Publish the model card as README so the limitations render on the repo page

README.md

@@ -0,0 +1,617 @@
+---
+license: mit
+tags:
+  - taiko
+  - rhythm-game
+  - chart-generation
+  - music
+  - audio-to-symbolic
+---
+
+# BarScript (experimental preview) — C1.3, denominator augmentation, seed 1234
+
+Audio log-mel + a bar grid → a Taiko no Tatsujin chart, decoded under a
+finite-state grammar over bar-scoped plan, skeleton and realization stages.
+
+**BarScript** is the release name of this charting model; it is also the name of
+the bar-scoped multi-stage token encoding it decodes into. It is a **preview
+under evaluation**, published so it can be tried, not as a finished model: the
+champion comparison below is `inconclusive`, and §2 lists measured defects — one
+of which fires a pre-registered kill condition. The seven published SoftChart
+1.x models are a separate, unaffected line.
+
+| | |
+|---|---|
+| source checkpoint | `runs/sc2_c1_v13_s1234/best.pt` |
+| sha256 | `44980bf64b1611ea73c1433c41adceff7596cfeece30dfaad5d228eae4a14e6e` |
+| seed / step | 1234 / 96 000 of 100 000 |
+| best validation CE | 0.5769365892960475 |
+| parameters | 9 107 291 |
+| shipped weights | bfloat16 safetensors, 18.2 MB (fp32 checkpoint preserved in the run dir) |
+| training data | `JacobLinCool/taiko-1000-parsed-clean`, revision `b72da4616d643018e81f372cea06ce51349285e0` |
+| label spec | `barscript_labels_v1`, frozen, shipped as `spec.json` (sha1 `cd876768…`) |
+| training flags vs C1 | `--cal-axes big span density` · `--balance-exclude-big` · `--denom-augment 0.5 --denom-augment-mode lattice` |
+| external pretraining | none |
+
+**This card is written for THIS checkpoint.** Every number below was measured on
+C1.3's own generations under the serving code this package ships
+(`generate.py` md5 `73b6e49b…`, `min_onset_gap_sec = 0.020`). Nothing is
+inherited from the C1 card. Where an item on the C1 card has no C1.3
+measurement, it is marked **not measured on this arm** rather than carried over.
+
+**Read the KNOWN LIMITATIONS section before using any number above it.** This
+model has real, measured defects, one of which fires a pre-registered kill
+condition, and its difficulty ladder is roughly 25 coverage points less nested
+than an authored one.
+
+---
+
+## 0. Status: the champion comparison is `inconclusive`
+
+`scripts/sc2_gates.py champion`, three pairings on one serving code base
+(`experiments/sc2_eval/FINAL_CHAMPION.md` §1):
+
+| pairing | `champion_verdict` | reason code |
+|---|---|---|
+| C1.2 vs C1 | `inconclusive` | `no_seed_replicate` |
+| **C1.3 vs C1** | **`inconclusive`** | `no_seed_replicate` |
+| C1.3 vs C1.2 | `inconclusive` | `no_seed_replicate` |
+
+`MIN_SEEDS_FOR_VERDICT` is 2 in **both** arms and C1.3 has one seed. The rule
+fires before any metric is read. **No arm has been declared champion.**
+
+What the clause ledger says, which is one-directional: C1.3 **passes** the
+required §4 clause `plan_token_calibration`, which **C1 fails on both its
+seeds**. No clause is passed by C1 and failed by C1.3. The two `a2` span clauses
+fail on **all four arms** and belong to the campaign, not to this checkpoint.
+
+The shipping rationale, the two named compromises and the decision rule for
+choosing a different checkpoint are in
+`experiments/sc2_eval/SHIP_DECISION.md`.
+
+Gate items on the 71-chart hard+oni population, this arm against the C1 seed-1234
+arm under identical code:
+
+| | C1 | **C1.3** |
+|---|---:|---:|
+| fail / pass / na | 9 / 10 / 6 | **8 / 11 / 6** |
+| newly passing | — | `control_own_axis_density`, `control_named_regression_density_big_count` |
+| newly failing | — | `a2_span_placement_fit` (knife-edge, §2.6) |
+
+---
+
+## 1. What it does
+
+* **Input** — 128-bin log-mel at 22 050 Hz (`preprocessor_config.json` freezes
+  the exact decode/STFT/mel contract), plus a bar grid with explicit measure
+  edges, exact rational meters and per-bar lattice denominators.
+* **Conditions** — course, authored level, density bucket, and the split-axis
+  knobs `big_rate` / `span_rate` / `stream` / `sync`.
+* **Output** — a BarScript token stream decoded under `BarscriptFSM`, exported
+  as TJA on the 96-slot lattice.
+* **Family decoding** — `generate_chart_family` decodes a song's courses as one
+  nested ladder, hardest first, each easier course biased toward its harder
+  sibling's onsets. **Off by default**; the campaign's headline population was
+  decoded independently (§2.1).
+
+Capabilities this checkpoint carries (`config.json:capabilities`): `aux`,
+`beat_head` (hi-res), `hierarchical_ctx`, `axis_knobs`, `sync_token`,
+`stage_emb`, `span_duration_head`, `tempo_head`, `rich_section_stats`. It
+carries **no** `style`, `sibling`, `ctx`, `plan`-prefix, `slot`, `dual`,
+`align`, `mask_infill`, `func_time`, `global_ctx` or `complexity` conditioning;
+requests on those axes reach nothing.
