Senba release: mars-fm-e010-a07350-exactlag-128x32-20260825-s1-v1 (validation-only)

LICENSE

@@ -0,0 +1,25 @@
+MIT License
+
+Senba adaptation: Copyright (c) 2026 Paper Crane (papercrane.bio)
+MarS-FM source model and code: Copyright (c) 2026 Valence Labs
+
+Permission is hereby granted, free of charge, to any person obtaining a copy
+of this software and associated documentation files (the "Software"), to deal
+in the Software without restriction, including without limitation the rights
+to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
+copies of the Software, and to permit persons to whom the Software is
+furnished to do so, subject to the following conditions:
+
+The above copyright notice and this permission notice shall be included in all
+copies or substantial portions of the Software.
+
+THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
+AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
+LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
+OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+SOFTWARE.
+
+The weights were adapted on mdCATH (compsciencelab/mdCATH), which is
+distributed under CC BY 4.0. Attribute mdCATH when you use these weights.

README.md

@@ -0,0 +1,208 @@
+---
+license: mit
+base_model: valencelabs/mars-fm
+datasets:
+- compsciencelab/mdCATH
+library_name: pytorch
+tags:
+- protein-dynamics
+- molecular-dynamics
+- flow-matching
+- mars-fm
+- mdcath
+- biology
+- pytorch
+---
+
+# Senba
+
+Senba is a source-anchored adaptation of [MarS-FM](https://github.com/valence-labs/mars-fm)
+(Valence Labs, ICLR 2026) for mdCATH backbone transitions at 450 K and one 50-frame lag.
+It is published as a **validation-stage candidate** together with the selection receipt, every
+comparison, and the preregistered protocols, so the claim below can be audited.
+
+**Claim, exactly:** on the complete 28-domain sequence-length-eligible mdCATH validation panel,
+Senba improves ensemble-mean endpoint C-alpha RMSD over source MarS-FM by 0.0326 Å (protein-bootstrap
+95% CI [0.0253, 0.0401]; 27/28 domains improved), improves over coordinate persistence by 0.436 Å
+(CI [0.203, 0.681]; 22/28), and keeps all seven frozen equilibrium-retention observables within
+their limits on the eight-domain equilibrium suite. The gain reproduced across three training seeds.
+
+**Boundary, exactly:** this is not test-set evidence (the 495-domain standard mdCATH test partition
+remains unopened; every receipt records `test_records_opened = 0`), not a state-of-the-art or
+field-leadership claim (the only matched baselines are source MarS-FM and persistence), not
+physical-time kinetics, rare-event, or long-timescale evidence (one lag only), not all-atom
+validation (C-alpha metrics), not broad unseen-protein generalization (28 validation domains at or
+below 256 residues), and not a final model.
+
+Release page: <https://senba.papercrane.bio/>
+
+## Identity
+
+| Field | Value |
+| --- | --- |
+| Run | `mars-fm-e010-a07350-exactlag-128x32-20260825-s1-v1` |
+| Checkpoint | `senba.ckpt` · 136,711,098 bytes · SHA-256 `f02dcf6a97848c834e3c0be7d848c1068d1d0eb7976d8d6bcfbc98d4114e9c7c` |
+| Base model | [`valencelabs/mars-fm`](https://huggingface.co/valencelabs/mars-fm) revision `1e957cd4b9f1a054ceb0c2104f40ba6f3cc0d8d0`, `mars_fm.ckpt` SHA-256 `f56794cc5d1a8e5263031944f932384dd785fe945553612fe76fe1e9191fa879` |
+| Source code | [`valence-labs/mars-fm`](https://github.com/valence-labs/mars-fm) commit `4fc17d86cd7e22dda6353f2f1f6bf4b40c9ce946` |
+| Architecture | Unchanged MarS-FM; 34,152,521 parameters, all trainable during adaptation |
+| Adaptation data | mdCATH, bounded development subset of 128 train and 32 validation domains (manifest `sha256:163ab0043d86b5a2731e1b1a233af27fb069049897d657fd5c39bd54b1343b3f`), 450 K replicas |
+| Objective | Observed same-replica frame pairs separated by exactly 50 frames (`exact_lag`), cluster-balanced starts, no replica crossing |
+| Anchoring | Source-checkpoint weight interpolation, α = 0.0735 (7.35% of the fine-tuning displacement retained) |
+| Frozen | 2026-08-25 |
+| Test records opened | 0 |
+
+The checkpoint is a PyTorch Lightning checkpoint. Its `hyper_parameters` contain an
+`argparse.Namespace`, so loading under `torch>=2.6` requires
+`torch.serialization.add_safe_globals([argparse.Namespace])` (the example script does this).
+
+## How it was made
+
+1. **Parent fine-tune.** All MarS-FM parameters were fine-tuned from the pinned source checkpoint
+   for 10 epochs (160 optimizer updates, 30,720 sampled exact-lag transitions), batch size 8
+   domains, learning rate 3e-6, fresh Adam, seed 20260825, on one NVIDIA H100 in 121.80 s (peak
+   64.7 GB allocated). Validation loss on a fixed transition population with replayed flow
+   randomness moved from the replayed source value 1.724479 to 1.693136. Parent run
+   `mars-fm-epoch-sweep-e010-exactlag-128x32-20260825-s1-v1`, checkpoint SHA-256
+   `1bd5c20b5f3aa8587dce6440af6580f0c2def98d0b868a39969ef854bd729a25`.
+2. **Anchoring.** The parent damaged the source model's equilibrium ensembles and was rejected as a
+   deployable candidate. Senba is `source + 0.0735 · (parent − source)` in weight space.
+3. **Selection.** α was chosen on the frozen epoch-sweep protocol, then the candidate had to pass the
+   full-panel transition comparison and the equilibrium-retention suite without any change to the
+   checkpoint, inference settings, metrics, or population.
+
+## Evidence
+
+All intervals are protein-level bootstraps with 100,000 resamples. Positive values mean Senba's
+sampled endpoints are closer to the observed endpoint than the baseline's. Inference used 10 ODE
+steps, four starts per replica, four samples per start, five 450 K replicas, lag 50.
+
+### Full eligible validation panel (28 domains, ≤256 residues)
+
+| Comparison | Ensemble-mean endpoint C-alpha RMSD improvement | 95% CI | Domains improved |
+| --- | --- | --- | --- |
+| vs source MarS-FM | +0.0326 Å | [0.0253, 0.0401] | 27 / 28 |
+| vs coordinate persistence | +0.436 Å | [0.203, 0.681] | 22 / 28 |
+| vs previous Senba (5-epoch, α = 0.08) | +0.00216 Å | [0.00086, 0.00345] | 22 / 28 |
+
+Secondary summaries on the same outputs: best-of-k RMSD improvement vs source +0.0372 Å,
+single-sample RMSD improvement vs source +0.0689 Å, displacement-magnitude MAE improvement
++0.0292 Å, direction-cosine improvement +0.0014. Per-domain rows are in
+`receipts/full-transition-comparison.json`.
+
+### Equilibrium retention (8 domains, official tree sampling, 500 samples)
+
+Degradation is measured in each observable's unfavorable direction against source MarS-FM;
+negative values are improvements. Limits were frozen before any candidate output existed.
+
+| Observable | Mean degradation | 95% CI | Mean limit | Worst domain / limit | Pass |
+| --- | --- | --- | --- | --- | --- |
+| DCCM MAE | −0.000660 | [−0.001625, 0.000285] | 0.005 | 0.001768 / 0.01 | ✓ |
+| Native-contact fraction W1 | +0.000271 | [−0.001594, 0.001946] | 0.015 | 0.003055 / 0.03 | ✓ |
+| Pairwise RMSD W1 (Å) | −0.0437 | [−0.1260, 0.0493] | 0.25 | 0.2076 / 0.5 | ✓ |
+| PCA subspace RMSIP | +0.00416 | [−0.00023, 0.01075] | 0.03 | 0.02536 / 0.06 | ✓ |
+| Clash fraction | −0.0000012 | [−0.0000246, 0.0000225] | 0.0005 | 0.0000575 / 0.001 | ✓ |
+| Radius of gyration W1 (Å) | −0.0422 | [−0.0775, −0.0080] | 0.15 | 0.0319 / 0.3 | ✓ |
+| RMSF RMSE (Å) | −0.0222 | [−0.0776, 0.0396] | 0.15 | 0.1471 / 0.3 | ✓ |
+
+### Replication (three independently trained parents, 8-domain compact panel)
+
+| Parent seed | Gain vs source | 95% CI | Domains improved | Matched difference vs previous Senba |
+| --- | --- | --- | --- | --- |
+| 1 (20260825) | +0.0301 Å | [0.0184, 0.0431] | 8 / 8 | +0.00308 Å |
+| 2 (20260826) | +0.0229 Å | [0.0133, 0.0343] | 8 / 8 | −0.00110 Å |
+| 3 (20260827) | +0.0290 Å | [0.0183, 0.0386] | 8 / 8 | +0.00557 Å |
+
+Three-seed mean gain over source: +0.0273 Å. Mean matched difference vs the previous Senba:
++0.00252 Å. The seed-2 difference is negative and its interval crosses zero, so the correct reading
+is a modest average improvement over the previous checkpoint, not a uniform win at every seed.
+
+### Negative results that remain part of the evidence
+
+The fully fine-tuned exact-lag parents (5, 10, 20, 40, 80, and 160 epochs), the 25% and 50%
+hybrid objectives, and the larger anchors (α ≥ 0.1) either damaged equilibrium retention or lost
+the transition gain. They are recorded in the protocols and are not substitutes for Senba.
+
+## Intended use
+
+Research inspection of single-chain, all-heavy-atom protein structures of at most 256 residues:
+draw independent samples of where MarS-FM-style flow matching places the backbone after the
+trained lag, and compare against source MarS-FM. Each draw is an independent sample, not a frame
+of a trajectory. Do not use Senba for clinical, diagnostic, or design decisions, and do not
+present its outputs as physical-time dynamics, kinetics, or equilibrium ensembles beyond the
+retention evidence above.
+
+## How to run
+
+```bash
+git clone https://github.com/valence-labs/mars-fm.git && cd mars-fm
+git checkout 4fc17d86cd7e22dda6353f2f1f6bf4b40c9ce946
+pip install "torch>=2.5" "pytorch-lightning>=2.4" torchdiffeq "numpy<2" \
+            dm-tree fair-esm biopython pandas scipy
+hf download pikachuandme1/senba senba.ckpt inference_example.py --local-dir senba
+
+PYTHONPATH=. python senba/inference_example.py \
+  --pdb 1ubq.pdb --chain A \
+  --checkpoint senba/senba.ckpt \
+  --expect-sha256 f02dcf6a97848c834e3c0be7d848c1068d1d0eb7976d8d6bcfbc98d4114e9c7c \
+  --samples 8 --ode-steps 10 --seed 20260904 \
+  --out senba-1ubq.pdb
+```
+
+The script writes a multi-model PDB: MODEL 1 is the input, MODELS 2 to K+1 are K independent
+draws. It refuses structures with missing heavy atoms, non-standard residues, or several chains
+rather than silently zero-filling, and warns when the input exceeds the validated 256 residues.
+A GPU is recommended; CPU works but is slow.
+
+To load the checkpoint yourself, follow `inference_example.py`: build the same batch layout as
+MarS-FM (`torsions`, `torsion_mask`, `trans`, `rots`, `seqres`, `mask`), then call
+`MarSModule.inference(batch, num_steps=10)`.
+
+## Files
+
+| Path | Contents |
+| --- | --- |
+| `senba.ckpt` | The selected checkpoint, byte-identical to the receipt hash |
+| `inference_example.py` | Sampling script against the pinned MarS-FM source |
+| `receipts/selection-v2.json` | Selection receipt (`dynaview-mars-fm-epoch-sweep-selection-v2`) |
+| `receipts/interpolation-config.json`, `receipts/interpolation-result.json` | Anchoring receipt for the selected checkpoint |
+| `receipts/parent/training-config.json`, `receipts/parent/training-result.json` | Seed-1 parent fine-tune receipt with per-epoch validation losses |
+| `receipts/full-transition-comparison.json` | 28-domain transition comparison vs source and persistence |
+| `receipts/head-to-head-vs-previous-senba.json` | 28-domain comparison vs the previous Senba checkpoint |
+| `receipts/equilibrium-suite-comparison.json` | Seven-observable retention suite with per-domain rows |
+| `receipts/replications/seed2/`, `receipts/replications/seed3/` | Seed-2 and seed-3 parent, anchoring, and compact-panel receipts |
+| `protocols/` | The four preregistered protocols, frozen before the outputs they govern existed |
+| `release-manifest.json` | SHA-256 of every published file |
+
+## Compute
+
+The three ten-epoch parents took 378.21 H100-seconds in aggregate (121.80, 126.07 mean). Observed
+Modal billing for the whole 2026-08-25 search and validation campaign was $19.45 ($10.60
+adaptation, $8.85 validation), including rejected, interrupted, and long-epoch runs; that is not
+the cost of the selected two-minute training run alone.
+
+## License and attribution
+
+Weights and code: MIT (see `LICENSE`). MarS-FM source model and code are MIT, Copyright (c) 2026
+Valence Labs. The adaptation data, mdCATH, is CC BY 4.0; attribute it when you use these weights.
+
+## Citation
+
+If you use Senba, cite MarS-FM and mdCATH:
+
+```bibtex
+@inproceedings{marsfm2026,
+  title     = {MarS-FM: Generative Modeling of Molecular Dynamics via Markov State Models},
+  booktitle = {International Conference on Learning Representations (ICLR)},
+  year      = {2026},
+  note      = {arXiv:2509.24779}
+}
+
+@article{mdcath2024,
+  title   = {mdCATH: A Large-Scale MD Dataset for Data-Driven Computational Biophysics},
+  journal = {arXiv preprint arXiv:2407.14794},
+  year    = {2024}
+}
+```
+
+Senba itself: Paper Crane, "Senba: a source-anchored MarS-FM adaptation for mdCATH transitions,"
+model release, 2026, <https://huggingface.co/pikachuandme1/senba>.

