,150,0.979866,0.973154,0.932886,139,11,9
+05_BN1-129-Eb_solo,highres,44,46,46,0.933333,0.933333,0.377778,17,29,27
+05_BN1-147-Gb_comp,highres,96,100,100,0.969388,0.938776,0.571429,56,44,40
+05_BN1-147-Gb_solo,highres,27,32,32,0.847458,0.847458,0.203390,6,26,21
+05_BN2-131-B_comp,highres,202,203,203,0.987654,0.972840,0.809877,164,39,38
+05_BN2-131-B_solo,highres,69,76,76,0.937931,0.937931,0.593103,43,33,26
+05_BN2-166-Ab_comp,highres,173,162,162,0.955224,0.931343,0.859701,144,18,29
+05_BN2-166-Ab_solo,highres,67,69,69,0.941176,0.926471,0.691176,47,22,20
+05_BN3-119-G_comp,highres,180,178,178,0.977654,0.977654,0.720670,129,49,51
+05_BN3-119-G_solo,highres,62,64,64,0.873016,0.873016,0.619048,39,25,23
+05_BN3-154-E_comp,highres,185,180,180,0.958904,0.926027,0.695890,127,53,58
+05_BN3-154-E_solo,highres,56,59,59,0.921739,0.904348,0.260870,15,44,41
+05_Funk1-114-Ab_comp,highres,181,162,153,0.844311,0.838323,0.808383,135,18,46
+05_Funk1-114-Ab_solo,highres,60,56,56,0.844828,0.793103,0.637931,37,19,23
+05_Funk1-97-C_comp,highres,177,172,172,0.974212,0.951289,0.796562,139,33,38
+05_Funk1-97-C_solo,highres,58,57,57,0.921739,0.921739,0.330435,19,38,39
+05_Funk2-108-Eb_comp,highres,245,271,260,0.843564,0.807921,0.673267,170,90,75
+05_Funk2-108-Eb_solo,highres,61,61,61,0.983607,0.983607,0.491803,30,31,31
+05_Funk2-119-G_comp,highres,182,192,184,0.907104,0.890710,0.841530,154,30,28
+05_Funk2-119-G_solo,highres,99,96,96,0.943590,0.923077,0.256410,25,71,74
+05_Funk3-112-C#_comp,highres,212,206,206,0.928230,0.913876,0.837321,175,31,37
+05_Funk3-112-C#_solo,highres,70,64,64,0.955224,0.955224,0.686567,46,18,24
+05_Funk3-98-A_comp,highres,184,182,162,0.901734,0.895954,0.815029,141,21,43
+05_Funk3-98-A_solo,highres,65,65,65,0.969231,0.969231,0.538462,35,30,30
+05_Jazz1-130-D_comp,highres,141,118,118,0.880309,0.849421,0.687259,89,29,52
+05_Jazz1-130-D_solo,highres,58,61,61,0.924370,0.890756,0.554622,33,28,25
+05_Jazz1-200-B_comp,highres,175,168,163,0.852071,0.781065,0.491124,83,80,92
+05_Jazz1-200-B_solo,highres,50,49,49,0.969697,0.969697,0.464646,23,26,27
+05_Jazz2-110-Bb_comp,highres,252,252,242,0.910931,0.890688,0.781377,193,49,59
+05_Jazz2-110-Bb_solo,highres,103,97,97,0.940000,0.940000,0.700000,70,27,33
+05_Jazz2-187-F#_comp,highres,247,226,219,0.862661,0.836910,0.669528,156,63,91
+05_Jazz2-187-F#_solo,highres,61,60,60,0.975207,0.975207,0.280992,17,43,44
+05_Jazz3-137-Eb_comp,highres,235,231,227,0.857143,0.800866,0.683983,158,69,77
+05_Jazz3-137-Eb_solo,highres,77,76,76,0.967320,0.967320,0.496732,38,38,39
+05_Jazz3-150-C_comp,highres,227,226,219,0.910314,0.874439,0.291480,65,154,162
+05_Jazz3-150-C_solo,highres,69,67,67,0.955882,0.955882,0.764706,52,15,17
+05_Rock1-130-A_comp,highres,312,349,349,0.883510,0.877458,0.166415,55,294,257
+05_Rock1-130-A_solo,highres,60,62,62,0.950820,0.950820,0.606557,37,25,23
+05_Rock1-90-C#_comp,highres,277,276,276,0.936709,0.871609,0.867993,240,36,37
+05_Rock1-90-C#_solo,highres,99,106,106,0.926829,0.907317,0.809756,83,23,16
+05_Rock2-142-D_comp,highres,460,450,450,0.890110,0.841758,0.336264,153,297,307