+
+### What works
+
+Evidence marks: **●** n ≥ 71 charts with a measured C1 seed band; **◐** n = 19–71
+charts or 8–12 sweep cells, one seed; **○** n < 19 or a CI covering zero.
+
+| axis | this checkpoint | reference | mark |
+|---|---|---|---|
+| **Density knob** — median within-song ρ **0.850**, sign p **0.0078**, 8/8 cells positive, **+1.413 nps** first→last, 12 % monotone adjacency | C1 moves the chart +0.024 / +0.151 nps over the same range and **fails** the gate; C1 seed-to-seed Δρ 0.003 (p 1.000) | 8 songs × 5 buckets, oni | ◐ |
+| **`big_rate` knob** — ρ **1.000**, p 0.00049, 91.7 % monotone, and **`calibrated: true`** — mean absolute bucket error **0.533**, within ±1 bucket **93.3 %** | first axis in the whole campaign to clear `calibrated`; C1 0.533 vs 1.200 / 0.967 | 12 songs × 5 buckets, oni | ◐ |
+| **`span_rate` knob** — ρ 0.810, p 0.0117, `calibrated: true`, mean abs err 1.483; and unlike C1 it **no longer drags hit density with it** | C1's span knob flags `onset_nps` / `hit_nps` / `n_onsets` as interference; C1.3 flags only the definitional `span_count` | 12 songs × 5 buckets, oni | ◐ |
+| **Span budget** — 739 spans against the authored 762 (**0.970×**); `span_rate_gen_per_min` gap to authored **0.049** against a C1 seed band of 0.198 | C1 1.160–1.244×, C1.2 1.259× | 97 charts | ● |
+| **FSM guarantees** — **0 unclosed, 0 orphan ends, 0 swallowed hits** over 739 generated spans | holds on all four arms | 97 charts | ● |
+| **Placement does not degrade** — precision 0.6477 (C1 band 0.6480–0.6502), median \|offset\| **0 ms**, `exact_slot_lift_over_null` **1.1755** (above the C1 band), long-song drift tests significant **2** vs C1's 6 / 5 | see §2.3 for why apparent recall drops | 71 charts | ● |
+| **Accent over-emission reduced furthest** — pooled hard+oni big share **0.0873** against C1's 0.1082 / 0.1027 (3.3 seed bands) and an authored 0.0589 | still outside the pre-registered stop window 0.045–0.075 | 71 charts | ● |
+| **Cheaper** — 1 618.6 tokens/chart, 19.22 tokens/bar, `max_window_tokens_p99` **449** | C1 1 731.2 / 20.51 / 515 | 71 charts | ● |
+| **Deployment grid repaired** — on a forced `/16` BPM grid, notes ÷ authored **1.130 / 1.112 / 1.067 / 1.027** and onset F1 **0.333 / 0.374 / 0.466 / 0.580** | C1 1.688 / 1.461 / 1.297 / 1.210 and F1 0.318 / 0.376 / 0.477 / 0.599; C1.3 beats C1.2 on **12 of 12** deployable cells | 13 songs × 4 courses, one seed | ◐ |
+| **Empty-bar declaration restored** — `/16` empty bars 12.82 / 8.60 / 4.55 / 3.14 % against C1.2's 5.96 / 3.23 / 1.08 / 0.33 %, sign-significant on 3 of 4 courses | authored 15.69 / 12.30 / 9.08 / 6.27 % | 13 songs | ◐ |
+
+---
+
+## 2. KNOWN LIMITATIONS
+
+Nothing in this section is softened. Where a limitation is invisible to the gate
+suite, that is said.
+
+### 2.1 The difficulty ladder is 25–32 coverage points less nested than authored
+
+Coverage = the fraction of the easier chart's notes that have a note in the
+harder chart within tolerance. **Two decode modes, two different numbers, and
+both belong on the record.**
+
+| adjacent pair | authored charts | **independent decode** (package default) | **family decode β = 2** (authored grid) | **family decode β = 2** (`/16` deploy grid) |
+|---|---:|---:|---:|---:|
+| easy → normal | 0.970 | **0.658** | 0.730 | 0.726 |
+| normal → hard | 0.979 | **0.667** | 0.797 | 0.811 |
+| hard → oni | 0.985 | **0.660** | 0.822 | 0.834 |
+
+Independent-decode figures: `FINAL_CHAMPION.md` §8.1, n = 13 / 13 / 35 songs, one
+seed, same serving code as this package (●/◐). Family-decode figures:
+`C13_PREDICTION.md` §3.3, n = 13 songs, one seed (◐).
+
+Five things make this worse than the headline:
+
+* **`nesting_coverage` and `hand_agreement` FAIL the family gate on every arm
+  measured**, C1 included. `difficulty_monotonicity` passes on all of them.