inference_example.py

@@ -0,0 +1,186 @@
+#!/usr/bin/env python
+"""Sample Senba (adapted MarS-FM) transition endpoints for one protein structure.
+
+Senba is a MarS-FM checkpoint, so it loads with the pinned MarS-FM source
+(https://github.com/valence-labs/mars-fm at 4fc17d86cd7e22dda6353f2f1f6bf4b40c9ce946).
+Put that repository on PYTHONPATH, then run:
+
+    python inference_example.py --pdb 1ubq.pdb --chain A \
+        --checkpoint senba.ckpt \
+        --expect-sha256 f02dcf6a97848c834e3c0be7d848c1068d1d0eb7976d8d6bcfbc98d4114e9c7c \
+        --samples 8 --ode-steps 10 --seed 20260904 --out senba-1ubq.pdb
+
+What you get: a multi-model PDB whose MODEL 1 is the input heavy-atom structure
+and whose MODELS 2..K+1 are K independent samples of where the model places the
+chain after the trained lag (50 mdCATH frames at 450 K). The samples are not a
+trajectory, carry no physical clock, and were validated only on single-chain
+mdCATH domains of at most 256 residues. Every claim about Senba is bounded to
+that validation evidence; see README.md.
+
+Input requirements: one protein chain, standard amino acids only, every heavy
+atom present (crystal structures with truncated side chains are rejected rather
+than silently zero-filled). HETATM records, hydrogens, alternate locations other
+than A, and other chains are dropped before parsing.
+"""
+
+from __future__ import annotations
+
+import argparse
+import hashlib
+import sys
+from pathlib import Path
+
+import numpy as np
+import torch
+
+torch.serialization.add_safe_globals([argparse.Namespace])
+
+from mars.data.geometry import atom14_to_atom37, atom14_to_frames, atom37_to_torsions  # noqa: E402
+from mars.model.module import MarSModule  # noqa: E402
+from mars.utils import atom14_to_pdb  # noqa: E402
+from mars.vendored.openfold import protein  # noqa: E402
+from mars.vendored.openfold import residue_constants as rc  # noqa: E402
+
+VALIDATED_MAX_RESIDUES = 256
+TRAINED_LAG_FRAMES = 50
+TRAINED_TEMPERATURE_K = 450
+
+
+def sha256_file(path: Path) -> str:
+    digest = hashlib.sha256()
+    with path.open("rb") as handle:
+        while chunk := handle.read(8 * 1024 * 1024):
+            digest.update(chunk)
+    return digest.hexdigest()
+
+
+def clean_pdb_text(pdb_text: str, chain_id: str | None) -> str:
+    """Keep ATOM records of one chain with primary alternate locations only."""
+    kept: list[str] = []
+    for line in pdb_text.splitlines():
+        if not line.startswith("ATOM  "):
+            continue
+        if len(line) < 54:
+            raise ValueError("truncated ATOM record in input PDB")
+        if line[16] not in (" ", "A"):
+            continue
+        if chain_id is not None and line[21].strip() != chain_id:
+            continue
+        kept.append(line[:16] + " " + line[17:])
+    if not kept:
+        raise ValueError("no ATOM records matched the requested chain")
+    return "\n".join(kept) + "\nEND\n"
+
+
+def pdb_to_atom14(pdb_text: str, chain_id: str | None) -> tuple[np.ndarray, str]:
+    prot = protein.from_pdb_string(clean_pdb_text(pdb_text, chain_id))
+    if prot.chain_index is not None and len(np.unique(prot.chain_index)) != 1:
+        raise ValueError("the structure has several chains; pass --chain")
+    if prot.aatype.size == 0:
+        raise ValueError("no residues were parsed")
+    if np.any(prot.aatype >= rc.restype_num):
+        raise ValueError("non-standard residues are not supported by MarS-FM")
+    n_res = int(prot.aatype.shape[0])
+    atom14 = np.zeros((n_res, 14, 3), dtype=np.float32)
+    for index, aatype in enumerate(prot.aatype.tolist()):
+        names = rc.restype_name_to_atom14_names[rc.restype_1to3[rc.restypes[aatype]]]
+        for slot, name in enumerate(names):
+            if not name:
+                continue
+            atom_index = rc.atom_order[name]
+            if prot.atom_mask[index, atom_index] <= 0:
+                raise ValueError(
+                    f"residue {index + 1} ({rc.restypes[aatype]}) is missing heavy atom {name}; "
+                    "supply a structure with complete side chains"
+                )
+            atom14[index, slot] = prot.atom_positions[index, atom_index]
+    sequence = "".join(rc.restypes[aatype] for aatype in prot.aatype.tolist())
+    return atom14, sequence
+
+
+def build_batch(atom14: np.ndarray, sequence: str, samples: int, device: torch.device) -> dict:
+    """Mirror the batch layout used by the frozen Senba validation runs."""
+    seqres = torch.tensor([rc.restype_order[residue] for residue in sequence])
+    seqres = seqres.unsqueeze(0).repeat(samples, 1)
+    repeated = np.repeat(atom14[None], samples, axis=0).astype(np.float32, copy=False)
+    frames = atom14_to_frames(torch.from_numpy(repeated))
+    atom37 = torch.from_numpy(atom14_to_atom37(repeated, seqres)).float()
+    torsions, torsion_mask = atom37_to_torsions(atom37, seqres)
+    batch = {
+        "torsions": torsions.unsqueeze(1),
+        "torsion_mask": torsion_mask,
+        "trans": frames._trans.unsqueeze(1),
+        "rots": frames._rots._rot_mats.unsqueeze(1),
+        "seqres": seqres,
+        "mask": torch.ones_like(seqres, dtype=torch.float32),
+    }
+    return {key: value.to(device) for key, value in batch.items()}
+
+
+def kabsch_rmsd(reference: np.ndarray, mobile: np.ndarray) -> float:
+    """C-alpha RMSD after optimal superposition (no reflection)."""
+    ref = reference - reference.mean(axis=0)
+    mob = mobile - mobile.mean(axis=0)
+    covariance = mob.T @ ref
+    u, _, vt = np.linalg.svd(covariance)
+    sign = np.sign(np.linalg.det(u @ vt))
+    correction = np.diag([1.0, 1.0, sign])
+    rotation = u @ correction @ vt
+    aligned = mob @ rotation
+    return float(np.sqrt(np.mean(np.sum((aligned - ref) ** 2, axis=1))))
+
+
+def main() -> int:
+    parser = argparse.ArgumentParser(description=__doc__, formatter_class=argparse.RawDescriptionHelpFormatter)
+    parser.add_argument("--pdb", required=True, type=Path, help="input PDB file (heavy atoms, one chain)")
+    parser.add_argument("--chain", default=None, help="chain identifier to use when the file has several")
+    parser.add_argument("--checkpoint", required=True, type=Path, help="path to senba.ckpt")
+    parser.add_argument("--expect-sha256", default=None, help="refuse to run unless the checkpoint hash matches")
+    parser.add_argument("--samples", type=int, default=8, help="independent endpoint draws")
+    parser.add_argument("--ode-steps", type=int, default=10, help="flow ODE steps (validation used 10)")
+    parser.add_argument("--seed", type=int, default=0)
+    parser.add_argument("--out", required=True, type=Path, help="multi-model PDB to write")
+    parser.add_argument("--device", default="cuda" if torch.cuda.is_available() else "cpu")
+    args = parser.parse_args()
+
+    if args.samples < 1 or args.ode_steps < 1:
+        parser.error("--samples and --ode-steps must be positive")
+    digest = sha256_file(args.checkpoint)
+    if args.expect_sha256 is not None and digest != args.expect_sha256.lower().removeprefix("sha256:"):
+        print(f"checkpoint hash {digest} does not match --expect-sha256", file=sys.stderr)
+        return 2
+
+    atom14, sequence = pdb_to_atom14(args.pdb.read_text(), args.chain)
+    if len(sequence) > VALIDATED_MAX_RESIDUES:
+        print(
+            f"warning: {len(sequence)} residues exceeds the validated scope of "
+            f"{VALIDATED_MAX_RESIDUES}; outputs are outside the published evidence",
+            file=sys.stderr,
+        )
+
+    device = torch.device(args.device)
+    torch.manual_seed(args.seed)
+    if device.type == "cuda":
+        torch.cuda.manual_seed_all(args.seed)
+    model = MarSModule.load_from_checkpoint(str(args.checkpoint), map_location="cpu")
+    model.eval().to(device)
+    batch = build_batch(atom14, sequence, args.samples, device)
+    with torch.no_grad():
+        generated_atom14, _ = model.inference(batch, num_steps=args.ode_steps)
+    samples = generated_atom14[:, 0].float().cpu().numpy()  # [K, L, 14, 3]
+
+    aatype = np.array([rc.restype_order[residue] for residue in sequence])
+    atom14_to_pdb(np.concatenate([atom14[None], samples], axis=0), aatype, str(args.out))
+
+    reference_ca = atom14[:, 1]
+    rmsds = [kabsch_rmsd(reference_ca, sample[:, 1]) for sample in samples]
+    print(f"checkpoint sha256 {digest}")
+    print(f"sequence length {len(sequence)}; lag {TRAINED_LAG_FRAMES} frames at {TRAINED_TEMPERATURE_K} K")
+    print(f"wrote {args.out} with MODEL 1 = input and MODELS 2..{args.samples + 1} = independent draws")
+    print("aligned C-alpha RMSD of each draw from the input (A): " + ", ".join(f"{value:.3f}" for value in rmsds))
+    print("these draws are independent samples at one lag, not a trajectory or a physical-time prediction")
+    return 0
+
+
+if __name__ == "__main__":
+    sys.exit(main())

protocols/MARS_FM_ANCHOR08_FULL_VALIDATION_V1_PROTOCOL.md

@@ -0,0 +1,125 @@
+# MarS-FM Anchor08 complete eligible validation protocol
+
+Status: frozen validation-stage protocol; no test, final-model, commercial,
+state-of-the-art, or field-changing claim.
+
+Frozen on 2026-08-02 after the seed-1 Anchor08 candidate passed the previously
+frozen eight-protein transition and equilibrium-retention suites, but before
+generating or inspecting exact-lag outputs for the remaining twenty eligible
+validation proteins.
+
+## Scientific question
+
+Does the small exact-lag transition gain of the frozen seed-1 Anchor08
+checkpoint survive the complete sequence-length-eligible validation population
+without changing the checkpoint, inference settings, metrics, or population?
+
+This is a candidate-qualification experiment. A pass does not establish
+physical kinetics, broad sequence generalization, field leadership, or test
+performance.
+
+## Immutable identities
+
+- Standard mdCATH manifest:
+  `sha256:53e81a90c0dcf3e00a814744f62d25db52f5aa8790475294dac451d36d728b8b`.
+- Development subset manifest:
+  `sha256:163ab0043d86b5a2731e1b1a233af27fb069049897d657fd5c39bd54b1343b3f`.
+- Source MarS-FM checkpoint:
+  `sha256:f56794cc5d1a8e5263031944f932384dd785fe945553612fe76fe1e9191fa879`.
+- Candidate run:
+  `mars-fm-mdcath-anchor08-exactlag-128x32-20260801-s1-v1`.
+- Candidate checkpoint:
+  `sha256:802042f2d5c1aa8adf07a9a935063ed970aa38fdfba1993b236cb0364c9a0fae`.
+- MarS-FM source revision:
+  `4fc17d86cd7e22dda6353f2f1f6bf4b40c9ce946`.
+- Standard test records opened: zero.
+
+## Complete eligible population
+
+Use every validation record in frozen manifest order whose sequence length is
+at most the pinned MarS-FM crop of 256 residues. No eligible protein may be
+removed after evaluation starts:
+
+1. `1e7uA04` (158)
+2. `1rquA01` (52)
+3. `1q16B03` (77)
+4. `1d1rA00` (83)
+5. `2k4qA00` (156)
+6. `2f2cA02` (106)
+7. `5k2mA01` (96)
+8. `2dp9A01` (100)
+9. `1nktA02` (122)
+10. `3mg3A01` (147)
+11. `2incC00` (83)
+12. `4nreA01` (120)
+13. `1h4yA00` (115)
+14. `2kjrA01` (80)
+15. `4qncB00` (82)
+16. `3aekA03` (123)
+17. `2ji7A03` (212)
+18. `3i01A03` (228)
+19. `2izwC00` (208)
+20. `2k54A00` (123)
+21. `4pkiG02` (171)
+22. `4b96A00` (151)
+23. `4deqA02` (53)
+24. `3pt1A02` (120)
+25. `3atsA01` (116)
+26. `3d8lA00` (91)
+27. `3tdgA01` (66)
+28. `1jeyB03` (102)
+
+The first eight existing receipts may be reused only after their settings,
+checkpoint identities, source transitions, and hashes pass the same aggregator
+validation as newly generated receipts.
+
+## Frozen transition and inference settings
+
+- Role: validation only.
+- Temperature: 450 K.
+- Replicas: all five available replicas.
+- Starts per replica: four deterministic evenly spaced starts.
+- Transition lag: exactly 50 source frames within one replica.
+- Stochastic samples per start: four.
+- ODE steps per sample: ten.
+- Inference seed: `20260801` for both source and candidate.
+- Alignment and C-alpha metric implementation: unchanged from
+  `dynaview-mars-fm-exact-lag-validation-v1`.
+- Bootstrap unit: protein.
+- Bootstrap resamples: 100,000.
+- Bootstrap seed: `20260801`.
+
+Best-of-four error is diagnostic only because selecting a sample using the
+unknown future is not an inference-time capability. The primary score is the
+ensemble-mean endpoint C-alpha RMSD.
+
+## Frozen advancement gate
+
+Anchor08 advances to multi-lag and kinetic validation only if all conditions
+hold across the complete 28-protein population:
+
+1. mean ensemble-endpoint RMSD improvement over source MarS-FM is positive and
+   its protein-bootstrap 95% interval excludes zero;
+2. at least 75% of proteins improve over source MarS-FM;
+3. mean ensemble-endpoint RMSD improvement over coordinate persistence is
+   positive and its protein-bootstrap 95% interval excludes zero;
+4. at least 75% of proteins improve over coordinate persistence;
+5. no source or candidate receipt uses a test record, crosses replicas, changes
+   the lag, or disagrees on the exact transition population;
+6. the candidate checkpoint and every receipt retain the immutable identities
+   above.
+
+No minimum practical effect size is inferred from this experiment. Even a pass
+only establishes broad validation nonregression/improvement at one lag. The
+later field-competitiveness gate must preregister a practically meaningful
+effect threshold before matched competitor outputs are inspected.
+
+## Failure and continuation policy
+
+If any advancement condition fails, do not open standard test data and do not
+delete proteins, change inference seeds, or relax the gate. Diagnose failures
+on train/validation data under a fresh run identity. If the gate passes, freeze
+the result and design the multi-lag, replica-noise, semigroup/kinetic, and
+matched-competitor protocols before generating those outputs.
+
+All generated arrays and receipts remain validation-only development evidence.

protocols/MARS_FM_EPOCH_SWEEP_V1_PROTOCOL.md

@@ -0,0 +1,108 @@
+# MarS-FM exact-lag epoch-sweep v1 development protocol
+
+Status: frozen development sweep; no test, final-model, state-of-the-art, or
+commercial claim.
+
+Frozen on 2026-08-25 before launching any run in this sweep or inspecting any
+new candidate output.
+
+## Question
+
+The existing five-epoch exact-lag parent ended with validation loss still
+decreasing. This sweep asks whether a longer exact-lag fine-tune, followed by
+source-weight anchoring, produces a larger endpoint improvement without
+damaging the source model's equilibrium ensemble.
+
+This is a retrospective development study. The 128-train/32-validation subset
+and the 28 sequence-length-eligible validation proteins were used by the prior
+Anchor08 program. Results from this sweep cannot be described as pristine
+validation. The 495-record standard test partition remains unopened.
+
+## Fixed training contract
+
+Every parent starts from the same pinned public MarS-FM checkpoint and changes
+only the epoch budget.
+
+- Source checkpoint SHA-256:
+  `f56794cc5d1a8e5263031944f932384dd785fe945553612fe76fe1e9191fa879`.
+- MarS-FM source revision:
+  `4fc17d86cd7e22dda6353f2f1f6bf4b40c9ce946`.
+- Development subset semantic SHA-256:
+  `163ab0043d86b5a2731e1b1a233af27fb069049897d657fd5c39bd54b1343b3f`.
+- Roles: 128 train domains and 32 validation domains, with no test records.
+- Temperature: 450 K.
+- Objective: observed same-replica pairs separated by exactly 50 frames.
+- Trainable scope: all 34,152,521 parameters.
+- Optimizer: fresh Adam state from the public checkpoint.
+- Learning rate: `3e-6`.
+- Batch size: 8.
+- Seed: `20260825`.
+- Validation population and flow randomness: fixed and replayed.
+- Epoch budgets: `[5, 10, 20, 40, 80, 160]`.
+
+Each epoch contains 16 optimizer steps, so the parent budgets contain
+`[80, 160, 320, 640, 1280, 2560]` optimizer steps.
+
+## Source-anchor grid
+
+Longer parents can move farther from the source checkpoint, so a fixed
+interpolation coefficient would confound training length with update size.
+Each parent is therefore interpolated at four coefficients chosen so that the
+heuristic product `epochs * alpha` is one of `[0.20, 0.30, 0.40, 0.60]`.
+This product is a grid-design device, not a physical quantity or an assumption
+that weight displacement is linear in epochs.
+
+| Epochs | Interpolation coefficients |
+| ---: | --- |
+| 5 | `0.04`, `0.06`, `0.08`, `0.12` |
+| 10 | `0.02`, `0.03`, `0.04`, `0.06` |
+| 20 | `0.01`, `0.015`, `0.02`, `0.03` |
+| 40 | `0.005`, `0.0075`, `0.01`, `0.015` |
+| 80 | `0.0025`, `0.00375`, `0.005`, `0.0075` |
+| 160 | `0.00125`, `0.001875`, `0.0025`, `0.00375` |
+
+All 24 interpolated checkpoints must be created and screened. A missing or
+failed run is reported, not silently replaced.
+
+## Development screen
+
+The source and every candidate use the same compact eight-protein panel, all
+five replicas, four fixed starts per replica, four samples per start, exact lag
+50, ten ODE steps, and inference seed `20260825`. Proteins are the bootstrap
+unit with 100,000 resamples and bootstrap seed `20260825`.
+
+A candidate remains eligible only if:
+
+1. its parent improves the fixed paired validation loss over the source;
+2. its mean ensemble-endpoint RMSD improvement over source is positive with a
+   95% protein-bootstrap interval above zero;
+3. at least six of eight proteins improve over source;
+4. its mean improvement over coordinate persistence is positive; and
+5. it passes all seven existing equilibrium-retention limits on the fixed
+   `1e7uA04` pilot using breadth 200, 500 retained samples, 50 ODE steps, and
+   seed `20260825`.
+
+Among eligible candidates, select the largest mean ensemble-endpoint RMSD
+improvement over source. An exact tie is broken by fewer epochs, then by the
+smaller interpolation coefficient.
+
+If the selected candidate uses 160 epochs, the epoch boundary is unresolved.
+Extend the parent grid to 320 and 640 epochs with the same four heuristic
+products before selecting a final configuration. Otherwise do not extend the
+grid after observing results.
+
+## Qualification after selection
+
+Freeze the selected epoch count, interpolation coefficient, checkpoint,
+inference settings, and selection receipt before further evaluation. Then:
+
+1. run the selected seed-1 candidate on all 28 eligible validation proteins;
+2. repeat the exact training and interpolation configuration with seeds
+   `20260826` and `20260827` and run the compact transition panel;
+3. run the existing eight-protein equilibrium-retention aggregate for the
+   selected seed-1 checkpoint; and
+4. leave the standard test partition unopened.
+
+These qualification results remain development evidence because the
+validation population was previously inspected. Any new external or test
+claim requires a separately frozen untouched evaluation boundary.

protocols/MARS_FM_EQUILIBRIUM_SUITE_PROTOCOL.md

@@ -0,0 +1,37 @@
+# MarS-FM equilibrium-suite validation protocol
+
+Status: frozen validation-stage protocol; no final-model or state-of-the-art
+claim.
+
+Frozen on 2026-07-31 after anchor08 passed the designated one-protein pilot,
+but before generating source or anchor08 equilibrium outputs for the other
+seven proteins.
+
+## Population and generation
+
+Use the first eight validation domains in the frozen 128x32 manifest whose
+sequence length is at most the pinned model crop of 256 residues:
+`1e7uA04`, `1rquA01`, `1q16B03`, `1d1rA00`, `2k4qA00`, `2f2cA02`,
+`5k2mA01`, and `2dp9A01`.
+
+For both pinned source MarS-FM and the seed-1 anchor08 candidate, generate the
+official breadth-first tree with 200 children per call, 500 retained samples,
+chunk size 100, 50 ODE steps, and seed `20260731`. Score against all five
+450 K replicas with the existing all-C-alpha aligned equilibrium metrics.
+
+## Protein-level retention gate
+
+For each metric, define degradation in its unfavorable direction and use
+proteins as the paired bootstrap unit with 100,000 resamples and seed
+`20260801`. A metric passes only if:
+
+1. the upper endpoint of its 95% bootstrap interval for mean degradation is
+   no greater than the frozen per-protein pilot limit;
+2. at least 75% of proteins individually remain within that limit; and
+3. no protein degrades by more than twice that limit.
+
+All seven metrics and their limits are inherited unchanged from
+`MARS_FM_HYBRID_V1_PROTOCOL.md`. Every metric must pass. This is an internal
+retention gate, not an equivalence test or evidence of field leadership.
+
+Standard mdCATH test records remain unopened.