+05_Rock2-142-D_solo,highres,86,86,86,0.906977,0.906977,0.290698,25,61,61
+05_Rock2-85-F_comp,highres,325,296,296,0.856683,0.821256,0.615137,191,105,134
+05_Rock2-85-F_solo,highres,125,124,124,0.955823,0.939759,0.698795,87,37,38
+05_Rock3-117-Bb_comp,highres,398,401,401,0.878598,0.868586,0.403004,161,240,237
+05_Rock3-117-Bb_solo,highres,70,65,65,0.962963,0.962963,0.518519,35,30,35
+05_Rock3-148-C_comp,highres,249,286,286,0.927103,0.923364,0.818692,219,67,30
+05_Rock3-148-C_solo,highres,61,60,60,0.991736,0.991736,0.495868,30,30,31
+05_SS1-100-C#_comp,highres,102,95,95,0.964467,0.964467,0.395939,39,56,63
+05_SS1-100-C#_solo,highres,60,64,64,0.887097,0.870968,0.774194,48,16,12
+05_SS1-68-E_comp,highres,143,148,148,0.982818,0.975945,0.783505,114,34,29
+05_SS1-68-E_solo,highres,78,74,74,0.894737,0.881579,0.328947,25,49,53
+05_SS2-107-Ab_comp,highres,222,227,227,0.971047,0.953229,0.726058,163,64,59
+05_SS2-107-Ab_solo,highres,87,85,85,0.941860,0.941860,0.546512,47,38,40
+05_SS2-88-F_comp,highres,196,190,190,0.963731,0.943005,0.621762,120,70,76
+05_SS2-88-F_solo,highres,93,97,97,0.978947,0.968421,0.589474,56,41,37
+05_SS3-84-Bb_comp,highres,211,215,215,0.985915,0.985915,0.671362,143,72,68
+05_SS3-84-Bb_solo,highres,111,119,119,0.947826,0.947826,0.330435,38,81,73
+05_SS3-98-C_comp,highres,199,187,187,0.937824,0.932642,0.735751,142,45,57
+05_SS3-98-C_solo,highres,93,94,94,0.973262,0.973262,0.288770,27,67,66

tabvision-server/tools/outputs/guitarset_audio_eval-highres-validation-guitarset-train-2026-05-07.md

@@ -0,0 +1,80 @@
+# GuitarSet Audio Eval (highres)
+
+Split: **validation**
+Position prior: **guitarset-train**
+Tracks: **60**
+Gold notes: **8715**
+Audio events: **8690**
+
+## Aggregate
+
+| Metric | Mean F1 | Micro P | Micro R | Micro F1 |
+| --- | ---: | ---: | ---: | ---: |
+| Onset | 0.930 | 0.928 | 0.916 | 0.922 |
+| Pitch | 0.915 | 0.908 | 0.897 | 0.902 |
+| Tab | 0.589 | 0.614 | 0.607 | 0.610 |
+
+## Per Track
+
+| Track | Gold | Audio | Decoded | Onset F1 | Pitch F1 | Tab F1 |
+| --- | ---: | ---: | ---: | ---: | ---: | ---: |
+| `05_BN1-129-Eb_comp` | 148 | 150 | 150 | 0.980 | 0.973 | 0.933 |
+| `05_BN1-129-Eb_solo` | 44 | 46 | 46 | 0.933 | 0.933 | 0.378 |
+| `05_BN1-147-Gb_comp` | 96 | 100 | 100 | 0.969 | 0.939 | 0.571 |
+| `05_BN1-147-Gb_solo` | 27 | 32 | 32 | 0.847 | 0.847 | 0.203 |
+| `05_BN2-131-B_comp` | 202 | 203 | 203 | 0.988 | 0.973 | 0.810 |
+| `05_BN2-131-B_solo` | 69 | 76 | 76 | 0.938 | 0.938 | 0.593 |
+| `05_BN2-166-Ab_comp` | 173 | 162 | 162 | 0.955 | 0.931 | 0.860 |
+| `05_BN2-166-Ab_solo` | 67 | 69 | 69 | 0.941 | 0.926 | 0.691 |
+| `05_BN3-119-G_comp` | 180 | 178 | 178 | 0.978 | 0.978 | 0.721 |
+| `05_BN3-119-G_solo` | 62 | 64 | 64 | 0.873 | 0.873 | 0.619 |
+| `05_BN3-154-E_comp` | 185 | 180 | 180 | 0.959 | 0.926 | 0.696 |
+| `05_BN3-154-E_solo` | 56 | 59 | 59 | 0.922 | 0.904 | 0.261 |
+| `05_Funk1-114-Ab_comp` | 181 | 162 | 153 | 0.844 | 0.838 | 0.808 |
+| `05_Funk1-114-Ab_solo` | 60 | 56 | 56 | 0.845 | 0.793 | 0.638 |
+| `05_Funk1-97-C_comp` | 177 | 172 | 172 | 0.974 | 0.951 | 0.797 |