+* **C1.3 is slightly worse than C1 here.** C1's independent-decode coverage on
+  the same population is 0.690 / 0.704 / 0.737; C1.3 is −0.032 / −0.037 / −0.077
+  against C1 seed bands of 0.010 / 0.021 / 0.025. The move is small against the
+  ~0.30 gap to authored that every arm shares, but it is in the wrong direction.
+* **Hand agreement at coinciding hits is near chance on hard→oni**: C1.3 0.539
+  against a marginal-chance null of 0.500 and an authored 0.753. When two
+  generated courses agree that a note belongs somewhere, which drum they pick is
+  near-independent across courses.
+* **The `/16` figure overstates the model's nesting.** The same checkpoint on a
+  `/96` grid drops to 0.649 / 0.730 / 0.752, worse on 11–12 of 13 songs
+  (p = .003 / .022 / .022). A coarse lattice manufactures agreement by leaving
+  few places to disagree (`C13_PREDICTION.md` §5.2).
+* **`family_bias = 2.0` / `family_hand_bias = 1.5` are PROVISIONAL** — described
+  in `generate.py` as logit offsets on a decoder never trained at that setting,
+  and never calibrated against the authored target.
+
+### 2.2 The note-type channel carries almost no information
+
+71 charts, **22 260 matched slots** (`FINAL_CHAMPION.md` §5.2, ●).
+
+| quantity | this model | its own floor | corpus §2.1 floor |
+|---|---:|---:|---:|
+| 4-way accuracy | **0.4776** | majority **0.5559** | majority 4-way **0.563** |
+| — margin over majority | **−0.0783**, 95 % CI [−0.1022, −0.0530] | | |
+| hand accuracy | **0.5493** | always-don **0.5975** | always-don **0.620** |
+| — margin | **−0.0482**, CI [−0.0721, −0.0229] | | |
+| **MI(generated; authored), 4-way** | **0.01028 bits** | — | **0.79 % of H(authored) = 1.300 bits** |
+| MI, hand | 0.00534 bits | — | 0.55 % of 0.972 bits |
+
+* **The model sits below its own majority floor on both endpoints**, and below
+  the campaign's corpus baselines (56.3 % / 62.0 %), which are a fixed reference
+  line from a separate 59 961-hit census, not this population.
+* **The honest caveat, which cuts the other way.** The published floors are
+  *argmax* predictors scored against a temperature-1.0 top-p-0.95 *sample*.
+  Against the sampler-appropriate i.i.d. floor (Σ p² = 0.4582 4-way, 0.5190
+  hand) this model is **above** baseline by **+0.0194** and **+0.0303** — the
+  largest 4-way excess of any arm in the campaign. Both readings belong on the
+  record.
+* **MI is flat at ~0.010–0.012 bits on all four arms** (seed band 0.0008). The
+  accuracy differences between arms are marginal-matching, not information.
+* Accuracies are conditional on coverage **0.6442** (precision side) /
+  **0.5842** (recall side); more than a third of generated hits have no authored
+  partner, and coverage is 9 % lower than C1's, so this population is smaller
+  and differently selected than C1's.
+* **No inter-charter agreement ceiling exists** for this split — no
+  (song, course) carries two independent authored charts — so 100 % is not a
+  legitimate target and is not used as one.
+* **Greedy re-decoding was not repeated on this arm.** The C1-era finding that
+  greedy closes part of the gap at an unacceptable `motif_reuse` cost is **not
+  measured on this arm** and must not be quoted for it.
+
+### 2.3 The note budget is 9 % short, and it is the whole of the apparent timing regression
+
+| | C1 (s1234 / s4321) | **C1.3** |
+|---|---:|---:|
+| `onset_precision` | 0.6480 / 0.6502 | **0.6477** |
+| note budget (gen ÷ authored notes) | 1.0046 / 0.9912 | **0.9069** |
+| `onset_recall` | 0.6510 / 0.6445 | **0.5874** |
+| `frac_of_authored_within_jnd` | 0.6463 / 0.6397 | 0.5840 |
+| `type_coverage_recall` | 0.6465 / 0.6401 | 0.5842 |
+
+`precision × budget` equals `onset_recall` to four decimals on **every** arm, by
+construction. Those three "timing" rows are one quantity, and it is the note
+budget, not placement: precision is inside the C1 seed band, median offset is
+0 ms, `exact_slot_lift_over_null` is *above* the band, and significant drift
+tests fall from 6 / 5 to 2.
+
+**The shortfall itself is a real regression and its cause is unknown.** It
+appears on both C1.2 and C1.3, so it belongs to the `--cal-axes` /
+`--balance-exclude-big` pair rather than to the denominator augmentation. It is
+the largest practical regression in this release.
+
+### 2.4 Raising density squeezes drumrolls out — this is the compromise a user can hit
+
+Median spans/min over eight oni songs, per requested density bucket
+(authored **3.885**; `FINAL_CHAMPION.md` §4.3, 8 songs × 5 buckets, ◐):
+
+| requested density | 0 | 3 | 7 | 11 | **15** |
+|---|---:|---:|---:|---:|---:|
+| C1 (s1234) | 3.629 | 3.055 | 4.153 | 2.818 | **2.971** |
+| **C1.3** | 3.550 | 4.413 | 2.172 | 2.727 | **1.279** |
+
+ρ(density → `span_per_min`) = **−0.759**, sign p 0.0078, first→last −2.443 /min.