protocols/MARS_FM_MDCATH_ADAPTATION_PROTOCOL.md

@@ -0,0 +1,88 @@
+# MarS-FM mdCATH adaptation protocol
+
+Status: frozen development protocol; no final-model or state-of-the-art claim.
+
+Frozen on 2026-07-31 before inspecting the first adapted checkpoint's
+structural or exact-lag benchmark outputs.
+
+## Scientific question
+
+Can the pinned public MarS-FM checkpoint be adapted on the standard mdCATH
+training split so that it preserves its strong equilibrium-ensemble behavior
+while becoming a useful predictor of observed, same-replica lagged
+transitions?
+
+The two development objectives are separate:
+
+- `msm`: reproduce MarS-FM's state-balanced transition sampling;
+- `exact_lag`: use observed `(t, t + 50 frames)` pairs from one replica, with
+  starting-state cluster balancing and no replica crossing.
+
+The exact-lag objective is a DynaView development contribution. It is not a
+validated kinetic model unless it passes the kinetic gates below.
+
+## Frozen identities and data boundary
+
+- MarS-FM source code commit:
+  `4fc17d86cd7e22dda6353f2f1f6bf4b40c9ce946`.
+- MarS-FM model revision:
+  `1e957cd4b9f1a054ceb0c2104f40ba6f3cc0d8d0`.
+- Source checkpoint SHA-256:
+  `f56794cc5d1a8e5263031944f932384dd785fe945553612fe76fe1e9191fa879`.
+- Standard split semantic SHA-256:
+  `53e81a90c0dcf3e00a814744f62d25db52f5aa8790475294dac451d36d728b8b`.
+- Development subset semantic SHA-256:
+  `163ab0043d86b5a2731e1b1a233af27fb069049897d657fd5c39bd54b1343b3f`.
+- Development subset: 128 train and 32 validation domains, all five 450 K
+  replicas, with byte-bound source and converted-array receipts.
+- Standard test records opened during adaptation and selection: zero.
+
+All validation transition items are materialized once. MarS-FM's stochastic
+flow noise and interpolation times are replayed from a dedicated validation
+seed, and the caller's training RNG state is restored afterward. Source loss
+is evaluated twice and must replay within `1e-6` before a run is valid.
+
+## Development scorecard
+
+### Transport-loss gate
+
+A checkpoint may enter molecular benchmarking only if its best loss is lower
+than the untouched source model on the identical validation transitions and
+identical flow randomness. Training loss alone is never a gate.
+
+### Observed-transition gate
+
+Use fixed same-replica transitions at exactly 50 frames. For each transition,
+score persistence, the mean error of individual stochastic samples, best of
+K, the ensemble-mean endpoint, displacement-magnitude error, and displacement
+direction cosine after the same C-alpha alignment policy.
+
+Before any standard-test access, a candidate must on the validation suite:
+
+1. improve ensemble-mean endpoint RMSD over the pinned source MarS-FM model;
+2. beat persistence with a protein-level 95% bootstrap confidence interval
+   excluding zero;
+3. improve on a majority of validation proteins, not only pooled transitions;
+4. retain the result for at least three training seeds.
+
+Best-of-K is diagnostic only because it uses the unknown future to select a
+sample.
+
+### Equilibrium-ensemble retention gate
+
+On matched tree samples and MD references, report DCCM error, RMSF error and
+correlation, pairwise-distance Wasserstein distance, reference-PCA covariance
+error/RMSIP, contact-transition Wasserstein distance, radius-of-gyration
+Wasserstein distance, and clash rate. A kinetic gain cannot promote a model
+that materially degrades the source model's equilibrium or structural
+behavior. The one-protein `1e7uA04` comparison is a development pilot only;
+promotion requires a protein-level validation aggregate.
+
+## Promotion boundary
+
+Passing this protocol only freezes a candidate. The 495-domain standard
+mdCATH test split remains unopened until architecture, checkpoint, sampling
+settings, seeds, metrics, and failure policy are immutable. A field-level
+claim additionally requires matched runnable baselines, protein-level
+uncertainty, replica-noise comparisons, and independent external replication.
+

receipts/equilibrium-suite-comparison.json

@@ -0,0 +1,616 @@
+{
+  "candidate_adaptation_run_id": "mars-fm-e010-a07350-exactlag-128x32-20260825-s1-v1",
+  "claim_boundary": "validation_comparison_not_test_or_sota_evidence",
+  "comparison": {
+    "bootstrap_resamples": 100000,
+    "bootstrap_seed": 20260825,
+    "bootstrap_unit": "protein",
+    "claim_boundary": "validation_equilibrium_suite_not_test_or_sota_evidence",
+    "metrics": {
+      "dccm_mae": {
+        "fraction_proteins_within_limit": 1.0,
+        "maximum_allowed_mean_degradation": 0.005,
+        "maximum_allowed_protein_degradation": 0.01,
+        "maximum_protein_degradation": 0.001768082357,
+        "mean_degradation": -0.000659761524,
+        "mean_degradation_95ci": [
+          -0.001624722174,
+          0.000285435304
+        ],
+        "passed": true
+      },
+      "native_contact_fraction_w1": {
+        "fraction_proteins_within_limit": 1.0,
+        "maximum_allowed_mean_degradation": 0.015,
+        "maximum_allowed_protein_degradation": 0.03,
+        "maximum_protein_degradation": 0.003055483871,
+        "mean_degradation": 0.000271095576,
+        "mean_degradation_95ci": [
+          -0.001593543408,
+          0.001945598306
+        ],
+        "passed": true
+      },
+      "pairwise_rmsd_w1_angstrom": {
+        "fraction_proteins_within_limit": 1.0,
+        "maximum_allowed_mean_degradation": 0.25,
+        "maximum_allowed_protein_degradation": 0.5,
+        "maximum_protein_degradation": 0.207604998691,
+        "mean_degradation": -0.043699034661,
+        "mean_degradation_95ci": [
+          -0.12602109911,
+          0.049296262623
+        ],
+        "passed": true
+      },
+      "pca_subspace_rmsip": {
+        "fraction_proteins_within_limit": 1.0,
+        "maximum_allowed_mean_degradation": 0.03,
+        "maximum_allowed_protein_degradation": 0.06,
+        "maximum_protein_degradation": 0.025361372291,
+        "mean_degradation": 0.004161025684,
+        "mean_degradation_95ci": [
+          -0.000227353353,
+          0.010754551963
+        ],
+        "passed": true
+      },
+      "prediction_clash_fraction": {
+        "fraction_proteins_within_limit": 1.0,
+        "maximum_allowed_mean_degradation": 0.0005,
+        "maximum_allowed_protein_degradation": 0.001,
+        "maximum_protein_degradation": 5.754386e-05,
+        "mean_degradation": -1.19536e-06,
+        "mean_degradation_95ci": [
+          -2.4638117e-05,
+          2.2484179e-05
+        ],
+        "passed": true
+      },
+      "radius_gyration_w1_angstrom": {
+        "fraction_proteins_within_limit": 1.0,
+        "maximum_allowed_mean_degradation": 0.15,
+        "maximum_allowed_protein_degradation": 0.3,
+        "maximum_protein_degradation": 0.031926049996,
+        "mean_degradation": -0.04215244792,
+        "mean_degradation_95ci": [
+          -0.077537352172,
+          -0.008042088849
+        ],
+        "passed": true
+      },
+      "rmsf_rmse_angstrom": {
+        "fraction_proteins_within_limit": 1.0,
+        "maximum_allowed_mean_degradation": 0.15,
+        "maximum_allowed_protein_degradation": 0.3,
+        "maximum_protein_degradation": 0.147062281582,
+        "mean_degradation": -0.022244200546,
+        "mean_degradation_95ci": [
+          -0.07759563382,
+          0.039555613954
+        ],
+        "passed": true
+      }
+    },
+    "passed": true,
+    "per_protein": [
+      {
+        "claim_boundary": "validation_retention_gate_not_test_or_sota_evidence",
+        "domain_id": "1d1rA00",
+        "metrics": {
+          "dccm_mae": {
+            "candidate": 0.15482688202531125,
+            "degradation": -0.000343473113,
+            "maximum_allowed_degradation": 0.005,
+            "passed": true,
+            "preferred_direction": "lower",
+            "source": 0.1551703551379965
+          },
+          "native_contact_fraction_w1": {
+            "candidate": 0.24524811017914466,
+            "degradation": 0.002894170032,
+            "maximum_allowed_degradation": 0.015,
+            "passed": true,
+            "preferred_direction": "lower",
+            "source": 0.2423539401470436
+          },
+          "pairwise_rmsd_w1_angstrom": {
+            "candidate": 6.40159036459235,
+            "degradation": 0.085398879573,
+            "maximum_allowed_degradation": 0.25,
+            "passed": true,
+            "preferred_direction": "lower",
+            "source": 6.316191485019427
+          },
+          "pca_subspace_rmsip": {
+            "candidate": 0.6821113516317411,
+            "degradation": 0.025361372291,
+            "maximum_allowed_degradation": 0.03,
+            "passed": true,
+            "preferred_direction": "higher",
+            "source": 0.7074727239227397
+          },
+          "prediction_clash_fraction": {
+            "candidate": 0.0022818428184281842,
+            "degradation": -5.1791629e-05,
+            "maximum_allowed_degradation": 0.0005,
+            "passed": true,
+            "preferred_direction": "lower",
+            "source": 0.002333634447455586
+          },
+          "radius_gyration_w1_angstrom": {
+            "candidate": 2.288233543617155,
+            "degradation": 0.012698668973,
+            "maximum_allowed_degradation": 0.15,
+            "passed": true,
+            "preferred_direction": "lower",
+            "source": 2.275534874643861
+          },
+          "rmsf_rmse_angstrom": {
+            "candidate": 5.077326957273818,
+            "degradation": 0.055022865248,
+            "maximum_allowed_degradation": 0.15,
+            "passed": true,
+            "preferred_direction": "lower",
+            "source": 5.022304092025405
+          }
+        },
+        "passed": true,
+        "test_records_opened": 0
+      },
+      {
+        "claim_boundary": "validation_retention_gate_not_test_or_sota_evidence",
+        "domain_id": "1e7uA04",
+        "metrics": {
+          "dccm_mae": {
+            "candidate": 0.0738950699689714,
+            "degradation": 0.001768082357,
+            "maximum_allowed_degradation": 0.005,
+            "passed": true,
+            "preferred_direction": "lower",
+            "source": 0.07212698761166055
+          },
+          "native_contact_fraction_w1": {
+            "candidate": 0.05295915207812313,
+            "degradation": 0.002907466952,
+            "maximum_allowed_degradation": 0.015,
+            "passed": true,
+            "preferred_direction": "lower",
+            "source": 0.050051685125640094
+          },
+          "pairwise_rmsd_w1_angstrom": {
+            "candidate": 1.7234786502233603,
+            "degradation": 0.207604998691,
+            "maximum_allowed_degradation": 0.25,
+            "passed": true,
+            "preferred_direction": "lower",
+            "source": 1.5158736515321163
+          },
+          "pca_subspace_rmsip": {
+            "candidate": 0.7409942065226867,
+            "degradation": -0.004018200837,
+            "maximum_allowed_degradation": 0.03,
+            "passed": true,
+            "preferred_direction": "higher",
+            "source": 0.736976005685482
+          },
+          "prediction_clash_fraction": {
+            "candidate": 0.002067777233382329,
+            "degradation": 1.8455006e-05,
+            "maximum_allowed_degradation": 0.0005,
+            "passed": true,
+            "preferred_direction": "lower",
+            "source": 0.002049322227666177
+          },
+          "radius_gyration_w1_angstrom": {
+            "candidate": 0.675267147009806,
+            "degradation": 0.031926049996,
+            "maximum_allowed_degradation": 0.15,
+            "passed": true,
+            "preferred_direction": "lower",
+            "source": 0.6433410970140768
+          },
+          "rmsf_rmse_angstrom": {
+            "candidate": 1.6840770833212975,
+            "degradation": 0.147062281582,
+            "maximum_allowed_degradation": 0.15,
+            "passed": true,
+            "preferred_direction": "lower",
+            "source": 1.5370148017397278
+          }
+        },
+        "passed": true,
+        "test_records_opened": 0
+      },
+      {
+        "claim_boundary": "validation_retention_gate_not_test_or_sota_evidence",
+        "domain_id": "1q16B03",
+        "metrics": {
+          "dccm_mae": {
+            "candidate": 0.053265546720762176,
+            "degradation": -0.003260047765,
+            "maximum_allowed_degradation": 0.005,
+            "passed": true,
+            "preferred_direction": "lower",
+            "source": 0.05652559448591472
+          },
+          "native_contact_fraction_w1": {
+            "candidate": 0.007023326572008125,
+            "degradation": -0.00037275706,
+            "maximum_allowed_degradation": 0.015,
+            "passed": true,
+            "preferred_direction": "lower",
+            "source": 0.007396083632391957
+          },
+          "pairwise_rmsd_w1_angstrom": {
+            "candidate": 1.3921825402100483,
+            "degradation": -0.123076031814,
+            "maximum_allowed_degradation": 0.25,
+            "passed": true,
+            "preferred_direction": "lower",
+            "source": 1.51525857202419
+          },
+          "pca_subspace_rmsip": {
+            "candidate": 0.8643176226646919,
+            "degradation": 0.001992735188,
+            "maximum_allowed_degradation": 0.03,
+            "passed": true,
+            "preferred_direction": "higher",
+            "source": 0.8663103578523526
+          },
+          "prediction_clash_fraction": {
+            "candidate": 0.0026449122807017545,
+            "degradation": 5.754386e-05,
+            "maximum_allowed_degradation": 0.0005,
+            "passed": true,
+            "preferred_direction": "lower",
+            "source": 0.0025873684210526316
+          },
+          "radius_gyration_w1_angstrom": {
+            "candidate": 1.2570520348810872,
+            "degradation": -0.126941573376,
+            "maximum_allowed_degradation": 0.15,
+            "passed": true,
+            "preferred_direction": "lower",
+            "source": 1.3839936082574305
+          },
+          "rmsf_rmse_angstrom": {
+            "candidate": 1.0680318039610732,
+            "degradation": -0.096485409347,
+            "maximum_allowed_degradation": 0.15,
+            "passed": true,
+            "preferred_direction": "lower",
+            "source": 1.1645172133082644
+          }
+        },
+        "passed": true,
+        "test_records_opened": 0
+      },
+      {
+        "claim_boundary": "validation_retention_gate_not_test_or_sota_evidence",
+        "domain_id": "1rquA01",
+        "metrics": {
+          "dccm_mae": {
+            "candidate": 0.030216005339397722,
+            "degradation": -0.001355403615,
+            "maximum_allowed_degradation": 0.005,
+            "passed": true,
+            "preferred_direction": "lower",
+            "source": 0.03157140895404707
+          },
+          "native_contact_fraction_w1": {
+            "candidate": 0.004222943722943736,
+            "degradation": -9.0909091e-05,
+            "maximum_allowed_degradation": 0.015,
+            "passed": true,
+            "preferred_direction": "lower",
+            "source": 0.004313852813852831
+          },
+          "pairwise_rmsd_w1_angstrom": {
+            "candidate": 0.6491749819309589,
+            "degradation": -0.017173560786,
+            "maximum_allowed_degradation": 0.25,
+            "passed": true,
+            "preferred_direction": "lower",
+            "source": 0.666348542716951
+          },
+          "pca_subspace_rmsip": {
+            "candidate": 0.9599878497108804,
+            "degradation": 0.003566240917,
+            "maximum_allowed_degradation": 0.03,
+            "passed": true,
+            "preferred_direction": "higher",
+            "source": 0.9635540906276152
+          },
+          "prediction_clash_fraction": {
+            "candidate": 0.0024329411764705882,
+            "degradation": -4.2352941e-05,
+            "maximum_allowed_degradation": 0.0005,
+            "passed": true,
+            "preferred_direction": "lower",
+            "source": 0.002475294117647059
+          },
+          "radius_gyration_w1_angstrom": {
+            "candidate": 0.7715122275935271,
+            "degradation": -0.064242733704,
+            "maximum_allowed_degradation": 0.15,
+            "passed": true,
+            "preferred_direction": "lower",
+            "source": 0.8357549612975174
+          },
+          "rmsf_rmse_angstrom": {
+            "candidate": 0.5261811583355112,
+            "degradation": -0.028756594432,
+            "maximum_allowed_degradation": 0.15,
+            "passed": true,
+            "preferred_direction": "lower",
+            "source": 0.5549377527679228
+          }
+        },
+        "passed": true,
+        "test_records_opened": 0
+      },
+      {
+        "claim_boundary": "validation_retention_gate_not_test_or_sota_evidence",
+        "domain_id": "2dp9A01",
+        "metrics": {
+          "dccm_mae": {
+            "candidate": 0.0969025883946428,
+            "degradation": 0.000567905859,
+            "maximum_allowed_degradation": 0.005,
+            "passed": true,
+            "preferred_direction": "lower",
+            "source": 0.09633468253613099
+          },
+          "native_contact_fraction_w1": {
+            "candidate": 0.11129813240677547,
+            "degradation": -0.001910529255,
+            "maximum_allowed_degradation": 0.015,
+            "passed": true,
+            "preferred_direction": "lower",
+            "source": 0.11320866166159957
+          },
+          "pairwise_rmsd_w1_angstrom": {
+            "candidate": 1.5752615766371485,
+            "degradation": -0.050518150252,
+            "maximum_allowed_degradation": 0.25,
+            "passed": true,
+            "preferred_direction": "lower",
+            "source": 1.625779726888966
+          },
+          "pca_subspace_rmsip": {
+            "candidate": 0.7561045693885684,
+            "degradation": 0.003853423393,
+            "maximum_allowed_degradation": 0.03,
+            "passed": true,
+            "preferred_direction": "higher",
+            "source": 0.7599579927814428
+          },
+          "prediction_clash_fraction": {
+            "candidate": 0.0019220779220779222,
+            "degradation": -2.4324881e-05,
+            "maximum_allowed_degradation": 0.0005,
+            "passed": true,
+            "preferred_direction": "lower",
+            "source": 0.0019464028035456606
+          },
+          "radius_gyration_w1_angstrom": {
+            "candidate": 0.26246381110195427,
+            "degradation": 0.003846185211,
+            "maximum_allowed_degradation": 0.15,
+            "passed": true,
+            "preferred_direction": "lower",
+            "source": 0.25861762589138254
+          },
+          "rmsf_rmse_angstrom": {
+            "candidate": 1.1443145000062587,
+            "degradation": -0.011688106824,
+            "maximum_allowed_degradation": 0.15,
+            "passed": true,
+            "preferred_direction": "lower",
+            "source": 1.1560026068300373
+          }
+        },
+        "passed": true,
+        "test_records_opened": 0
+      },
+      {
+        "claim_boundary": "validation_retention_gate_not_test_or_sota_evidence",
+        "domain_id": "2f2cA02",
+        "metrics": {
+          "dccm_mae": {
+            "candidate": 0.05804454311556474,
+            "degradation": -0.000519282801,
+            "maximum_allowed_degradation": 0.005,
+            "passed": true,
+            "preferred_direction": "lower",
+            "source": 0.05856382591643404
+          },
+          "native_contact_fraction_w1": {
+            "candidate": 0.044334999999999986,
+            "degradation": 0.000525,
+            "maximum_allowed_degradation": 0.015,
+            "passed": true,
+            "preferred_direction": "lower",
+            "source": 0.04380999999999999
+          },
+          "pairwise_rmsd_w1_angstrom": {
+            "candidate": 0.46429456601829555,
+            "degradation": -0.096452375141,
+            "maximum_allowed_degradation": 0.25,
+            "passed": true,
+            "preferred_direction": "lower",
+            "source": 0.5607469411588364
+          },
+          "pca_subspace_rmsip": {
+            "candidate": 0.8402476658310097,
+            "degradation": 0.002448361066,
+            "maximum_allowed_degradation": 0.03,
+            "passed": true,
+            "preferred_direction": "higher",
+            "source": 0.8426960268968243
+          },
+          "prediction_clash_fraction": {
+            "candidate": 0.002286813186813187,
+            "degradation": 2.3443223e-05,
+            "maximum_allowed_degradation": 0.0005,
+            "passed": true,
+            "preferred_direction": "lower",
+            "source": 0.0022633699633699635
+          },
+          "radius_gyration_w1_angstrom": {
+            "candidate": 0.47719223951019296,
+            "degradation": -0.069887401157,
+            "maximum_allowed_degradation": 0.15,
+            "passed": true,
+            "preferred_direction": "lower",
+            "source": 0.5470796406671098
+          },
+          "rmsf_rmse_angstrom": {
+            "candidate": 1.2610296796524307,
+            "degradation": -0.033795235512,
+            "maximum_allowed_degradation": 0.15,
+            "passed": true,
+            "preferred_direction": "lower",
+            "source": 1.2948249151639777
+          }
+        },
+        "passed": true,
+        "test_records_opened": 0
+      },
+      {
+        "claim_boundary": "validation_retention_gate_not_test_or_sota_evidence",
+        "domain_id": "2k4qA00",
+        "metrics": {
+          "dccm_mae": {
+            "candidate": 0.08346559114199999,
+            "degradation": -0.000941171831,
+            "maximum_allowed_degradation": 0.005,
+            "passed": true,
+            "preferred_direction": "lower",
+            "source": 0.08440676297266357
+          },
+          "native_contact_fraction_w1": {
+            "candidate": 0.07879300699300698,
+            "degradation": -0.004839160839,
+            "maximum_allowed_degradation": 0.015,
+            "passed": true,
+            "preferred_direction": "lower",
+            "source": 0.08363216783216781
+          },
+          "pairwise_rmsd_w1_angstrom": {
+            "candidate": 1.3449775622982791,
+            "degradation": -0.223476841546,
+            "maximum_allowed_degradation": 0.25,
+            "passed": true,
+            "preferred_direction": "lower",
+            "source": 1.5684544038439028
+          },
+          "pca_subspace_rmsip": {
+            "candidate": 0.8026720708814058,
+            "degradation": 0.000304196004,
+            "maximum_allowed_degradation": 0.03,
+            "passed": true,
+            "preferred_direction": "higher",
+            "source": 0.802976266885178
+          },
+          "prediction_clash_fraction": {
+            "candidate": 0.0018865521575198994,
+            "degradation": -8.37872e-07,
+            "maximum_allowed_degradation": 0.0005,
+            "passed": true,
+            "preferred_direction": "lower",
+            "source": 0.0018873900293255132
+          },
+          "radius_gyration_w1_angstrom": {
+            "candidate": 0.8941144484858697,
+            "degradation": -0.063311253212,
+            "maximum_allowed_degradation": 0.15,
+            "passed": true,
+            "preferred_direction": "lower",
+            "source": 0.9574257016976239
+          },
+          "rmsf_rmse_angstrom": {
+            "candidate": 1.4991205013517446,
+            "degradation": -0.146117205874,
+            "maximum_allowed_degradation": 0.15,
+            "passed": true,
+            "preferred_direction": "lower",
+            "source": 1.645237707226197
+          }
+        },
+        "passed": true,
+        "test_records_opened": 0
+      },
+      {
+        "claim_boundary": "validation_retention_gate_not_test_or_sota_evidence",
+        "domain_id": "5k2mA01",
+        "metrics": {
+          "dccm_mae": {
+            "candidate": 0.07823249656049976,
+            "degradation": -0.001194701284,
+            "maximum_allowed_degradation": 0.005,
+            "passed": true,
+            "preferred_direction": "lower",
+            "source": 0.07942719784484631
+          },
+          "native_contact_fraction_w1": {
+            "candidate": 0.05564129032258064,
+            "degradation": 0.003055483871,
+            "maximum_allowed_degradation": 0.015,
+            "passed": true,
+            "preferred_direction": "lower",
+            "source": 0.0525858064516129
+          },
+          "pairwise_rmsd_w1_angstrom": {
+            "candidate": 0.6455713530814601,
+            "degradation": -0.13189919601,
+            "maximum_allowed_degradation": 0.25,
+            "passed": true,
+            "preferred_direction": "lower",
+            "source": 0.7774705490911311
+          },
+          "pca_subspace_rmsip": {
+            "candidate": 0.7562751919299002,
+            "degradation": -0.000219922551,
+            "maximum_allowed_degradation": 0.03,
+            "passed": true,
+            "preferred_direction": "higher",
+            "source": 0.7560552693785566
+          },
+          "prediction_clash_fraction": {
+            "candidate": 0.0023435610302351625,
+            "degradation": 1.0302352e-05,
+            "maximum_allowed_degradation": 0.0005,
+            "passed": true,
+            "preferred_direction": "lower",
+            "source": 0.002333258678611422
+          },
+          "radius_gyration_w1_angstrom": {
+            "candidate": 0.915345146943571,
+            "degradation": -0.061307526094,
+            "maximum_allowed_degradation": 0.15,
+            "passed": true,
+            "preferred_direction": "lower",
+            "source": 0.9766526730372651
+          },
+          "rmsf_rmse_angstrom": {
+            "candidate": 0.943766604396447,
+            "degradation": -0.063196199208,
+            "maximum_allowed_degradation": 0.15,
+            "passed": true,
+            "preferred_direction": "lower",
+            "source": 1.0069628036043834
+          }
+        },
+        "passed": true,
+        "test_records_opened": 0
+      }
+    ],
+    "proteins": 8,
+    "test_records_opened": 0
+  },
+  "schema_version": "dynaview-mars-fm-equilibrium-suite-comparison-v1",
+  "source_checkpoint_identity": "1e957cd4b9f1a054ceb0c2104f40ba6f3cc0d8d0",
+  "test_records_opened": 0
+}