+| `05_Funk1-97-C_solo` | 58 | 57 | 57 | 0.922 | 0.922 | 0.330 |
+| `05_Funk2-108-Eb_comp` | 245 | 271 | 260 | 0.844 | 0.808 | 0.673 |
+| `05_Funk2-108-Eb_solo` | 61 | 61 | 61 | 0.984 | 0.984 | 0.492 |
+| `05_Funk2-119-G_comp` | 182 | 192 | 184 | 0.907 | 0.891 | 0.842 |
+| `05_Funk2-119-G_solo` | 99 | 96 | 96 | 0.944 | 0.923 | 0.256 |
+| `05_Funk3-112-C#_comp` | 212 | 206 | 206 | 0.928 | 0.914 | 0.837 |
+| `05_Funk3-112-C#_solo` | 70 | 64 | 64 | 0.955 | 0.955 | 0.687 |
+| `05_Funk3-98-A_comp` | 184 | 182 | 162 | 0.902 | 0.896 | 0.815 |
+| `05_Funk3-98-A_solo` | 65 | 65 | 65 | 0.969 | 0.969 | 0.538 |
+| `05_Jazz1-130-D_comp` | 141 | 118 | 118 | 0.880 | 0.849 | 0.687 |
+| `05_Jazz1-130-D_solo` | 58 | 61 | 61 | 0.924 | 0.891 | 0.555 |
+| `05_Jazz1-200-B_comp` | 175 | 168 | 163 | 0.852 | 0.781 | 0.491 |
+| `05_Jazz1-200-B_solo` | 50 | 49 | 49 | 0.970 | 0.970 | 0.465 |
+| `05_Jazz2-110-Bb_comp` | 252 | 252 | 242 | 0.911 | 0.891 | 0.781 |
+| `05_Jazz2-110-Bb_solo` | 103 | 97 | 97 | 0.940 | 0.940 | 0.700 |
+| `05_Jazz2-187-F#_comp` | 247 | 226 | 219 | 0.863 | 0.837 | 0.670 |
+| `05_Jazz2-187-F#_solo` | 61 | 60 | 60 | 0.975 | 0.975 | 0.281 |
+| `05_Jazz3-137-Eb_comp` | 235 | 231 | 227 | 0.857 | 0.801 | 0.684 |
+| `05_Jazz3-137-Eb_solo` | 77 | 76 | 76 | 0.967 | 0.967 | 0.497 |
+| `05_Jazz3-150-C_comp` | 227 | 226 | 219 | 0.910 | 0.874 | 0.291 |
+| `05_Jazz3-150-C_solo` | 69 | 67 | 67 | 0.956 | 0.956 | 0.765 |
+| `05_Rock1-130-A_comp` | 312 | 349 | 349 | 0.884 | 0.877 | 0.166 |
+| `05_Rock1-130-A_solo` | 60 | 62 | 62 | 0.951 | 0.951 | 0.607 |
+| `05_Rock1-90-C#_comp` | 277 | 276 | 276 | 0.937 | 0.872 | 0.868 |
+| `05_Rock1-90-C#_solo` | 99 | 106 | 106 | 0.927 | 0.907 | 0.810 |
+| `05_Rock2-142-D_comp` | 460 | 450 | 450 | 0.890 | 0.842 | 0.336 |
+| `05_Rock2-142-D_solo` | 86 | 86 | 86 | 0.907 | 0.907 | 0.291 |
+| `05_Rock2-85-F_comp` | 325 | 296 | 296 | 0.857 | 0.821 | 0.615 |
+| `05_Rock2-85-F_solo` | 125 | 124 | 124 | 0.956 | 0.940 | 0.699 |
+| `05_Rock3-117-Bb_comp` | 398 | 401 | 401 | 0.879 | 0.869 | 0.403 |
+| `05_Rock3-117-Bb_solo` | 70 | 65 | 65 | 0.963 | 0.963 | 0.519 |
+| `05_Rock3-148-C_comp` | 249 | 286 | 286 | 0.927 | 0.923 | 0.819 |
+| `05_Rock3-148-C_solo` | 61 | 60 | 60 | 0.992 | 0.992 | 0.496 |
+| `05_SS1-100-C#_comp` | 102 | 95 | 95 | 0.964 | 0.964 | 0.396 |
+| `05_SS1-100-C#_solo` | 60 | 64 | 64 | 0.887 | 0.871 | 0.774 |
+| `05_SS1-68-E_comp` | 143 | 148 | 148 | 0.983 | 0.976 | 0.784 |
+| `05_SS1-68-E_solo` | 78 | 74 | 74 | 0.895 | 0.882 | 0.329 |
+| `05_SS2-107-Ab_comp` | 222 | 227 | 227 | 0.971 | 0.953 | 0.726 |
+| `05_SS2-107-Ab_solo` | 87 | 85 | 85 | 0.942 | 0.942 | 0.547 |
+| `05_SS2-88-F_comp` | 196 | 190 | 190 | 0.964 | 0.943 | 0.622 |
+| `05_SS2-88-F_solo` | 93 | 97 | 97 | 0.979 | 0.968 | 0.589 |
+| `05_SS3-84-Bb_comp` | 211 | 215 | 215 | 0.986 | 0.986 | 0.671 |
+| `05_SS3-84-Bb_solo` | 111 | 119 | 119 | 0.948 | 0.948 | 0.330 |
+| `05_SS3-98-C_comp` | 199 | 187 | 187 | 0.938 | 0.933 | 0.736 |
+| `05_SS3-98-C_solo` | 93 | 94 | 94 | 0.973 | 0.973 | 0.289 |