+At the top of the density request C1.3 delivers **33 %** of the authored span
+rate; C1 delivers 76–79 %, C1.2 44 %.
+
+* **Do not advertise the density knob at its top setting** until this is fixed.
+* There is a named suspect: `cal_span` never left **0.1667** on either
+  challenger, so this is plausibly a calibration term that never trained rather
+  than an intrinsic cost of the knob.
+* The mirror image is a *gain*: C1.3's `span_rate` knob no longer drags hit
+  density, which C1's does.
+
+### 2.5 Absolute bucket calibration is still wrong on the density axis
+
+| axis | exact bucket | within ±1 | mean abs error | mean signed | `calibrated` |
+|---|---:|---:|---:|---:|:--:|
+| density | 5.0 % | 25.0 % | 4.125 | −1.075 | **false** |
+| `big_rate` | **53.3 %** | **93.3 %** | **0.533** | +0.267 | **true** |
+| `span_rate` | — | — | 1.483 | — | true |
+
+The density knob orders the request correctly and lands in the wrong bucket,
+under-shooting by ~1 bucket on average — more than C1.2 does. Rank control is
+real; absolute density targeting is not delivered.
+
+Note the cost that comes with the `big_rate` calibration win: C1.3 has the
+**shortest** `big_rate` ladder of any arm (first→last +0.159 against C1's
+0.183 / 0.224), so the top of that request now delivers less accent than C1's
+did.
+
+### 2.6 It fails its own span-quality gate and triggers a pre-registered kill condition
+
+71 charts hard+oni (467 closed spans) and 26 charts easy+normal (272):
+
+| axis | hard+oni | easy+normal |
+|---|---|---|
+| `duration_low_tail` | **fail** (rate verdict passes; balloon 0.250 beats = 0.75× the shortest authored balloon, roll 0.250 = 0.60×) | **fail** (balloon 0.250 beats = 0.20× the shortest authored balloon; 45/159 balloons below authored support) |
+| `forced_close` | **fail** — **7 / 467 = 1.499 %** (0 unclosed, 7 clamp-forced) | **fail** — **7 / 272 = 2.574 %** |
+| `over_span_max` | **fail** — one balloon 7.72 s = 1.19× the 6.5 s ceiling | **fail** — one balloon 7.58 s = 1.17× |
+| `placement_fit` | **fail** — AUC **0.560** (CI [0.529, 0.593]) against a 0.56 floor; authored reference 0.668 | pass — AUC 0.604 |
+| `orphans` / `swallowed` / `rate` / `duration` | pass | pass |
+| **`a2_kill_condition_not_triggered`** | **fail — the kill condition FIRED** | fail |
+
+The pre-registered kill condition (`SOFTCHART2_DESIGN.md` §4-A2) reads: *rate is
+calibrated but placement/length quality is bad → concede that the bar-scoped
+commitment mechanism is insufficient and escalate to an independent span-state
+objective.* It fires on **all four arms**, including C1. Shipping does not
+discharge it.
+
+**`placement_fit` is a knife-edge, not a finding.** C1.2 passes at 0.561 and
+C1.3 fails at 0.560 with CIs overlapping every other arm across their whole
+width. One thousandth of AUC decides the verdict. It is reported, not used.
+
+### 2.7 One span in seven is a micro-span
+
+All generated spans, 97 charts, `FINAL_CHAMPION.md` §2 (●):
+
+| | n spans | p1 | p5 | p50 | p90 | min | **share < 0.25 s** |
+|---|---:|---:|---:|---:|---:|---:|---:|
+| authored | 762 | 0.177 | 0.273 | 0.963 | 2.419 | 0.083 | **3.8 %** |
+| **this model** | 739 | 0.089 | 0.150 | 0.818 | 2.258 | **0.072** | **14.7 %** |
+
+Per course: easy 6.6 %, normal 8.1 %, hard 13.7 %, oni **25.2 %** — authored
+3.0 / 3.3 / 2.1 / 7.1 %.
+
+* C1.3 does **not** inherit C1.2's worsening (16.4 %); 14.7 % is exactly at the
+  top of the C1 seed band (13.6–14.7 %). It is not an improvement either.
+* **Mechanism is known and unfixed on every arm.** Duration bucket 0 is open at
+  the bottom; 123 of C1.3's 739 spans sit in it, at a median 0.5 beats against
+  the authored 0.917.
+* **The shipped floor does not fix it.** `span_min_sec` / `span_min_beats` fired
+  263 times (`span_close_floor`) plus 33 carries across 97 charts and the
+  < 0.25 s share is still 14.7 %.
+* **The mitigation is unbuilt.** `span_bucket0_profile` is `null` in this
+  package because `scripts/build_span_bucket0_profile.py` has never been run on
+  the train split. This is the shortest available fix and it is not in the box.
+* **The gate cannot see it directly.** `a2_span_duration` compares p50/p90/p99
+  only and passes; `duration_low_tail` catches the support violation but not the
+  rate (its rate verdict passes on this arm).
+* On a `/16` deployment grid the same quantity is 14.7 / 12.4 / **30.6** / 21.8 %
+  — hard is untouched by the augmentation and is the worst cell in the release.