receipts/full-transition-comparison.json

@@ -0,0 +1,284 @@
+{
+  "candidate_adaptation_run_id": "mars-fm-e010-a07350-exactlag-128x32-20260825-s1-v1",
+  "claim_boundary": "validation_comparison_not_test_or_sota_evidence",
+  "comparison": {
+    "best_of_k_rmsd_improvement_vs_source_angstrom": 0.03724134251807005,
+    "bootstrap_resamples": 100000,
+    "bootstrap_seed": 20260825,
+    "bootstrap_unit": "protein",
+    "direction_cosine_improvement_vs_source": 0.0013599728514569792,
+    "displacement_magnitude_mae_improvement_vs_source_angstrom": 0.029187859054032816,
+    "ensemble_mean_improvement_vs_persistence_95ci_angstrom": [
+      0.20302960346670967,
+      0.6806461308413764
+    ],
+    "ensemble_mean_improvement_vs_persistence_angstrom": 0.4364052826817221,
+    "ensemble_mean_improvement_vs_source_95ci_angstrom": [
+      0.02525014950148637,
+      0.040114625364512255
+    ],
+    "ensemble_mean_improvement_vs_source_angstrom": 0.0325922111581212,
+    "ensemble_mean_rmsd_improvement_vs_source_angstrom": 0.0325922111581212,
+    "fraction_proteins_better_than_persistence": 0.7857142857142857,
+    "fraction_proteins_better_than_source": 0.9642857142857143,
+    "per_protein": [
+      {
+        "best_of_k_rmsd_improvement_vs_source_angstrom": 0.01646343856639909,
+        "direction_cosine_improvement_vs_source": 0.0023076805253925414,
+        "displacement_magnitude_mae_improvement_vs_source_angstrom": 0.016322871227970825,
+        "domain_id": "1d1rA00",
+        "ensemble_mean_improvement_vs_persistence_angstrom": 0.9537240495659791,
+        "ensemble_mean_rmsd_improvement_vs_source_angstrom": 0.0217256570667832,
+        "single_sample_rmsd_improvement_vs_source_angstrom": 0.03951161703407102
+      },
+      {
+        "best_of_k_rmsd_improvement_vs_source_angstrom": 0.059424884910713516,
+        "direction_cosine_improvement_vs_source": 0.0026543662306548477,
+        "displacement_magnitude_mae_improvement_vs_source_angstrom": 0.011610491113109411,
+        "domain_id": "1e7uA04",
+        "ensemble_mean_improvement_vs_persistence_angstrom": 0.5149769102211295,
+        "ensemble_mean_rmsd_improvement_vs_source_angstrom": 0.06635897186434292,
+        "single_sample_rmsd_improvement_vs_source_angstrom": 0.11754119743891422
+      },
+      {
+        "best_of_k_rmsd_improvement_vs_source_angstrom": 0.04172957928759047,
+        "direction_cosine_improvement_vs_source": 0.002846304044747394,
+        "displacement_magnitude_mae_improvement_vs_source_angstrom": 0.041301447179341366,
+        "domain_id": "1h4yA00",
+        "ensemble_mean_improvement_vs_persistence_angstrom": 0.22033741973402954,
+        "ensemble_mean_rmsd_improvement_vs_source_angstrom": 0.0418509559295579,
+        "single_sample_rmsd_improvement_vs_source_angstrom": 0.07409587219493474
+      },
+      {
+        "best_of_k_rmsd_improvement_vs_source_angstrom": 0.07014363542276669,
+        "direction_cosine_improvement_vs_source": 0.003166029676787696,
+        "displacement_magnitude_mae_improvement_vs_source_angstrom": 0.051838844900964354,
+        "domain_id": "1jeyB03",
+        "ensemble_mean_improvement_vs_persistence_angstrom": 1.2044744714309843,
+        "ensemble_mean_rmsd_improvement_vs_source_angstrom": 0.08156112205738353,
+        "single_sample_rmsd_improvement_vs_source_angstrom": 0.12569450249841907
+      },
+      {
+        "best_of_k_rmsd_improvement_vs_source_angstrom": 0.04457590997010996,
+        "direction_cosine_improvement_vs_source": 0.002184110543618112,
+        "displacement_magnitude_mae_improvement_vs_source_angstrom": -0.0043289986049730445,
+        "domain_id": "1nktA02",
+        "ensemble_mean_improvement_vs_persistence_angstrom": 0.44746261411252775,
+        "ensemble_mean_rmsd_improvement_vs_source_angstrom": 0.04742557231089606,
+        "single_sample_rmsd_improvement_vs_source_angstrom": 0.09028303341979083
+      },
+      {
+        "best_of_k_rmsd_improvement_vs_source_angstrom": 0.03662634979693635,
+        "direction_cosine_improvement_vs_source": 0.0012805379688710095,
+        "displacement_magnitude_mae_improvement_vs_source_angstrom": 0.03468276666002312,
+        "domain_id": "1q16B03",
+        "ensemble_mean_improvement_vs_persistence_angstrom": 0.7257780606234068,
+        "ensemble_mean_rmsd_improvement_vs_source_angstrom": 0.03660349491889647,
+        "single_sample_rmsd_improvement_vs_source_angstrom": 0.058337339941360256
+      },
+      {
+        "best_of_k_rmsd_improvement_vs_source_angstrom": 0.013836531763145032,
+        "direction_cosine_improvement_vs_source": 0.00024276392576128014,
+        "displacement_magnitude_mae_improvement_vs_source_angstrom": -0.002192331873721365,
+        "domain_id": "1rquA01",
+        "ensemble_mean_improvement_vs_persistence_angstrom": 1.8858914144742513,
+        "ensemble_mean_rmsd_improvement_vs_source_angstrom": 0.008323592408094882,
+        "single_sample_rmsd_improvement_vs_source_angstrom": 0.026547073748048433
+      },
+      {
+        "best_of_k_rmsd_improvement_vs_source_angstrom": 0.011527843979859398,
+        "direction_cosine_improvement_vs_source": -5.03211283424676e-05,
+        "displacement_magnitude_mae_improvement_vs_source_angstrom": 0.019818897677486014,
+        "domain_id": "2dp9A01",
+        "ensemble_mean_improvement_vs_persistence_angstrom": -0.6272755183938967,
+        "ensemble_mean_rmsd_improvement_vs_source_angstrom": 0.008758476198289067,
+        "single_sample_rmsd_improvement_vs_source_angstrom": 0.036204654243105594
+      },
+      {
+        "best_of_k_rmsd_improvement_vs_source_angstrom": 0.036318539434503805,
+        "direction_cosine_improvement_vs_source": 0.001004197797784434,
+        "displacement_magnitude_mae_improvement_vs_source_angstrom": 0.03017962829367038,
+        "domain_id": "2f2cA02",
+        "ensemble_mean_improvement_vs_persistence_angstrom": 0.7645881213225749,
+        "ensemble_mean_rmsd_improvement_vs_source_angstrom": 0.031179340722999882,
+        "single_sample_rmsd_improvement_vs_source_angstrom": 0.08004612169317937
+      },
+      {
+        "best_of_k_rmsd_improvement_vs_source_angstrom": 0.027488650238129964,
+        "direction_cosine_improvement_vs_source": 0.0022487422442267757,
+        "displacement_magnitude_mae_improvement_vs_source_angstrom": 0.01013227478770995,
+        "domain_id": "2incC00",
+        "ensemble_mean_improvement_vs_persistence_angstrom": 0.4610114264775973,
+        "ensemble_mean_rmsd_improvement_vs_source_angstrom": 0.030418514688641096,
+        "single_sample_rmsd_improvement_vs_source_angstrom": 0.061988680732824264
+      },
+      {
+        "best_of_k_rmsd_improvement_vs_source_angstrom": 0.07090862131514619,
+        "direction_cosine_improvement_vs_source": 0.0009985778057123862,
+        "displacement_magnitude_mae_improvement_vs_source_angstrom": 0.06321654704817181,
+        "domain_id": "2izwC00",
+        "ensemble_mean_improvement_vs_persistence_angstrom": -0.13981274361730378,
+        "ensemble_mean_rmsd_improvement_vs_source_angstrom": 0.04142448078263605,
+        "single_sample_rmsd_improvement_vs_source_angstrom": 0.12186975015593049
+      },
+      {
+        "best_of_k_rmsd_improvement_vs_source_angstrom": 0.05547176355797667,
+        "direction_cosine_improvement_vs_source": 0.0005487120056557604,
+        "displacement_magnitude_mae_improvement_vs_source_angstrom": 0.07110735039435179,
+        "domain_id": "2ji7A03",
+        "ensemble_mean_improvement_vs_persistence_angstrom": 0.051978667255809796,
+        "ensemble_mean_rmsd_improvement_vs_source_angstrom": 0.0337787132934011,
+        "single_sample_rmsd_improvement_vs_source_angstrom": 0.09840405804953356
+      },
+      {
+        "best_of_k_rmsd_improvement_vs_source_angstrom": 0.03301877379466944,
+        "direction_cosine_improvement_vs_source": 9.640421573570146e-05,
+        "displacement_magnitude_mae_improvement_vs_source_angstrom": 0.01114392549299481,
+        "domain_id": "2k4qA00",
+        "ensemble_mean_improvement_vs_persistence_angstrom": 0.32883278099041746,
+        "ensemble_mean_rmsd_improvement_vs_source_angstrom": 0.024334332199050124,
+        "single_sample_rmsd_improvement_vs_source_angstrom": 0.06936441511294511
+      },
+      {
+        "best_of_k_rmsd_improvement_vs_source_angstrom": 0.05946884802215102,
+        "direction_cosine_improvement_vs_source": 0.0016945195278653125,
+        "displacement_magnitude_mae_improvement_vs_source_angstrom": 0.04858287356966784,
+        "domain_id": "2k54A00",
+        "ensemble_mean_improvement_vs_persistence_angstrom": -0.46020310958796884,
+        "ensemble_mean_rmsd_improvement_vs_source_angstrom": 0.040033867338310714,
+        "single_sample_rmsd_improvement_vs_source_angstrom": 0.08235325209592581
+      },
+      {
+        "best_of_k_rmsd_improvement_vs_source_angstrom": 0.028910811993508645,
+        "direction_cosine_improvement_vs_source": 0.0011774412263663714,
+        "displacement_magnitude_mae_improvement_vs_source_angstrom": 0.018033156181977983,
+        "domain_id": "2kjrA01",
+        "ensemble_mean_improvement_vs_persistence_angstrom": 1.4604137236244572,
+        "ensemble_mean_rmsd_improvement_vs_source_angstrom": 0.016205079020728874,
+        "single_sample_rmsd_improvement_vs_source_angstrom": 0.04195554998556261
+      },
+      {
+        "best_of_k_rmsd_improvement_vs_source_angstrom": 0.050691172350521896,
+        "direction_cosine_improvement_vs_source": 0.0022433936983189406,
+        "displacement_magnitude_mae_improvement_vs_source_angstrom": 0.022114787535346014,
+        "domain_id": "3aekA03",
+        "ensemble_mean_improvement_vs_persistence_angstrom": 0.5786139214024963,
+        "ensemble_mean_rmsd_improvement_vs_source_angstrom": 0.03710378408937132,
+        "single_sample_rmsd_improvement_vs_source_angstrom": 0.0664340032846038
+      },
+      {
+        "best_of_k_rmsd_improvement_vs_source_angstrom": 0.04497904644005857,
+        "direction_cosine_improvement_vs_source": 0.0023567328636533436,
+        "displacement_magnitude_mae_improvement_vs_source_angstrom": 0.035524930464325255,
+        "domain_id": "3atsA01",
+        "ensemble_mean_improvement_vs_persistence_angstrom": 0.20788321390406317,
+        "ensemble_mean_rmsd_improvement_vs_source_angstrom": 0.04148384120566462,
+        "single_sample_rmsd_improvement_vs_source_angstrom": 0.0860958046522029
+      },
+      {
+        "best_of_k_rmsd_improvement_vs_source_angstrom": 0.013697406094962616,
+        "direction_cosine_improvement_vs_source": -0.00013762585012866424,
+        "displacement_magnitude_mae_improvement_vs_source_angstrom": -0.005477332506224286,
+        "domain_id": "3d8lA00",
+        "ensemble_mean_improvement_vs_persistence_angstrom": 2.0063157112889467,
+        "ensemble_mean_rmsd_improvement_vs_source_angstrom": 0.004197499422030404,
+        "single_sample_rmsd_improvement_vs_source_angstrom": 0.03751472449189208
+      },
+      {
+        "best_of_k_rmsd_improvement_vs_source_angstrom": 0.05490254596090516,
+        "direction_cosine_improvement_vs_source": 0.002250155452739433,
+        "displacement_magnitude_mae_improvement_vs_source_angstrom": 0.10179220887019103,
+        "domain_id": "3i01A03",
+        "ensemble_mean_improvement_vs_persistence_angstrom": -0.647370447977841,
+        "ensemble_mean_rmsd_improvement_vs_source_angstrom": 0.04505940499652006,
+        "single_sample_rmsd_improvement_vs_source_angstrom": 0.10122106731273384
+      },
+      {
+        "best_of_k_rmsd_improvement_vs_source_angstrom": 0.05826094337685639,
+        "direction_cosine_improvement_vs_source": 0.00029652449148337423,
+        "displacement_magnitude_mae_improvement_vs_source_angstrom": 0.05863624999975281,
+        "domain_id": "3mg3A01",
+        "ensemble_mean_improvement_vs_persistence_angstrom": -0.3028312878629542,
+        "ensemble_mean_rmsd_improvement_vs_source_angstrom": 0.016587115498079896,
+        "single_sample_rmsd_improvement_vs_source_angstrom": 0.09502677544444538
+      },
+      {
+        "best_of_k_rmsd_improvement_vs_source_angstrom": 0.03983387710605868,
+        "direction_cosine_improvement_vs_source": 0.000730346391562775,
+        "displacement_magnitude_mae_improvement_vs_source_angstrom": 0.03342570543575407,
+        "domain_id": "3pt1A02",
+        "ensemble_mean_improvement_vs_persistence_angstrom": 0.6971612857436362,
+        "ensemble_mean_rmsd_improvement_vs_source_angstrom": 0.0493324027839499,
+        "single_sample_rmsd_improvement_vs_source_angstrom": 0.09043735897854077
+      },
+      {
+        "best_of_k_rmsd_improvement_vs_source_angstrom": 0.009444895851252966,
+        "direction_cosine_improvement_vs_source": 0.0009112996701411413,
+        "displacement_magnitude_mae_improvement_vs_source_angstrom": 0.017329693769446042,
+        "domain_id": "3tdgA01",
+        "ensemble_mean_improvement_vs_persistence_angstrom": 0.48438079640615683,
+        "ensemble_mean_rmsd_improvement_vs_source_angstrom": 0.011343461526418963,
+        "single_sample_rmsd_improvement_vs_source_angstrom": 0.0267872591475129
+      },
+      {
+        "best_of_k_rmsd_improvement_vs_source_angstrom": 0.015091725001004086,
+        "direction_cosine_improvement_vs_source": 0.001749461987328449,
+        "displacement_magnitude_mae_improvement_vs_source_angstrom": 0.013485576196493154,
+        "domain_id": "4b96A00",
+        "ensemble_mean_improvement_vs_persistence_angstrom": 0.27100829259371206,
+        "ensemble_mean_rmsd_improvement_vs_source_angstrom": 0.01762705665974007,
+        "single_sample_rmsd_improvement_vs_source_angstrom": 0.039824214715261874
+      },
+      {
+        "best_of_k_rmsd_improvement_vs_source_angstrom": 0.005297614946439211,
+        "direction_cosine_improvement_vs_source": -0.001143902539429753,
+        "displacement_magnitude_mae_improvement_vs_source_angstrom": 0.0005222666785464902,
+        "domain_id": "4deqA02",
+        "ensemble_mean_improvement_vs_persistence_angstrom": 0.6473776731505758,
+        "ensemble_mean_rmsd_improvement_vs_source_angstrom": -0.0048515754505995545,
+        "single_sample_rmsd_improvement_vs_source_angstrom": 0.0062746228583989705
+      },
+      {
+        "best_of_k_rmsd_improvement_vs_source_angstrom": -0.002997467814806143,
+        "direction_cosine_improvement_vs_source": 0.0003445721453008699,
+        "displacement_magnitude_mae_improvement_vs_source_angstrom": -0.007371030205435947,
+        "domain_id": "4nreA01",
+        "ensemble_mean_improvement_vs_persistence_angstrom": 0.31957326632080196,
+        "ensemble_mean_rmsd_improvement_vs_source_angstrom": 0.007202589613928723,
+        "single_sample_rmsd_improvement_vs_source_angstrom": 0.02279038285857382
+      },
+      {
+        "best_of_k_rmsd_improvement_vs_source_angstrom": 0.0706757992718785,
+        "direction_cosine_improvement_vs_source": 0.0014261487311752474,
+        "displacement_magnitude_mae_improvement_vs_source_angstrom": 0.04737084883148501,
+        "domain_id": "4pkiG02",
+        "ensemble_mean_improvement_vs_persistence_angstrom": 0.11107006735801939,
+        "ensemble_mean_rmsd_improvement_vs_source_angstrom": 0.051415757478777024,
+        "single_sample_rmsd_improvement_vs_source_angstrom": 0.11893457676300834
+      },
+      {
+        "best_of_k_rmsd_improvement_vs_source_angstrom": 0.02496120742352481,
+        "direction_cosine_improvement_vs_source": 0.0022999471197731047,
+        "displacement_magnitude_mae_improvement_vs_source_angstrom": 0.04598881868845961,
+        "domain_id": "4qncB00",
+        "ensemble_mean_improvement_vs_persistence_angstrom": -0.1476554018994971,
+        "ensemble_mean_rmsd_improvement_vs_source_angstrom": 0.0626686097356064,
+        "single_sample_rmsd_improvement_vs_source_angstrom": 0.0454132602121593
+      },
+      {
+        "best_of_k_rmsd_improvement_vs_source_angstrom": 0.0520046424436984,
+        "direction_cosine_improvement_vs_source": 0.002352119068040004,
+        "displacement_magnitude_mae_improvement_vs_source_angstrom": 0.03246758570603436,
+        "domain_id": "5k2mA01",
+        "ensemble_mean_improvement_vs_persistence_angstrom": 0.20164253642610674,
+        "ensemble_mean_rmsd_improvement_vs_source_angstrom": 0.04342979406789382,
+        "single_sample_rmsd_improvement_vs_source_angstrom": 0.06816632978683401
+      }
+    ],
+    "proteins": 28,
+    "single_sample_rmsd_improvement_vs_source_angstrom": 0.06889705353038263
+  },
+  "schema_version": "dynaview-mars-fm-exact-lag-suite-comparison-v1",
+  "source_checkpoint_identity": "1e957cd4b9f1a054ceb0c2104f40ba6f3cc0d8d0",
+  "test_records_opened": 0
+}