+
+### 2.8 The deployment grid has a two-role conflict that no checkpoint dissolves
+
+`BAR_DENOM` is simultaneously an input the grid supplies and a **trained output
+token** the model reads as a density announcement. `bar_denoms="supplied"`, the
+shipped default, teacher-forces it.
+
+13 songs × 4 courses, one seed, `--cond authored`, `C13_PREDICTION.md`:
+
+| easy / normal / hard / oni | authored grid | **forced `/16`** | `/96` | `/24` |
+|---|---|---|---|---|
+| notes ÷ authored | 0.917 / 0.919 / 0.875 / 0.899 | **1.130 / 1.112 / 1.067 / 1.027** | 0.944 / 0.950 / 0.943 / 1.027 | 1.102 / 1.070 / 0.950 / 0.921 |
+| onset F1 | 0.613 / 0.565 / 0.595 / 0.647 | **0.333 / 0.374 / 0.466 / 0.580** | 0.303 / 0.325 / 0.393 / 0.535 | 0.342 / 0.370 / 0.442 / 0.542 |
+| genuine triplets (den 3/6/12/24) % | 1.13 / 2.48 / 6.34 / 9.43 | **0.00** (arithmetically impossible) | 3.45 / 5.78 / 10.57 / 11.46 | 14.13 / 20.31 / 28.57 / 38.70 |
+| ultra-fine (den 48/96) % | ~0 | 0.00 | **11.53 / 13.22 / 15.62 / 6.60** | 0.00 |
+| nesting (family β = 2) | 0.730 / 0.797 / 0.822 | 0.726 / 0.811 / 0.834 | 0.649 / 0.730 / 0.752 | 0.714 / 0.769 / 0.779 |
+
+Authored triplet rates for reference: 4.20 / 5.72 / 7.43 / 9.95 %; ultra-fine
+~0.00 %.
+
+**The recommended deployment configuration is `--grid bpm --grid-bpm-denom 16`
+with the package default `bar_denoms="supplied"`.** Its two costs, stated
+plainly: **triplets are exactly 0 and cannot be otherwise**, and hard micro-spans
+stay at 30.6 % against an authored 2.1 %.
+
+Four consequences a deployer must carry:
+
+* **Deployment costs 10–46 % of onset F1** relative to the authored grid:
+  easy −46 %, normal −34 %, hard −22 %, oni −10 %. Every `--grid authored`
+  number in this card is therefore optimistic for a user who supplies only BPM.
+* **`p90 |offset|` is 0 ms on the authored grid and 14.6–36.7 ms on the bpm
+  grids.** That residual belongs to the bar edges; no denominator choice touches
+  it.
+* **The model's off-dyadic rate is a property of the model, not the lattice.**
+  Given any 3-divisible grid it puts 2–3× the authored share off the dyadic
+  grid; the grid only decides whether that mass gets called "triplets" (`/24`)
+  or "ultra-fine jitter" (`/96`). This is a training-side defect and no
+  deployment grid fixes it.
+* **The gate suite cannot score the deployment path.** `robustness.py:629-642`
+  raises `ContractError` on uniform-BPM arms, so every gate timing metric is
+  `null` for the path the model ships on; the `/16` figures above come from
+  `timing.py`, a second implementation that reproduces the gate on the authored
+  grid.
+
+**The `deploy_bpm_grid` profile in this package is NOT the recommendation.** It
+sets `bar_denoms="model"`, `bar_denom_mask="div96"`, per `DEPLOY_GRID_FIX.md`
+§7 — a recommendation made explicitly conditional on C1.3 failing, which it did
+not. That arm was measured on **C1 only**: density 1.06 / 1.08 / 0.97 / 1.01 and
+nesting 0.635 / 0.706 / 0.778. **It has never been run on this checkpoint.** The
+profile is kept because it is implemented and because the arm is the obvious
+next experiment, not because it is advised here.
+
+**One contradiction inside this package, stated so nobody has to find it.**
+`config.json:serving_profiles.deploy_bpm_grid.basis` still quotes
+`DEPLOY_GRID_FIX.md` §7 verbatim, including the words *"Recommended now"*, and
+still carries C1's nesting figures. That string is generated by
+`scripts/sc2_package.py`, which is read-only for this release. **This section
+supersedes it.** The generated string is left intact rather than silently
+diverging from the shipped code.
+
+### 2.9 Playability floor: gaps no human hand can play
+
+Recomputed on this arm's own events, 71 charts hard+oni, generated vs authored
+on the same songs (producer `experiments/sc2_eval/ship_c13/playability.py`,
+records `.../records/playability_all_arms.json`; ●):
+
+| | this model | authored |
+|---|---:|---:|
+| inter-onset gaps < 40 ms | **102** across **29** charts | 2 across 1 chart |
+| — as a share of adjacent pairs | **0.296 %** (102 / 34 484) | **0.0053 %** (2 / 38 032) |
+| gaps below the series record 29.4118 ms | **19** | 0 |
+| shortest gap | **20.8 ms** (the serving floor) | 39.7 ms |
+| peak burst in any 1 s window | **17 notes** | 15 notes |
+
+C1 on the same population: 151 gaps (0.395 %), shortest 16.7 ms, peak 22
+notes/s. C1.3 is better on every row and still **56× the authored rate**.