receipts/head-to-head-vs-previous-senba.json

@@ -0,0 +1,284 @@
+{
+  "candidate_adaptation_run_id": "mars-fm-e010-a07350-exactlag-128x32-20260825-s1-v1",
+  "claim_boundary": "validation_comparison_not_test_or_sota_evidence",
+  "comparison": {
+    "best_of_k_rmsd_improvement_vs_source_angstrom": 0.002203602852091345,
+    "bootstrap_resamples": 100000,
+    "bootstrap_seed": 20260825,
+    "bootstrap_unit": "protein",
+    "direction_cosine_improvement_vs_source": 0.00031542212283628454,
+    "displacement_magnitude_mae_improvement_vs_source_angstrom": 0.0017132670015839815,
+    "ensemble_mean_improvement_vs_persistence_95ci_angstrom": [
+      0.20302960346670967,
+      0.6806461308413764
+    ],
+    "ensemble_mean_improvement_vs_persistence_angstrom": 0.4364052826817221,
+    "ensemble_mean_improvement_vs_source_95ci_angstrom": [
+      0.0008567748166488626,
+      0.0034470447723082133
+    ],
+    "ensemble_mean_improvement_vs_source_angstrom": 0.00216273137725704,
+    "ensemble_mean_rmsd_improvement_vs_source_angstrom": 0.00216273137725704,
+    "fraction_proteins_better_than_persistence": 0.7857142857142857,
+    "fraction_proteins_better_than_source": 0.7857142857142857,
+    "per_protein": [
+      {
+        "best_of_k_rmsd_improvement_vs_source_angstrom": 0.0006785778092410766,
+        "direction_cosine_improvement_vs_source": 0.0002993262239820438,
+        "displacement_magnitude_mae_improvement_vs_source_angstrom": 0.002646677662391017,
+        "domain_id": "1d1rA00",
+        "ensemble_mean_improvement_vs_persistence_angstrom": 0.9537240495659791,
+        "ensemble_mean_rmsd_improvement_vs_source_angstrom": 0.0008050506502961241,
+        "single_sample_rmsd_improvement_vs_source_angstrom": -0.001427499129412979
+      },
+      {
+        "best_of_k_rmsd_improvement_vs_source_angstrom": 0.008536098296413996,
+        "direction_cosine_improvement_vs_source": 0.0006249384468047792,
+        "displacement_magnitude_mae_improvement_vs_source_angstrom": 0.0009249568649503104,
+        "domain_id": "1e7uA04",
+        "ensemble_mean_improvement_vs_persistence_angstrom": 0.5149769102211295,
+        "ensemble_mean_rmsd_improvement_vs_source_angstrom": 0.007822896565652826,
+        "single_sample_rmsd_improvement_vs_source_angstrom": 0.014963149879591242
+      },
+      {
+        "best_of_k_rmsd_improvement_vs_source_angstrom": 0.001631685651380721,
+        "direction_cosine_improvement_vs_source": 0.0003551876703023482,
+        "displacement_magnitude_mae_improvement_vs_source_angstrom": 0.00239515056673989,
+        "domain_id": "1h4yA00",
+        "ensemble_mean_improvement_vs_persistence_angstrom": 0.22033741973402954,
+        "ensemble_mean_rmsd_improvement_vs_source_angstrom": 0.002630855706875934,
+        "single_sample_rmsd_improvement_vs_source_angstrom": 0.004541147721052852
+      },
+      {
+        "best_of_k_rmsd_improvement_vs_source_angstrom": 0.010435189880567464,
+        "direction_cosine_improvement_vs_source": 0.0003776889447406617,
+        "displacement_magnitude_mae_improvement_vs_source_angstrom": 0.006782602181466757,
+        "domain_id": "1jeyB03",
+        "ensemble_mean_improvement_vs_persistence_angstrom": 1.2044744714309843,
+        "ensemble_mean_rmsd_improvement_vs_source_angstrom": 0.008852667441143325,
+        "single_sample_rmsd_improvement_vs_source_angstrom": 0.01458084865376108
+      },
+      {
+        "best_of_k_rmsd_improvement_vs_source_angstrom": 0.008408357798330712,
+        "direction_cosine_improvement_vs_source": 0.0006075578164836948,
+        "displacement_magnitude_mae_improvement_vs_source_angstrom": 0.0009957319559803501,
+        "domain_id": "1nktA02",
+        "ensemble_mean_improvement_vs_persistence_angstrom": 0.44746261411252775,
+        "ensemble_mean_rmsd_improvement_vs_source_angstrom": 0.008032838923188734,
+        "single_sample_rmsd_improvement_vs_source_angstrom": 0.010277157430774508
+      },
+      {
+        "best_of_k_rmsd_improvement_vs_source_angstrom": 0.0044042275243789675,
+        "direction_cosine_improvement_vs_source": 0.0006619484248290286,
+        "displacement_magnitude_mae_improvement_vs_source_angstrom": 0.00479673986355289,
+        "domain_id": "1q16B03",
+        "ensemble_mean_improvement_vs_persistence_angstrom": 0.7257780606234068,
+        "ensemble_mean_rmsd_improvement_vs_source_angstrom": 0.007218884144174709,
+        "single_sample_rmsd_improvement_vs_source_angstrom": 0.008520653654500165
+      },
+      {
+        "best_of_k_rmsd_improvement_vs_source_angstrom": 0.0034132866535561135,
+        "direction_cosine_improvement_vs_source": 0.0002211525032043138,
+        "displacement_magnitude_mae_improvement_vs_source_angstrom": 0.0023920425513326293,
+        "domain_id": "1rquA01",
+        "ensemble_mean_improvement_vs_persistence_angstrom": 1.8858914144742513,
+        "ensemble_mean_rmsd_improvement_vs_source_angstrom": 0.0032131526333465388,
+        "single_sample_rmsd_improvement_vs_source_angstrom": 0.005400757589313798
+      },
+      {
+        "best_of_k_rmsd_improvement_vs_source_angstrom": -0.004939055741098208,
+        "direction_cosine_improvement_vs_source": -7.18377050291652e-05,
+        "displacement_magnitude_mae_improvement_vs_source_angstrom": -0.007474717067504688,
+        "domain_id": "2dp9A01",
+        "ensemble_mean_improvement_vs_persistence_angstrom": -0.6272755183938967,
+        "ensemble_mean_rmsd_improvement_vs_source_angstrom": -0.00471624337370713,
+        "single_sample_rmsd_improvement_vs_source_angstrom": -0.005000387800890849
+      },
+      {
+        "best_of_k_rmsd_improvement_vs_source_angstrom": 0.0026886262320342524,
+        "direction_cosine_improvement_vs_source": 0.00046674352693365506,
+        "displacement_magnitude_mae_improvement_vs_source_angstrom": 0.0026163037987427806,
+        "domain_id": "2f2cA02",
+        "ensemble_mean_improvement_vs_persistence_angstrom": 0.7645881213225749,
+        "ensemble_mean_rmsd_improvement_vs_source_angstrom": 0.005100551362795969,
+        "single_sample_rmsd_improvement_vs_source_angstrom": 0.007259712735745438
+      },
+      {
+        "best_of_k_rmsd_improvement_vs_source_angstrom": 0.0014735429150700696,
+        "direction_cosine_improvement_vs_source": -1.6987974106430492e-05,
+        "displacement_magnitude_mae_improvement_vs_source_angstrom": -0.0005273012751021433,
+        "domain_id": "2incC00",
+        "ensemble_mean_improvement_vs_persistence_angstrom": 0.4610114264775973,
+        "ensemble_mean_rmsd_improvement_vs_source_angstrom": -0.0013825947753112189,
+        "single_sample_rmsd_improvement_vs_source_angstrom": 0.0010851136470950706
+      },
+      {
+        "best_of_k_rmsd_improvement_vs_source_angstrom": 0.004618185039049649,
+        "direction_cosine_improvement_vs_source": 0.00022458160722271803,
+        "displacement_magnitude_mae_improvement_vs_source_angstrom": 0.002944538827348353,
+        "domain_id": "2izwC00",
+        "ensemble_mean_improvement_vs_persistence_angstrom": -0.13981274361730378,
+        "ensemble_mean_rmsd_improvement_vs_source_angstrom": 0.000908701305505133,
+        "single_sample_rmsd_improvement_vs_source_angstrom": 0.008094738753165132
+      },
+      {
+        "best_of_k_rmsd_improvement_vs_source_angstrom": 0.004690081539150626,
+        "direction_cosine_improvement_vs_source": 0.00027002374019979847,
+        "displacement_magnitude_mae_improvement_vs_source_angstrom": 0.0038744868310840452,
+        "domain_id": "2ji7A03",
+        "ensemble_mean_improvement_vs_persistence_angstrom": 0.051978667255809796,
+        "ensemble_mean_rmsd_improvement_vs_source_angstrom": 0.0019738157747113405,
+        "single_sample_rmsd_improvement_vs_source_angstrom": 0.009367578036330926
+      },
+      {
+        "best_of_k_rmsd_improvement_vs_source_angstrom": 0.002254093837734672,
+        "direction_cosine_improvement_vs_source": 0.00024184954588191143,
+        "displacement_magnitude_mae_improvement_vs_source_angstrom": 0.002039409201823439,
+        "domain_id": "2k4qA00",
+        "ensemble_mean_improvement_vs_persistence_angstrom": 0.32883278099041746,
+        "ensemble_mean_rmsd_improvement_vs_source_angstrom": 0.002613130698424726,
+        "single_sample_rmsd_improvement_vs_source_angstrom": 0.006066778150380259
+      },
+      {
+        "best_of_k_rmsd_improvement_vs_source_angstrom": -0.003391002322111092,
+        "direction_cosine_improvement_vs_source": 0.00040376653071927393,
+        "displacement_magnitude_mae_improvement_vs_source_angstrom": -0.000742828194239209,
+        "domain_id": "2k54A00",
+        "ensemble_mean_improvement_vs_persistence_angstrom": -0.46020310958796884,
+        "ensemble_mean_rmsd_improvement_vs_source_angstrom": 0.00028833541840889154,
+        "single_sample_rmsd_improvement_vs_source_angstrom": 0.0010953910690503221
+      },
+      {
+        "best_of_k_rmsd_improvement_vs_source_angstrom": -0.0028422551199938084,
+        "direction_cosine_improvement_vs_source": 0.00013719243815440985,
+        "displacement_magnitude_mae_improvement_vs_source_angstrom": 0.001100068283368838,
+        "domain_id": "2kjrA01",
+        "ensemble_mean_improvement_vs_persistence_angstrom": 1.4604137236244572,
+        "ensemble_mean_rmsd_improvement_vs_source_angstrom": -0.0012405314421162217,
+        "single_sample_rmsd_improvement_vs_source_angstrom": 1.6873812349160744e-05
+      },
+      {
+        "best_of_k_rmsd_improvement_vs_source_angstrom": -0.0006394626098664702,
+        "direction_cosine_improvement_vs_source": 0.0004587659683807921,
+        "displacement_magnitude_mae_improvement_vs_source_angstrom": 0.0011039961008028065,
+        "domain_id": "3aekA03",
+        "ensemble_mean_improvement_vs_persistence_angstrom": 0.5786139214024963,
+        "ensemble_mean_rmsd_improvement_vs_source_angstrom": 0.0026328318889170532,
+        "single_sample_rmsd_improvement_vs_source_angstrom": 0.002256060844173291
+      },
+      {
+        "best_of_k_rmsd_improvement_vs_source_angstrom": 0.002274048944386209,
+        "direction_cosine_improvement_vs_source": 0.0004879876252218973,
+        "displacement_magnitude_mae_improvement_vs_source_angstrom": 0.0049290344078136705,
+        "domain_id": "3atsA01",
+        "ensemble_mean_improvement_vs_persistence_angstrom": 0.20788321390406317,
+        "ensemble_mean_rmsd_improvement_vs_source_angstrom": 0.0030849113918467452,
+        "single_sample_rmsd_improvement_vs_source_angstrom": 0.004409217652737496
+      },
+      {
+        "best_of_k_rmsd_improvement_vs_source_angstrom": 0.002083973073787604,
+        "direction_cosine_improvement_vs_source": 0.0001422769015084202,
+        "displacement_magnitude_mae_improvement_vs_source_angstrom": 0.0008420164080931158,
+        "domain_id": "3d8lA00",
+        "ensemble_mean_improvement_vs_persistence_angstrom": 2.0063157112889467,
+        "ensemble_mean_rmsd_improvement_vs_source_angstrom": 0.001536007970695863,
+        "single_sample_rmsd_improvement_vs_source_angstrom": 0.004381875167545957
+      },
+      {
+        "best_of_k_rmsd_improvement_vs_source_angstrom": 0.006087507168309969,
+        "direction_cosine_improvement_vs_source": 0.0005503094262716557,
+        "displacement_magnitude_mae_improvement_vs_source_angstrom": 0.008695777987576925,
+        "domain_id": "3i01A03",
+        "ensemble_mean_improvement_vs_persistence_angstrom": -0.647370447977841,
+        "ensemble_mean_rmsd_improvement_vs_source_angstrom": 0.004014637285723488,
+        "single_sample_rmsd_improvement_vs_source_angstrom": 0.010684784268560676
+      },
+      {
+        "best_of_k_rmsd_improvement_vs_source_angstrom": 0.007816674671877522,
+        "direction_cosine_improvement_vs_source": 0.00052402673518398,
+        "displacement_magnitude_mae_improvement_vs_source_angstrom": 0.0026880143766545572,
+        "domain_id": "3mg3A01",
+        "ensemble_mean_improvement_vs_persistence_angstrom": -0.3028312878629542,
+        "ensemble_mean_rmsd_improvement_vs_source_angstrom": 0.00249418589050876,
+        "single_sample_rmsd_improvement_vs_source_angstrom": 0.010427843978586537
+      },
+      {
+        "best_of_k_rmsd_improvement_vs_source_angstrom": 0.003852814324588394,
+        "direction_cosine_improvement_vs_source": 0.0003274225748183257,
+        "displacement_magnitude_mae_improvement_vs_source_angstrom": 0.003002600232765662,
+        "domain_id": "3pt1A02",
+        "ensemble_mean_improvement_vs_persistence_angstrom": 0.6971612857436362,
+        "ensemble_mean_rmsd_improvement_vs_source_angstrom": 0.004758955800639342,
+        "single_sample_rmsd_improvement_vs_source_angstrom": 0.008898808253889001
+      },
+      {
+        "best_of_k_rmsd_improvement_vs_source_angstrom": -0.00415646349399168,
+        "direction_cosine_improvement_vs_source": -0.0002966016217157974,
+        "displacement_magnitude_mae_improvement_vs_source_angstrom": -0.004025043599632294,
+        "domain_id": "3tdgA01",
+        "ensemble_mean_improvement_vs_persistence_angstrom": 0.48438079640615683,
+        "ensemble_mean_rmsd_improvement_vs_source_angstrom": -0.006059125194271431,
+        "single_sample_rmsd_improvement_vs_source_angstrom": -0.00811735792116508
+      },
+      {
+        "best_of_k_rmsd_improvement_vs_source_angstrom": -0.0011907149408427742,
+        "direction_cosine_improvement_vs_source": 0.000505883491922765,
+        "displacement_magnitude_mae_improvement_vs_source_angstrom": -0.0004697086602298839,
+        "domain_id": "4b96A00",
+        "ensemble_mean_improvement_vs_persistence_angstrom": 0.27100829259371206,
+        "ensemble_mean_rmsd_improvement_vs_source_angstrom": 0.0004107570848308484,
+        "single_sample_rmsd_improvement_vs_source_angstrom": -0.0006578234498135771
+      },
+      {
+        "best_of_k_rmsd_improvement_vs_source_angstrom": -0.005616419462839417,
+        "direction_cosine_improvement_vs_source": 3.423014781078759e-05,
+        "displacement_magnitude_mae_improvement_vs_source_angstrom": 0.0019134151453981119,
+        "domain_id": "4deqA02",
+        "ensemble_mean_improvement_vs_persistence_angstrom": 0.6473776731505758,
+        "ensemble_mean_rmsd_improvement_vs_source_angstrom": -0.0011615737905632528,
+        "single_sample_rmsd_improvement_vs_source_angstrom": -0.0038480792455537483
+      },
+      {
+        "best_of_k_rmsd_improvement_vs_source_angstrom": -0.0016432280572065139,
+        "direction_cosine_improvement_vs_source": 0.0002699097766834524,
+        "displacement_magnitude_mae_improvement_vs_source_angstrom": -0.001425850638268944,
+        "domain_id": "4nreA01",
+        "ensemble_mean_improvement_vs_persistence_angstrom": 0.31957326632080196,
+        "ensemble_mean_rmsd_improvement_vs_source_angstrom": 0.00017752350699318242,
+        "single_sample_rmsd_improvement_vs_source_angstrom": -0.00292232864859443
+      },
+      {
+        "best_of_k_rmsd_improvement_vs_source_angstrom": 0.009997843482297597,
+        "direction_cosine_improvement_vs_source": 0.0003801411148291467,
+        "displacement_magnitude_mae_improvement_vs_source_angstrom": 0.0012432784091056526,
+        "domain_id": "4pkiG02",
+        "ensemble_mean_improvement_vs_persistence_angstrom": 0.11107006735801939,
+        "ensemble_mean_rmsd_improvement_vs_source_angstrom": 0.004888569947910426,
+        "single_sample_rmsd_improvement_vs_source_angstrom": 0.013191271406777005
+      },
+      {
+        "best_of_k_rmsd_improvement_vs_source_angstrom": -0.0005199002795901464,
+        "direction_cosine_improvement_vs_source": 0.0002491926766371999,
+        "displacement_magnitude_mae_improvement_vs_source_angstrom": 0.0030215865642047035,
+        "domain_id": "4qncB00",
+        "ensemble_mean_improvement_vs_persistence_angstrom": -0.1476554018994971,
+        "ensemble_mean_rmsd_improvement_vs_source_angstrom": -0.0009444544986543235,
+        "single_sample_rmsd_improvement_vs_source_angstrom": 0.004683197692969543
+      },
+      {
+        "best_of_k_rmsd_improvement_vs_source_angstrom": 0.001294567043942152,
+        "direction_cosine_improvement_vs_source": 0.0003951428815403002,
+        "displacement_magnitude_mae_improvement_vs_source_angstrom": 0.0016884972581321378,
+        "domain_id": "5k2mA01",
+        "ensemble_mean_improvement_vs_persistence_angstrom": 0.20164253642610674,
+        "ensemble_mean_rmsd_improvement_vs_source_angstrom": 0.0026017402452307437,
+        "single_sample_rmsd_improvement_vs_source_angstrom": 0.0029315910766793962
+      }
+    ],
+    "proteins": 28,
+    "single_sample_rmsd_improvement_vs_source_angstrom": 0.004684324117128507
+  },
+  "schema_version": "dynaview-mars-fm-exact-lag-suite-comparison-v1",
+  "source_checkpoint_identity": "mars-fm-mdcath-anchor08-exactlag-128x32-20260801-s1-v1",
+  "test_records_opened": 0
+}

receipts/interpolation-config.json

@@ -0,0 +1 @@
+{"alpha":0.0735,"parent_run_id":"mars-fm-epoch-sweep-e010-exactlag-128x32-20260825-s1-v1","run_id":"mars-fm-e010-a07350-exactlag-128x32-20260825-s1-v1","source_checkpoint_sha256":"sha256:f56794cc5d1a8e5263031944f932384dd785fe945553612fe76fe1e9191fa879","test_records_opened":0}

receipts/interpolation-result.json

@@ -0,0 +1 @@
+{"ablation_hash":"sha256:163ab0043d86b5a2731e1b1a233af27fb069049897d657fd5c39bd54b1343b3f","ablation_track":"128x32","adaptation_method":"source_checkpoint_weight_interpolation","best_checkpoint":"training/mars-fm-e010-a07350-exactlag-128x32-20260825-s1-v1/checkpoints/best.ckpt","best_checkpoint_sha256":"sha256:f02dcf6a97848c834e3c0be7d848c1068d1d0eb7976d8d6bcfbc98d4114e9c7c","best_checkpoint_size_bytes":136711098,"claim_boundary":"validation_source_anchored_candidate_not_test_or_sota_evidence","interpolation_alpha":0.0735,"interpolation_elapsed_seconds":1.336162495,"interpolation_parent_checkpoint_sha256":"sha256:1bd5c20b5f3aa8587dce6440af6580f0c2def98d0b868a39969ef854bd729a25","interpolation_parent_run_id":"mars-fm-epoch-sweep-e010-exactlag-128x32-20260825-s1-v1","last_checkpoint":"training/mars-fm-e010-a07350-exactlag-128x32-20260825-s1-v1/checkpoints/best.ckpt","last_checkpoint_sha256":"sha256:f02dcf6a97848c834e3c0be7d848c1068d1d0eb7976d8d6bcfbc98d4114e9c7c","last_checkpoint_size_bytes":136711098,"mars_repository_revision":"4fc17d86cd7e22dda6353f2f1f6bf4b40c9ce946","parent_best_validation_improvement":0.03134322166442871,"run_id":"mars-fm-e010-a07350-exactlag-128x32-20260825-s1-v1","schema_version":"dynaview-mars-fm-adaptation-training-result-v1","source_checkpoint_sha256":"sha256:f56794cc5d1a8e5263031944f932384dd785fe945553612fe76fe1e9191fa879","test_records_opened":0}

receipts/parent/training-config.json

@@ -0,0 +1 @@
+{"ablation_hash":"sha256:163ab0043d86b5a2731e1b1a233af27fb069049897d657fd5c39bd54b1343b3f","ablation_track":"128x32","batch_size":8,"claim_boundary":"train_validation_adaptation_only","epochs":10,"exact_lag_fraction":1.0,"learning_rate":3e-06,"optimizer_initialization":"fresh_adam","resume_optimizer_state":false,"run_id":"mars-fm-epoch-sweep-e010-exactlag-128x32-20260825-s1-v1","schema_version":"dynaview-mars-fm-adaptation-run-spec-v1","seed":20260825,"source_checkpoint_sha256":"sha256:f56794cc5d1a8e5263031944f932384dd785fe945553612fe76fe1e9191fa879","source_epoch_counter_resumed":false,"split_receipt":{"ablation_hash":"sha256:163ab0043d86b5a2731e1b1a233af27fb069049897d657fd5c39bd54b1343b3f","ablation_track":"128x32","claim_boundary":"train_validation_adaptation_splits_only","roles":{"train":{"count":128,"path":"baselines/mars_fm/adaptation_data/128x32/train.csv","sha256":"sha256:ce2c0f3408e4934161535b519552d72f27c7e4d8973cb9fa55a4cebce014bba4","size_bytes":17659},"val":{"count":32,"path":"baselines/mars_fm/adaptation_data/128x32/val.csv","sha256":"sha256:311704d26216300bc6606546e6699a3a3efce676497ed5177a639fde1175f799","size_bytes":4822}},"schema_version":"dynaview-mars-fm-adaptation-splits-v1","standard_manifest_hash":"sha256:53e81a90c0dcf3e00a814744f62d25db52f5aa8790475294dac451d36d728b8b","test_records_opened":0},"test_records_opened":0,"train_batches":0,"train_domains":128,"trainable_scope":"all","transition_lag_frames":50,"transition_sampling":"exact_lag","validation_batches":0,"validation_domains":32,"validation_flow_seed":21260828,"validation_randomness":"paired_and_replayed"}

receipts/parent/training-result.json

@@ -0,0 +1 @@
+{"ablation_hash":"sha256:163ab0043d86b5a2731e1b1a233af27fb069049897d657fd5c39bd54b1343b3f","ablation_track":"128x32","batch_size":8,"best_checkpoint":"training/mars-fm-epoch-sweep-e010-exactlag-128x32-20260825-s1-v1/checkpoints/best.ckpt","best_checkpoint_sha256":"sha256:1bd5c20b5f3aa8587dce6440af6580f0c2def98d0b868a39969ef854bd729a25","best_checkpoint_size_bytes":136727442,"best_validation_improvement":0.03134322166442871,"best_validation_loss":1.6931358575820923,"claim_boundary":"train_validation_adaptation_only_not_test_or_sota_evidence","epochs":10,"exact_lag_fraction":1.0,"final_callback_metrics":{"train/loss":1.7669486999511719,"val/loss":1.6931358575820923},"fixed_validation_transition_population_sha256":"sha256:bf8856eda4a2a72fb82e3c8dda6fccd26135d3b17acb6db07992796339a19e15","last_checkpoint":"training/mars-fm-epoch-sweep-e010-exactlag-128x32-20260825-s1-v1/checkpoints/last.ckpt","last_checkpoint_sha256":"sha256:1bd5c20b5f3aa8587dce6440af6580f0c2def98d0b868a39969ef854bd729a25","last_checkpoint_size_bytes":136727442,"learning_rate":3e-06,"mars_repository_revision":"4fc17d86cd7e22dda6353f2f1f6bf4b40c9ce946","max_cuda_memory_bytes":64684237824,"optimizer_initialization":"fresh_adam","parameter_count":34152521,"repeated_source_validation_loss":1.724479079246521,"resume_optimizer_state":false,"run_id":"mars-fm-epoch-sweep-e010-exactlag-128x32-20260825-s1-v1","schema_version":"dynaview-mars-fm-adaptation-training-result-v1","seed":20260825,"source_checkpoint_sha256":"sha256:f56794cc5d1a8e5263031944f932384dd785fe945553612fe76fe1e9191fa879","source_epoch_counter_resumed":false,"source_validation_loss":1.724479079246521,"source_validation_replay_absolute_error":0.0,"split_receipt_sha256":"sha256:8afe1912883b1e2030e43127ae68c49540d51a14de0aa1418237a9d2bd902f6d","test_records_opened":0,"train_batches":0,"train_domains":128,"trainable_parameter_count":34152521,"trainable_scope":"all","training_elapsed_seconds":121.79760014299998,"transition_lag_frames":50,"transition_sampling":"exact_lag","validation_batches":0,"validation_domains":32,"validation_flow_seed":21260828,"validation_history":[{"epoch":0,"validation_loss":1.7130178213119507},{"epoch":1,"validation_loss":1.707754135131836},{"epoch":2,"validation_loss":1.7043360471725464},{"epoch":3,"validation_loss":1.7015782594680786},{"epoch":4,"validation_loss":1.69961416721344},{"epoch":5,"validation_loss":1.6989654302597046},{"epoch":6,"validation_loss":1.6973251104354858},{"epoch":7,"validation_loss":1.695460319519043},{"epoch":8,"validation_loss":1.6943145990371704},{"epoch":9,"validation_loss":1.6931358575820923}],"validation_randomness":"paired_and_replayed"}