+
+* `min_onset_gap_sec = 0.020` is a **degeneracy guard, not a fix**: it is set
+  below the fastest thing the series has shipped (29.4118 ms,
+  TAIKO-TONGUE-TWISTER oni, BPM 170, 48th notes) precisely so it cannot refuse a
+  chart the domain writes. It fired **125 times** across 97 charts with **0**
+  fail-opens. *A mask cannot fix a distribution.*
+* The `generate.py` constant comment quotes the generated sub-40 ms rate as
+  ~0.35 % against ~0.045 % authored. **That authored figure does not reconcile
+  with the 0.0053 % measured here**, and the two have never been put on the same
+  denominator. The excess is 8× on the comment's accounting and 56× on this one.
+* **Good news that replaces a C1-card claim.** The C1 card reported the model
+  placing hits in 19.6 % of sub-0.25 s scaffolding bars. Under today's serving
+  floors this arm places **0 hits in all 112 such bars** across 97 charts, exactly
+  as the authored side does; `degenerate_bar_hits_blocked` fired 12 times with 0
+  fail-opens.
+
+### 2.10 Pattern proxies regress against C1 on hard+oni
+
+71 charts, one code version, C1 band = |s1234 − s4321| (`FINAL_CHAMPION.md` §3):
+
+| metric (lower is better) | C1 mean | C1 band | **C1.3** | ×band | relative |
+|---|---:|---:|---:|---:|---:|
+| **`motif_reuse_gap`** | 0.3630 | 0.0042 | **0.4222** | +14.1 | **+16.3 %** |
+| `compression_gap` | 0.0976 | 0.0006 | 0.1116 | +23.3 | +14.3 % |
+| `motif_ref_marginal_js` | 0.0331 | 0.0046 | 0.0452 | +2.6 | +36.6 % |
+| `ioi_js_per_chart` | 0.0295 | 0.0006 | **0.0352** | **+9.7** | +19.3 % |
+| `ul_4gram_js_per_chart` | 0.2402 | 0.0055 | 0.2530 | +2.3 | +5.3 % |
+| `class_4gram_js_per_chart` | 0.1337 | 0.0080 | 0.1357 | +0.2 | +1.5 % |
+
+The `×band` column overstates the case — several bands are under 1 % of their own
+level — so the relative column is the honest one. On `motif_reuse` the bootstrap
+CIs are nonetheless **disjoint**: C1 [−0.392, −0.334] / [−0.388, −0.334] against
+C1.3 [−0.447, −0.396]. **The regression is real and not seed noise.**
+
+Four bounds, all measured, none of them a dismissal:
+
+* **It does not happen on easy+normal.** Every pattern proxy there is inside the
+  C1 seed band, `motif_reuse_gap` is marginally *better* than C1's mean, and
+  `class_4gram_js` is better on both challengers.
+* **The decomposition puts the loss on rhythm and hand, while the accent layer
+  improves**: rhythm 0.1251 → 0.1640, hand 0.1774 → 0.2102, accent 0.0606 →
+  **0.0480**.
+* **`ioi_js_per_chart` is where the augmentation shows up** — it reads the exact
+  rational IOI lattice, and it is C1.3's worst pattern metric relative to C1.2
+  (+9.7 band against +1.5).
+* **The proxy reads mostly a channel carrying ~0.010 bits** about the authored
+  type (§2.2), and it has **never been validated against a listener**. It is
+  quoted because §4 names it, not because it is strong.
+
+Realized `motif_reuse` is **0.3165** against an authored **0.7387** — under half
+the authored repetition.
+
+### 2.11 The note-type marginal, and one C1 finding that does NOT reproduce
+
+Big notes are **1.48× the authored rate** pooled hard+oni (0.0873 vs 0.0589),
+worst on the easiest course:
+
+| course | authored | **C1.3** | C1 (s1234 / s4321) |
+|---|---:|---:|---:|
+| easy | 18.80 % | **26.22 % (1.40×)** | 32.68 % / 29.65 % |
+| normal | 11.75 % | **17.41 % (1.48×)** | 19.21 % / 20.77 % |
+| hard | 7.38 % | **10.61 % (1.44×)** | 13.11 % / 12.73 % |
+| oni | 4.88 % | **7.40 % (1.52×)** | 9.29 % / 8.58 % |
+
+The pre-registered stop window (0.045–0.075 pooled hard+oni) is **not met**.
+
+**The C1 card's §2.8 claim does not reproduce here and is corrected.** On C1 the
+realized class marginal was 13× closer to the training loss weights than to the
+corpus. On C1.3 it is the other way round: JS(gen ‖ authored) = **0.00588**
+against JS(gen ‖ class-weight prediction) = **0.00746**
+(JS(authored ‖ weights) = 0.01452). `--balance-exclude-big` moved the marginal
+off the loss weights and toward the corpus. The log-log fit of realized share
+against class weight collapses from slope 0.378 (ρ 0.436) on C1 to slope 0.078
+(ρ 0.156) on C1.3.
+
+### 2.12 Fine-lattice, tuplet and syncopation accuracy are much worse than average
+
+`robustness.py` timing strata, 71 charts, exact-slot rate (overall **0.5840**):
+
+| stratum | n authored | exact-slot | C1 (s1234) |
+|---|---:|---:|---:|
+| lattice step 11.6–23.2 ms | 114 | **0.342** | 0.456 |
+| lattice step 23.2–46.4 ms | 790 | **0.443** | 0.465 |
+| lattice step ≥ 92.9 ms | 24 228 | 0.594 | 0.651 |
+| positions with a denominator divisible by 3 | 3 480 | **0.345** | 0.496 |
+| syncopation band 0 → band 5 | 20 846 → 1 669 | **0.615 → 0.434** | 0.672 → 0.514 |
+| local IOI 25–50 ms | 189 | **0.402** | — |
+| BPM > 250 | 2 570 | **0.499** | — |
+
+Every stratum is lower than C1's, which is the note-budget effect of §2.3 acting
+on a per-stratum recall, but the *shape* is the finding: tuplet positions and
+fast lattice steps are 0.24 below the overall rate, and accuracy decays
+monotonically with syncopation.
+
+### 2.13 Smaller, but on the record
+
+| finding | number | n | source |
+|---|---|---|---|
+| `long_song_no_drift` fails | **2** of 21 metrics significant at BH q ≤ 0.05 (`dens_signed_err`, `plan_dens_bucket_tvd`) — C1 fails 6 / 5 | 71 charts | `robustness.json` ● |
+| Harness-level generation failures | **6** of 71 song-courses refused (`#BRANCHSTART` and non-representable meters); **1** chart exported `gen_tja = null` — identical to C1 on the same population | 71 charts | `index.json` ● |
+| Slot-export off-lattice events | 8 across 97 charts (C1: 5) — the documented contract path, not a crash | 97 charts | decoder counters ● |
+| `escape_hatch` fires | 91 times across 97 charts (C1: 84) | 97 charts | decoder counters ● |
+| Generation wall time | 9.29 s/chart, 4.33 s per audio-minute — **contended**, the GPU was shared, quoted as a declaration only | 71 charts | ● |
+| Complete-bar rest placement | **not measured on this arm** (C1: Jaccard 0.474, recall 0.623) | — | — |
+| Greedy-decode contrast | **not measured on this arm** | — | — |
+| Micro-span seed sensitivity | **not measured on this arm** (C1: 24.0 % vs 32.4 % across two sampling seeds on 19 oni charts) | — | — |
+
+### 2.14 The evidence base is thinner than it looks
+
+* **One seed.** Every number in this card is n = 1 in the training seed. The C1
+  seed band quoted throughout is C1's, used as a reproducibility scale; it is a
+  range over n = 2 and carries no confidence statement.
+* **easy and normal are evaluated on 13 songs.** The 71-chart population is
+  `hard` + `oni` only; the easy/normal population is 26 charts from 13 songs, and
+  its seed band is 4–5× wider than hard+oni's.
+* Every difficulty-family and deployment-grid number is **13 songs, one seed**
+  (◐). Every knob sweep is **8–12 songs × 5 buckets** (◐).
+* At these sizes a gate `pass` carries little information. Worked example from
+  this campaign: the `big_share_of_hits` gate **flips between two C1 training
+  seeds** while the failing seed's point estimate is *smaller*. At n = 36 that
+  gate cannot rank checkpoints, and no pass/fail on it should be quoted as
+  evidence for any arm.
+* **Nothing here is audio-referenced** except `a2_span_placement_fit`. Every
+  other quantity is chart-vs-chart on the exact rational lattice or on hit order.
+  No timing window and no game judgement parameter appears anywhere in this path.
+* **`best_val` is comparable across C1 / C1.2 / C1.3** (one training cache) and
+  **not** comparable to C1.1 or to any arm on a different cache. C1.3's 0.5769 is
+  **2.9 % worse** than C1.2's 0.5605; that was the registered trade and the
+  deployment side won it.
+* **No plan-neutral arm was generated**, so the `plan_neutral_fallback` §4 clause
+  is `na` and the ledger is incomplete by one required clause.
+
+### 2.15 Serving regime: what `motif_constraint: auto` resolves to, and why it matters less here
+
+`motif_constraint: "auto"` resolves to **OFF** for this checkpoint: the training
+cache's `barscript_md5` (`634e3dc3…`) differs from the serving `barscript.py`
+(`9afc715a…`), so the gold sequences it trained on never satisfied the MOTIF hard
+constraint. `train_serve_matched = true` — OFF is the train/serve-matched choice
+and it is the right default.
+
+**Unlike the C1 package, this card needs no correction for it.** Every number in
+this card was measured with the constraint **OFF**, i.e. under exactly this
+package's resolved default; the 20 evaluation runs all record
+`serving_fsm.motif_constraint: off, verified: true`. In particular
+`motif_ref_marginal_js` here is **0.0452** (hard+oni) and **0.0114**
+(easy+normal), both measured OFF.
+
+Two things that still belong on the record:
+
+* **The ON/OFF blast radius was measured on C1 only**, where switching the
+  constraint on moved `motif_ref_marginal_js` 0.0311 → 0.0189 (−39 %). On the C1
+  package the shipped default therefore makes the correct value **0.0311, not the
+  0.0189 in the older tables**. **That contrast has never been measured on
+  C1.3**, so no ON-constraint number should be quoted for this checkpoint.