receipts/replications/seed2/compact-transition-comparison.json

@@ -0,0 +1,104 @@
+{
+  "candidate_adaptation_run_id": "mars-fm-e010-a07350-exactlag-128x32-20260826-s2-v1",
+  "claim_boundary": "validation_comparison_not_test_or_sota_evidence",
+  "comparison": {
+    "best_of_k_rmsd_improvement_vs_source_angstrom": 0.02399681045633617,
+    "bootstrap_resamples": 100000,
+    "bootstrap_seed": 20260826,
+    "bootstrap_unit": "protein",
+    "direction_cosine_improvement_vs_source": 0.001034606180841044,
+    "displacement_magnitude_mae_improvement_vs_source_angstrom": 0.01742144458609146,
+    "ensemble_mean_improvement_vs_persistence_95ci_angstrom": [
+      0.10598212322871658,
+      1.0117286611348109
+    ],
+    "ensemble_mean_improvement_vs_persistence_angstrom": 0.5729732270971675,
+    "ensemble_mean_improvement_vs_source_95ci_angstrom": [
+      0.013268226510653514,
+      0.034294909639703275
+    ],
+    "ensemble_mean_improvement_vs_source_angstrom": 0.022905611767840894,
+    "ensemble_mean_rmsd_improvement_vs_source_angstrom": 0.022905611767840894,
+    "fraction_proteins_better_than_persistence": 0.75,
+    "fraction_proteins_better_than_source": 1.0,
+    "per_protein": [
+      {
+        "best_of_k_rmsd_improvement_vs_source_angstrom": -0.0008694710315904075,
+        "direction_cosine_improvement_vs_source": 0.001918252697266587,
+        "displacement_magnitude_mae_improvement_vs_source_angstrom": -0.0029419518339022233,
+        "domain_id": "1d1rA00",
+        "ensemble_mean_improvement_vs_persistence_angstrom": 0.950610844040142,
+        "ensemble_mean_rmsd_improvement_vs_source_angstrom": 0.021374124533442895,
+        "single_sample_rmsd_improvement_vs_source_angstrom": 0.03324448982969841
+      },
+      {
+        "best_of_k_rmsd_improvement_vs_source_angstrom": 0.06168167157157356,
+        "direction_cosine_improvement_vs_source": 0.002410559126106826,
+        "displacement_magnitude_mae_improvement_vs_source_angstrom": 0.046758928152479395,
+        "domain_id": "1e7uA04",
+        "ensemble_mean_improvement_vs_persistence_angstrom": -0.28791989974639876,
+        "ensemble_mean_rmsd_improvement_vs_source_angstrom": 0.05298996493052499,
+        "single_sample_rmsd_improvement_vs_source_angstrom": 0.10255997231742242
+      },
+      {
+        "best_of_k_rmsd_improvement_vs_source_angstrom": 0.031974860285679085,
+        "direction_cosine_improvement_vs_source": 0.002046900010626407,
+        "displacement_magnitude_mae_improvement_vs_source_angstrom": 0.021767141963866532,
+        "domain_id": "1q16B03",
+        "ensemble_mean_improvement_vs_persistence_angstrom": 1.4799616169884597,
+        "ensemble_mean_rmsd_improvement_vs_source_angstrom": 0.04316121024603348,
+        "single_sample_rmsd_improvement_vs_source_angstrom": 0.05581355803711929
+      },
+      {
+        "best_of_k_rmsd_improvement_vs_source_angstrom": 0.012715207805458206,
+        "direction_cosine_improvement_vs_source": 0.00018689100951307935,
+        "displacement_magnitude_mae_improvement_vs_source_angstrom": 0.012691674691646782,
+        "domain_id": "1rquA01",
+        "ensemble_mean_improvement_vs_persistence_angstrom": 1.2044382003048426,
+        "ensemble_mean_rmsd_improvement_vs_source_angstrom": 0.009539505320024233,
+        "single_sample_rmsd_improvement_vs_source_angstrom": 0.027734242948016075
+      },
+      {
+        "best_of_k_rmsd_improvement_vs_source_angstrom": 0.004118743835924477,
+        "direction_cosine_improvement_vs_source": 0.001960149882962392,
+        "displacement_magnitude_mae_improvement_vs_source_angstrom": 0.021771542078383277,
+        "domain_id": "2dp9A01",
+        "ensemble_mean_improvement_vs_persistence_angstrom": -0.5233299115380774,
+        "ensemble_mean_rmsd_improvement_vs_source_angstrom": 0.019940417693107726,
+        "single_sample_rmsd_improvement_vs_source_angstrom": 0.0339049615477105
+      },
+      {
+        "best_of_k_rmsd_improvement_vs_source_angstrom": 0.021460772584791954,
+        "direction_cosine_improvement_vs_source": -0.00027067762721477706,
+        "displacement_magnitude_mae_improvement_vs_source_angstrom": 0.024562766082155463,
+        "domain_id": "2f2cA02",
+        "ensemble_mean_improvement_vs_persistence_angstrom": 0.5992039630504458,
+        "ensemble_mean_rmsd_improvement_vs_source_angstrom": 0.014655311244951008,
+        "single_sample_rmsd_improvement_vs_source_angstrom": 0.04586298579887682
+      },
+      {
+        "best_of_k_rmsd_improvement_vs_source_angstrom": 0.02529209095375151,
+        "direction_cosine_improvement_vs_source": 0.00033254269489152444,
+        "displacement_magnitude_mae_improvement_vs_source_angstrom": 0.0007115998451308059,
+        "domain_id": "2k4qA00",
+        "ensemble_mean_improvement_vs_persistence_angstrom": 0.8579005098932964,
+        "ensemble_mean_rmsd_improvement_vs_source_angstrom": 0.012123044189367604,
+        "single_sample_rmsd_improvement_vs_source_angstrom": 0.05201049304342931
+      },
+      {
+        "best_of_k_rmsd_improvement_vs_source_angstrom": 0.03560060764510098,
+        "direction_cosine_improvement_vs_source": -0.00030776834742368697,
+        "displacement_magnitude_mae_improvement_vs_source_angstrom": 0.01404985570897166,
+        "domain_id": "5k2mA01",
+        "ensemble_mean_improvement_vs_persistence_angstrom": 0.3029204937846295,
+        "ensemble_mean_rmsd_improvement_vs_source_angstrom": 0.00946131598527522,
+        "single_sample_rmsd_improvement_vs_source_angstrom": 0.04177296456846413
+      }
+    ],
+    "proteins": 8,
+    "single_sample_rmsd_improvement_vs_source_angstrom": 0.04911295851134212
+  },
+  "schema_version": "dynaview-mars-fm-exact-lag-suite-comparison-v1",
+  "source_checkpoint_identity": "1e957cd4b9f1a054ceb0c2104f40ba6f3cc0d8d0",
+  "test_records_opened": 0
+}

receipts/replications/seed2/interpolation-result.json

@@ -0,0 +1 @@
+{"ablation_hash":"sha256:163ab0043d86b5a2731e1b1a233af27fb069049897d657fd5c39bd54b1343b3f","ablation_track":"128x32","adaptation_method":"source_checkpoint_weight_interpolation","best_checkpoint":"training/mars-fm-e010-a07350-exactlag-128x32-20260826-s2-v1/checkpoints/best.ckpt","best_checkpoint_sha256":"sha256:2989729c979840bc915fde80ebdbb7b5f4953b1169f7877248522014cb7d651a","best_checkpoint_size_bytes":136711098,"claim_boundary":"validation_source_anchored_candidate_not_test_or_sota_evidence","interpolation_alpha":0.0735,"interpolation_elapsed_seconds":1.1135783679999998,"interpolation_parent_checkpoint_sha256":"sha256:c3bc7d728b5b09b6b8665cffdbdfc6c2b60028a5c41bbd685daaf97e41b8ab14","interpolation_parent_run_id":"mars-fm-e010-repl-exactlag-128x32-20260826-s2-v1","last_checkpoint":"training/mars-fm-e010-a07350-exactlag-128x32-20260826-s2-v1/checkpoints/best.ckpt","last_checkpoint_sha256":"sha256:2989729c979840bc915fde80ebdbb7b5f4953b1169f7877248522014cb7d651a","last_checkpoint_size_bytes":136711098,"mars_repository_revision":"4fc17d86cd7e22dda6353f2f1f6bf4b40c9ce946","parent_best_validation_improvement":0.054322242736816406,"run_id":"mars-fm-e010-a07350-exactlag-128x32-20260826-s2-v1","schema_version":"dynaview-mars-fm-adaptation-training-result-v1","source_checkpoint_sha256":"sha256:f56794cc5d1a8e5263031944f932384dd785fe945553612fe76fe1e9191fa879","test_records_opened":0}

receipts/replications/seed2/parent-result.json

@@ -0,0 +1 @@
+{"ablation_hash":"sha256:163ab0043d86b5a2731e1b1a233af27fb069049897d657fd5c39bd54b1343b3f","ablation_track":"128x32","batch_size":8,"best_checkpoint":"training/mars-fm-e010-repl-exactlag-128x32-20260826-s2-v1/checkpoints/best.ckpt","best_checkpoint_sha256":"sha256:c3bc7d728b5b09b6b8665cffdbdfc6c2b60028a5c41bbd685daaf97e41b8ab14","best_checkpoint_size_bytes":136727442,"best_validation_improvement":0.054322242736816406,"best_validation_loss":1.7257102727890015,"claim_boundary":"train_validation_adaptation_only_not_test_or_sota_evidence","epochs":10,"exact_lag_fraction":1.0,"final_callback_metrics":{"train/loss":2.8650994300842285,"val/loss":1.7257102727890015},"fixed_validation_transition_population_sha256":"sha256:02c595bdb4867950a12b833ed30e8f201c87b14078d1505888d461b2881d3f29","last_checkpoint":"training/mars-fm-e010-repl-exactlag-128x32-20260826-s2-v1/checkpoints/last.ckpt","last_checkpoint_sha256":"sha256:c3bc7d728b5b09b6b8665cffdbdfc6c2b60028a5c41bbd685daaf97e41b8ab14","last_checkpoint_size_bytes":136727442,"learning_rate":3e-06,"mars_repository_revision":"4fc17d86cd7e22dda6353f2f1f6bf4b40c9ce946","max_cuda_memory_bytes":64684237824,"optimizer_initialization":"fresh_adam","parameter_count":34152521,"repeated_source_validation_loss":1.7800325155258179,"resume_optimizer_state":false,"run_id":"mars-fm-e010-repl-exactlag-128x32-20260826-s2-v1","schema_version":"dynaview-mars-fm-adaptation-training-result-v1","seed":20260826,"source_checkpoint_sha256":"sha256:f56794cc5d1a8e5263031944f932384dd785fe945553612fe76fe1e9191fa879","source_epoch_counter_resumed":false,"source_validation_loss":1.7800325155258179,"source_validation_replay_absolute_error":0.0,"split_receipt_sha256":"sha256:8afe1912883b1e2030e43127ae68c49540d51a14de0aa1418237a9d2bd902f6d","test_records_opened":0,"train_batches":0,"train_domains":128,"trainable_parameter_count":34152521,"trainable_scope":"all","training_elapsed_seconds":125.98805148599999,"transition_lag_frames":50,"transition_sampling":"exact_lag","validation_batches":0,"validation_domains":32,"validation_flow_seed":21260829,"validation_history":[{"epoch":0,"validation_loss":1.7621299028396606},{"epoch":1,"validation_loss":1.753602147102356},{"epoch":2,"validation_loss":1.7482267618179321},{"epoch":3,"validation_loss":1.743962287902832},{"epoch":4,"validation_loss":1.7389475107192993},{"epoch":5,"validation_loss":1.7350735664367676},{"epoch":6,"validation_loss":1.7322125434875488},{"epoch":7,"validation_loss":1.730080008506775},{"epoch":8,"validation_loss":1.7274216413497925},{"epoch":9,"validation_loss":1.7257102727890015}],"validation_randomness":"paired_and_replayed"}

receipts/replications/seed3/compact-transition-comparison.json

@@ -0,0 +1,104 @@
+{
+  "candidate_adaptation_run_id": "mars-fm-e010-a07350-exactlag-128x32-20260827-s3-v1",
+  "claim_boundary": "validation_comparison_not_test_or_sota_evidence",
+  "comparison": {
+    "best_of_k_rmsd_improvement_vs_source_angstrom": 0.03265513870177583,
+    "bootstrap_resamples": 100000,
+    "bootstrap_seed": 20260827,
+    "bootstrap_unit": "protein",
+    "direction_cosine_improvement_vs_source": 0.0012009501987395171,
+    "displacement_magnitude_mae_improvement_vs_source_angstrom": 0.029808345153726268,
+    "ensemble_mean_improvement_vs_persistence_95ci_angstrom": [
+      0.24623704931391632,
+      1.3283872719106213
+    ],
+    "ensemble_mean_improvement_vs_persistence_angstrom": 0.7592305062020892,
+    "ensemble_mean_improvement_vs_source_95ci_angstrom": [
+      0.01834818349385936,
+      0.038582393840849424
+    ],
+    "ensemble_mean_improvement_vs_source_angstrom": 0.029012580049456393,
+    "ensemble_mean_rmsd_improvement_vs_source_angstrom": 0.029012580049456393,
+    "fraction_proteins_better_than_persistence": 0.875,
+    "fraction_proteins_better_than_source": 1.0,
+    "per_protein": [
+      {
+        "best_of_k_rmsd_improvement_vs_source_angstrom": 0.01921723029970579,
+        "direction_cosine_improvement_vs_source": 0.0013604111102288874,
+        "displacement_magnitude_mae_improvement_vs_source_angstrom": -0.00041034829667774275,
+        "domain_id": "1d1rA00",
+        "ensemble_mean_improvement_vs_persistence_angstrom": 0.9717345589304287,
+        "ensemble_mean_rmsd_improvement_vs_source_angstrom": 0.01664366887661295,
+        "single_sample_rmsd_improvement_vs_source_angstrom": 0.03485632986522269
+      },
+      {
+        "best_of_k_rmsd_improvement_vs_source_angstrom": 0.032583939449395416,
+        "direction_cosine_improvement_vs_source": 0.0011998338200834846,
+        "displacement_magnitude_mae_improvement_vs_source_angstrom": 0.05330692934103354,
+        "domain_id": "1e7uA04",
+        "ensemble_mean_improvement_vs_persistence_angstrom": 0.3570836917065421,
+        "ensemble_mean_rmsd_improvement_vs_source_angstrom": 0.0407681305555645,
+        "single_sample_rmsd_improvement_vs_source_angstrom": 0.09766544013473677
+      },
+      {
+        "best_of_k_rmsd_improvement_vs_source_angstrom": 0.03177574820822748,
+        "direction_cosine_improvement_vs_source": 0.0016955534497122837,
+        "displacement_magnitude_mae_improvement_vs_source_angstrom": 0.028718329995466974,
+        "domain_id": "1q16B03",
+        "ensemble_mean_improvement_vs_persistence_angstrom": 1.6310801948658256,
+        "ensemble_mean_rmsd_improvement_vs_source_angstrom": 0.03932070991751857,
+        "single_sample_rmsd_improvement_vs_source_angstrom": 0.06809330087664733
+      },
+      {
+        "best_of_k_rmsd_improvement_vs_source_angstrom": 0.0025001240399848967,
+        "direction_cosine_improvement_vs_source": -0.0009554111097714868,
+        "displacement_magnitude_mae_improvement_vs_source_angstrom": 0.01765874386209898,
+        "domain_id": "1rquA01",
+        "ensemble_mean_improvement_vs_persistence_angstrom": 2.1871347248664534,
+        "ensemble_mean_rmsd_improvement_vs_source_angstrom": 0.0005443498683685988,
+        "single_sample_rmsd_improvement_vs_source_angstrom": 0.019244098277658495
+      },
+      {
+        "best_of_k_rmsd_improvement_vs_source_angstrom": 0.020662162381801075,
+        "direction_cosine_improvement_vs_source": 0.0009058806930637986,
+        "displacement_magnitude_mae_improvement_vs_source_angstrom": 0.028104229525728552,
+        "domain_id": "2dp9A01",
+        "ensemble_mean_improvement_vs_persistence_angstrom": -0.39076498806906557,
+        "ensemble_mean_rmsd_improvement_vs_source_angstrom": 0.022575114995732726,
+        "single_sample_rmsd_improvement_vs_source_angstrom": 0.046863434966580186
+      },
+      {
+        "best_of_k_rmsd_improvement_vs_source_angstrom": 0.0876791618003665,
+        "direction_cosine_improvement_vs_source": 0.0016582268590448757,
+        "displacement_magnitude_mae_improvement_vs_source_angstrom": 0.06424743380299702,
+        "domain_id": "2f2cA02",
+        "ensemble_mean_improvement_vs_persistence_angstrom": 0.13539797852797086,
+        "ensemble_mean_rmsd_improvement_vs_source_angstrom": 0.04844871967104858,
+        "single_sample_rmsd_improvement_vs_source_angstrom": 0.09289941545821279
+      },
+      {
+        "best_of_k_rmsd_improvement_vs_source_angstrom": 0.03674204942535475,
+        "direction_cosine_improvement_vs_source": 0.0011215370140124081,
+        "displacement_magnitude_mae_improvement_vs_source_angstrom": 0.017421877465952917,
+        "domain_id": "2k4qA00",
+        "ensemble_mean_improvement_vs_persistence_angstrom": 0.8428801934645023,
+        "ensemble_mean_rmsd_improvement_vs_source_angstrom": 0.02569358306975289,
+        "single_sample_rmsd_improvement_vs_source_angstrom": 0.07478952117848436
+      },
+      {
+        "best_of_k_rmsd_improvement_vs_source_angstrom": 0.030080694009370745,
+        "direction_cosine_improvement_vs_source": 0.002621569753541886,
+        "displacement_magnitude_mae_improvement_vs_source_angstrom": 0.0294195655332099,
+        "domain_id": "5k2mA01",
+        "ensemble_mean_improvement_vs_persistence_angstrom": 0.3392976953240563,
+        "ensemble_mean_rmsd_improvement_vs_source_angstrom": 0.03810636344105234,
+        "single_sample_rmsd_improvement_vs_source_angstrom": 0.06607787337544835
+      }
+    ],
+    "proteins": 8,
+    "single_sample_rmsd_improvement_vs_source_angstrom": 0.06256117676662387
+  },
+  "schema_version": "dynaview-mars-fm-exact-lag-suite-comparison-v1",
+  "source_checkpoint_identity": "1e957cd4b9f1a054ceb0c2104f40ba6f3cc0d8d0",
+  "test_records_opened": 0
+}