+* Serving under the constraint ON would be train/serve **mismatched** for this
+  checkpoint and is not a supported configuration.
+
+---
+
+## 3. Serving contract
+
+Every value below is in `config.json:serving`, with its justification in
+`config.json:serving_basis`. Pass them explicitly — the package's
+`serving_kwargs()` does — so the recorded contract is the one that reaches the
+decoder.
+
+| parameter | value | one-line basis |
+|---|---|---|
+| `min_onset_gap_sec` | **0.020** | Degeneracy guard, **not** a corpus percentile. Must stay strictly below the series record of 29.4118 ms (BPM 170, 48ths). Two earlier corpus-derived values (0.0395, 0.0300) were both wrong. See §2.9. |
+| `degenerate_bar_sec` | **0.25** | Authored charts place zero hits in any bar under 0.375 s (3 010 bars). Largest round threshold with zero counterexamples and 1.5× margin; identical to the gate's threshold. Blocks hits only, never span geometry. |
+| `span_min_sec` / `span_min_beats` | 0.0833 / 1⁄3 | Authored population minima. Fail-open, counted. Does not fix §2.7. |
+| `span_bucket0_profile` | **null** | Not built — see §2.7. |
+| `family_bias` / `family_hand_bias` / `family_mode` | 2.0 / 1.5 / `bias` | **Provisional**, never calibrated — see §2.1. Family decode is opt-in. |
+| `motif_constraint` | `auto` → resolves **off** here | Train/serve matching by `barscript.py` md5 — see §2.15. |
+| `bar_denoms` / `bar_denom_mask` | `supplied` / `lattice` | The evaluated regime, and the recommended deployment regime on a uniform `/16` grid. The `deploy_bpm_grid` profile exists but is **not** recommended for this checkpoint — see §2.8. |
+| `greedy` / `temperature` / `top_p` | false / 1.0 / 0.95 | As evaluated. |
+| `plan_temperature` | `null` (follows `temperature`) | Never exercised in evaluation; shipped unset rather than tuned. |
+
+`_has_sync` **must** be forced on at load. `train.py` records
+`sync_token=False` while the training prefix carries the SYNC slot, so a loader
+that trusts the recorded flag drops the slot and every sync bucket reaches the
+decoder as an identical prefix. The package loader does this and records why in
+`config.json:serving_prefix_fix`. This is a workaround; the fix belongs in
+`train.py`.
+
+### Recommended deployment configuration
+
+```
+--grid bpm --grid-bpm-denom 16          # uniform /16 grid
+bar_denoms = "supplied"                 # the package default
+motif_constraint = auto  -> off         # resolved at package time
+family decode: optional; β = 2.0 / 1.5 is provisional
+```
+
+Expected behaviour under it, 13 songs × 4 courses, one seed: notes within
+13 / 11 / 7 / 3 % of authored; onset F1 0.333 / 0.374 / 0.466 / 0.580;
+**zero triplets**; hard micro-spans ~30 %.
+
+---
+
+## 4. Weights are bfloat16
+
+Training ran bf16 autocast and CUDA inference runs bf16 autocast, so fp32
+storage carried no information the forward pass could use — the argument the
+v1.5 release made and verified. Measured cast cost: max absolute delta
+**0.00711** on `frontend.2.weight` (0.349 % of that tensor's max) over 9 107 291
+float elements.
+
+This is an argument about the **compute** dtype, not a proof that the two
+checkpoints decode identically. Sampling is chaotic in the logits, so individual
+charts can differ. The paired fp32-vs-bf16 evaluation has **not** been repeated
+for BarScript. The fp32 checkpoint is preserved in the source run directory and
+is the reference for any bit-level comparison.
+
+---
+
+## 5. Provenance
+
+| | |
+|---|---|
+| serving code | `generate.py` md5 `73b6e49b00fb60f3229c8539bc7a1029`; all 18 `src/softchart/*.py` md5s in `config.json:code.module_md5` |
+| evaluation harness | `sc2_generate_eval_v2_2_serving_floors`, contract `sc2_eval_v2` |
+| evaluation runs | 20 run indices, 1 028 charts, one `serving_floors` signature, `comparable: true`, 6/6 floor counters present |
+| headline population | 71 charts hard+oni (36 songs) + 26 charts easy+normal (13 songs), `--cond authored`, `--grid authored`, seed 1 |
+| deployment population | 13 songs × 4 courses × 10 arms, `--cond authored`, one serving tree, one signature |
+| label spec | `spec.json`, field-for-field equal to the cache manifest's copy (`spec_parameters_match_cache_manifest: true`) |
+| uploaded | **no** (`release_manifest.json: "uploaded": false`) |
+
+Reports: `experiments/sc2_eval/SHIP_DECISION.md`,
+`experiments/sc2_eval/FINAL_CHAMPION.md`,
+`experiments/sc2_eval/C13_PREDICTION.md`,
+`experiments/sc2_eval/C12_EVAL.md`, `experiments/sc2_eval/DEPLOY_GRID_FIX.md`.