receipts/replications/seed3/interpolation-result.json

@@ -0,0 +1 @@
+{"ablation_hash":"sha256:163ab0043d86b5a2731e1b1a233af27fb069049897d657fd5c39bd54b1343b3f","ablation_track":"128x32","adaptation_method":"source_checkpoint_weight_interpolation","best_checkpoint":"training/mars-fm-e010-a07350-exactlag-128x32-20260827-s3-v1/checkpoints/best.ckpt","best_checkpoint_sha256":"sha256:6b97920739aec292acb1b1b9e23c731ac1cdf51f582984acc7a045588ba94e40","best_checkpoint_size_bytes":136711098,"claim_boundary":"validation_source_anchored_candidate_not_test_or_sota_evidence","interpolation_alpha":0.0735,"interpolation_elapsed_seconds":1.5678450369999997,"interpolation_parent_checkpoint_sha256":"sha256:727bbb1e078ccd083ddf3867a9475934d411e23888b12c39140181dd8e70e5a2","interpolation_parent_run_id":"mars-fm-e010-repl-exactlag-128x32-20260827-s3-v1","last_checkpoint":"training/mars-fm-e010-a07350-exactlag-128x32-20260827-s3-v1/checkpoints/best.ckpt","last_checkpoint_sha256":"sha256:6b97920739aec292acb1b1b9e23c731ac1cdf51f582984acc7a045588ba94e40","last_checkpoint_size_bytes":136711098,"mars_repository_revision":"4fc17d86cd7e22dda6353f2f1f6bf4b40c9ce946","parent_best_validation_improvement":0.04138755798339844,"run_id":"mars-fm-e010-a07350-exactlag-128x32-20260827-s3-v1","schema_version":"dynaview-mars-fm-adaptation-training-result-v1","source_checkpoint_sha256":"sha256:f56794cc5d1a8e5263031944f932384dd785fe945553612fe76fe1e9191fa879","test_records_opened":0}

receipts/replications/seed3/parent-result.json

@@ -0,0 +1 @@
+{"ablation_hash":"sha256:163ab0043d86b5a2731e1b1a233af27fb069049897d657fd5c39bd54b1343b3f","ablation_track":"128x32","batch_size":8,"best_checkpoint":"training/mars-fm-e010-repl-exactlag-128x32-20260827-s3-v1/checkpoints/best.ckpt","best_checkpoint_sha256":"sha256:727bbb1e078ccd083ddf3867a9475934d411e23888b12c39140181dd8e70e5a2","best_checkpoint_size_bytes":136727442,"best_validation_improvement":0.04138755798339844,"best_validation_loss":1.76681649684906,"claim_boundary":"train_validation_adaptation_only_not_test_or_sota_evidence","epochs":10,"exact_lag_fraction":1.0,"final_callback_metrics":{"train/loss":1.6300885677337646,"val/loss":1.76681649684906},"fixed_validation_transition_population_sha256":"sha256:88e0a38e24c58dc81e1316cbbce067d8080a09daae3bb2700f1fcc3a782cc850","last_checkpoint":"training/mars-fm-e010-repl-exactlag-128x32-20260827-s3-v1/checkpoints/last.ckpt","last_checkpoint_sha256":"sha256:727bbb1e078ccd083ddf3867a9475934d411e23888b12c39140181dd8e70e5a2","last_checkpoint_size_bytes":136727442,"learning_rate":3e-06,"mars_repository_revision":"4fc17d86cd7e22dda6353f2f1f6bf4b40c9ce946","max_cuda_memory_bytes":64683481088,"optimizer_initialization":"fresh_adam","parameter_count":34152521,"repeated_source_validation_loss":1.8082040548324585,"resume_optimizer_state":false,"run_id":"mars-fm-e010-repl-exactlag-128x32-20260827-s3-v1","schema_version":"dynaview-mars-fm-adaptation-training-result-v1","seed":20260827,"source_checkpoint_sha256":"sha256:f56794cc5d1a8e5263031944f932384dd785fe945553612fe76fe1e9191fa879","source_epoch_counter_resumed":false,"source_validation_loss":1.8082040548324585,"source_validation_replay_absolute_error":0.0,"split_receipt_sha256":"sha256:8afe1912883b1e2030e43127ae68c49540d51a14de0aa1418237a9d2bd902f6d","test_records_opened":0,"train_batches":0,"train_domains":128,"trainable_parameter_count":34152521,"trainable_scope":"all","training_elapsed_seconds":130.42531005700002,"transition_lag_frames":50,"transition_sampling":"exact_lag","validation_batches":0,"validation_domains":32,"validation_flow_seed":21260830,"validation_history":[{"epoch":0,"validation_loss":1.795013427734375},{"epoch":1,"validation_loss":1.787873387336731},{"epoch":2,"validation_loss":1.7828435897827148},{"epoch":3,"validation_loss":1.778768539428711},{"epoch":4,"validation_loss":1.7758903503417969},{"epoch":5,"validation_loss":1.7742332220077515},{"epoch":6,"validation_loss":1.7724345922470093},{"epoch":7,"validation_loss":1.7703598737716675},{"epoch":8,"validation_loss":1.7684249877929688},{"epoch":9,"validation_loss":1.76681649684906}],"validation_randomness":"paired_and_replayed"}

receipts/selection-v2.json

@@ -0,0 +1,120 @@
+{
+  "schema_version": "dynaview-mars-fm-epoch-sweep-selection-v2",
+  "selected_name": "Senba",
+  "selected_run_id": "mars-fm-e010-a07350-exactlag-128x32-20260825-s1-v1",
+  "selected_checkpoint_sha256": "sha256:f02dcf6a97848c834e3c0be7d848c1068d1d0eb7976d8d6bcfbc98d4114e9c7c",
+  "selection_reason": "The 10-epoch parent with 7.35% source-weight interpolation retained all seven equilibrium observables, improved the endpoint ensemble mean on 27 of 28 validation proteins, and reproduced a positive gain over MarS-FM on every protein in two additional eight-protein panels.",
+  "training": {
+    "parent_run_id": "mars-fm-epoch-sweep-e010-exactlag-128x32-20260825-s1-v1",
+    "parent_checkpoint_sha256": "sha256:1bd5c20b5f3aa8587dce6440af6580f0c2def98d0b868a39969ef854bd729a25",
+    "epochs": 10,
+    "optimizer_updates": 160,
+    "sampled_transitions": 30720,
+    "batch_size_domains": 8,
+    "learning_rate": 0.000003,
+    "trainable_scope": "all",
+    "parameter_count": 34152521,
+    "trainable_parameter_count": 34152521,
+    "train_domains": 128,
+    "validation_domains": 32,
+    "transition_sampling": "exact_lag",
+    "transition_lag_frames": 50,
+    "accelerator": "one NVIDIA H100",
+    "training_elapsed_seconds": 121.79760014299998,
+    "max_cuda_memory_bytes": 64684237824,
+    "source_validation_loss": 1.724479079246521,
+    "best_validation_loss": 1.6931358575820923,
+    "best_validation_improvement": 0.03134322166442871
+  },
+  "source_weight_interpolation": {
+    "alpha": 0.0735,
+    "interpretation": "The final checkpoint keeps 7.35% of the full fine-tuning displacement from the MarS-FM source weights.",
+    "checkpoint_size_bytes": 136711098
+  },
+  "full_transition_validation": {
+    "proteins": 28,
+    "proteins_better_than_source": 27,
+    "ensemble_mean_rmsd_improvement_vs_source_angstrom": 0.0325922111581212,
+    "ensemble_mean_rmsd_improvement_vs_source_95ci_angstrom": [
+      0.02525014950148637,
+      0.040114625364512255
+    ],
+    "proteins_better_than_persistence": 22,
+    "ensemble_mean_rmsd_improvement_vs_persistence_angstrom": 0.4364052826817221,
+    "ensemble_mean_rmsd_improvement_vs_persistence_95ci_angstrom": [
+      0.20302960346670967,
+      0.6806461308413764
+    ],
+    "bootstrap_unit": "protein",
+    "bootstrap_resamples": 100000,
+    "bootstrap_seed": 20260825
+  },
+  "full_head_to_head_vs_previous_senba": {
+    "proteins": 28,
+    "proteins_better_than_previous_senba": 22,
+    "ensemble_mean_rmsd_improvement_angstrom": 0.00216273137725704,
+    "ensemble_mean_rmsd_improvement_95ci_angstrom": [
+      0.0008567748166488626,
+      0.0034470447723082133
+    ]
+  },
+  "equilibrium_retention": {
+    "proteins": 8,
+    "observables_passed": 7,
+    "observables_total": 7,
+    "passed": true,
+    "bootstrap_unit": "protein",
+    "bootstrap_resamples": 100000,
+    "bootstrap_seed": 20260825
+  },
+  "replication": {
+    "training_seeds": [
+      20260825,
+      20260826,
+      20260827
+    ],
+    "training_elapsed_seconds": [
+      121.79760014299998,
+      125.98805148599999,
+      130.42531005700002
+    ],
+    "training_elapsed_seconds_total": 378.210961686,
+    "training_elapsed_seconds_mean": 126.070320562,
+    "compact_panel_proteins_per_seed": 8,
+    "ensemble_mean_rmsd_improvement_vs_source_angstrom": [
+      0.030089207430793796,
+      0.022905611767840894,
+      0.029012580049456393
+    ],
+    "ensemble_mean_rmsd_improvement_vs_source_angstrom_mean": 0.027335799749363696,
+    "proteins_better_than_source_per_seed": [
+      8,
+      8,
+      8
+    ],
+    "matched_improvement_vs_previous_senba_angstrom": [
+      0.0030823953657768133,
+      -0.0011042861471590726,
+      0.005569173818678608
+    ],
+    "matched_improvement_vs_previous_senba_angstrom_mean": 0.0025157610124321165
+  },
+  "campaign_compute": {
+    "workspace": "daddyjj43",
+    "billing_date": "2026-08-25",
+    "observed_modal_cost_usd": 19.451563639999996,
+    "adaptation_cost_usd": 10.601034570000001,
+    "validation_cost_usd": 8.85052907,
+    "note": "This includes the epoch sweep, interrupted and rejected runs, anchor search, equilibrium validation, full-panel validation, and seed replications. It is larger than the cost of the selected training run alone."
+  },
+  "artifacts": {
+    "checkpoint": "frozen-candidate/seed1/mars-fm-e010-a07350-exactlag-128x32-20260825-s1-v1/checkpoints/best.ckpt",
+    "full_transition_comparison": "full-transition/e010-a07350-comparison.json",
+    "full_head_to_head_comparison": "head-to-head/e010-vs-old-full.json",
+    "equilibrium_comparison": "equilibrium-suite/e010-a07350/comparison.json",
+    "seed_2_replication": "e010-seed-replication/s2/comparison.json",
+    "seed_3_replication": "e010-seed-replication/s3/comparison.json"
+  },
+  "test_records_opened": 0,
+  "claim_boundary": "validation_only_not_test_or_sota_evidence"
+}

release-manifest.json

@@ -0,0 +1,121 @@
+{
+  "schema_version": "papercrane-senba-release-manifest-v1",
+  "repository": "pikachuandme1/senba",
+  "name": "Senba",
+  "run_id": "mars-fm-e010-a07350-exactlag-128x32-20260825-s1-v1",
+  "checkpoint_sha256": "sha256:f02dcf6a97848c834e3c0be7d848c1068d1d0eb7976d8d6bcfbc98d4114e9c7c",
+  "claim_boundary": "validation_only_not_test_or_sota_evidence",
+  "test_records_opened": 0,
+  "files": [
+    {
+      "path": "LICENSE",
+      "size_bytes": 1310,
+      "sha256": "sha256:d496726bd0f4f32c2ac80a1b64f511716e3877ec9ff19afb8641e291af3c1747"
+    },
+    {
+      "path": "README.md",
+      "size_bytes": 11374,
+      "sha256": "sha256:12c7c1f8e7088a4c28d99a76e5fb1d68880cc9206f56976880a3725d402f073b"
+    },
+    {
+      "path": "inference_example.py",
+      "size_bytes": 8683,
+      "sha256": "sha256:295e88d21551c455c28a0223e928257672a3f3e0b49e8713f295a77047231ae9"
+    },
+    {
+      "path": "protocols/MARS_FM_ANCHOR08_FULL_VALIDATION_V1_PROTOCOL.md",
+      "size_bytes": 4777,
+      "sha256": "sha256:779988158703632aa2e043a44f50b2f8b1112ff7a08e87d848c62cef67fea9d1"
+    },
+    {
+      "path": "protocols/MARS_FM_EPOCH_SWEEP_V1_PROTOCOL.md",
+      "size_bytes": 4773,
+      "sha256": "sha256:c922519170603aa71d94ed3f6bab9cd2c2463575d820173ea83b7d660f913a7c"
+    },
+    {
+      "path": "protocols/MARS_FM_EQUILIBRIUM_SUITE_PROTOCOL.md",
+      "size_bytes": 1605,
+      "sha256": "sha256:1fcd7674c305d43ed96e26ecc0ea0a7ca45fec2081478ff2781ee0966ecf979f"
+    },
+    {
+      "path": "protocols/MARS_FM_MDCATH_ADAPTATION_PROTOCOL.md",
+      "size_bytes": 3895,
+      "sha256": "sha256:8d5ba3e0a015d85164da7164b84876e3e742e45b5e206bb3ea1be35da89be3d8"
+    },
+    {
+      "path": "receipts/equilibrium-suite-comparison.json",
+      "size_bytes": 22201,
+      "sha256": "sha256:74b59827a85e2f868f8cc5d8301981aadeadf8a4a8c6777d7f50818975b1559f"
+    },
+    {
+      "path": "receipts/full-transition-comparison.json",
+      "size_bytes": 16425,
+      "sha256": "sha256:9a6ea0073fd7a3af50c987ecbb1b5abec365689ccdd1acd1089eb0ff270e9c54"
+    },
+    {
+      "path": "receipts/head-to-head-vs-previous-senba.json",
+      "size_bytes": 16591,
+      "sha256": "sha256:3e6cd8467a2c4ca390011efd12d846d0ffb64c8f72bf1eb4348bf07ba5f66d3e"
+    },
+    {
+      "path": "receipts/interpolation-config.json",
+      "size_bytes": 278,
+      "sha256": "sha256:f2f1a8e1f462e48e196f91d5854474f02b78cbf9b1d564abdba794f27fb05019"
+    },
+    {
+      "path": "receipts/interpolation-result.json",
+      "size_bytes": 1397,
+      "sha256": "sha256:75289e17c293a0ae60ea99d4c5512d6177f0e7123d6d1b112efdba4fa4d07682"
+    },
+    {
+      "path": "receipts/parent/training-config.json",
+      "size_bytes": 1612,
+      "sha256": "sha256:cb9228e34e3db544c5cc18fa154ce9d43e7a04e2e7fefc36af49bb5465d0e21d"
+    },
+    {
+      "path": "receipts/parent/training-result.json",
+      "size_bytes": 2656,
+      "sha256": "sha256:5bdb3b2b91e4c6138f619dcd32b065a4fd7719382bc2150b0a500aa464ef73a5"
+    },
+    {
+      "path": "receipts/replications/seed2/compact-transition-comparison.json",
+      "size_bytes": 5685,
+      "sha256": "sha256:aa059c76970a4a7048764b7761b7347bbb1d3692b3875b0852d34c3e5f39133d"
+    },
+    {
+      "path": "receipts/replications/seed2/interpolation-result.json",
+      "size_bytes": 1398,
+      "sha256": "sha256:4e51d9e2a2d94d6cba9819b795592e9161115868094c671b30ce955f8fdfd34e"
+    },
+    {
+      "path": "receipts/replications/seed2/parent-result.json",
+      "size_bytes": 2639,
+      "sha256": "sha256:1cf4b89bbce75dc9a7ca4c36ba6dbbf63cd0389b3ccb8661de834087792ffc3a"
+    },
+    {
+      "path": "receipts/replications/seed3/compact-transition-comparison.json",
+      "size_bytes": 5678,
+      "sha256": "sha256:80f9cc8e85abe3f03b5bc9df8f8fd72f7e26b9938717e4535c145d946de2c954"
+    },
+    {
+      "path": "receipts/replications/seed3/interpolation-result.json",
+      "size_bytes": 1397,
+      "sha256": "sha256:d2eba5a0475e344f9522b0010bbcd37894874bd034dcda060428e4b776eb4b47"
+    },
+    {
+      "path": "receipts/replications/seed3/parent-result.json",
+      "size_bytes": 2632,
+      "sha256": "sha256:006621399c33632fe68295039df5120a86162de4844c50e0f0eca00c1d7efb0c"
+    },
+    {
+      "path": "receipts/selection-v2.json",
+      "size_bytes": 4823,
+      "sha256": "sha256:1f1429751c9526a969aab72cfdc71b4304dc297f47af4b5884bcda63d1b0bd2b"
+    },
+    {
+      "path": "senba.ckpt",
+      "size_bytes": 136711098,
+      "sha256": "sha256:f02dcf6a97848c834e3c0be7d848c1068d1d0eb7976d8d6bcfbc98d4114e9c7c"
+    }
+  ]
+}

senba.ckpt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:f02dcf6a97848c834e3c0be7d848c1068d1d0eb7976d8d6bcfbc98d4114e9c7c
+size 136711098