NMD-VCellNeuromuscular Virtual Cell Research PlatformModel Evaluation Gateway · Evidence → perturbation → experiment → outcomeNMD = neuromuscular disorders
For Reviewers Evidence Status Claims

Evidence Release 2026.08DMD context observedDMD candidate-conditioned prediction locked

View scientific status
Evidence freeze: 3 August 2026 P59 hold · scorer failure retained · no target score NAR working package · 6 main + 31 supplementary figures Resource: v1.2.0-measured-dmd-evidence Schema: 1.1 Open platform state → Open Trust Center → Open evidence dashboard → Open claim registry → Open release status →

Model Evaluation Gateway · local V201 candidate

Admit models, metrics and generalization claims through one refusal-first contract.

Model Adapter, Model Gateway, Metric Lab and Generalization Cube now resolve against the same released task, run and evidence objects. The surface tells you what can be compared—and why the rest must stop.

Inspect adaptersOpen machine contract
Contract closure, not model execution. V200 created no inference endpoint, ran no container or weights, rescaled no score and produced no DMD prediction; V201 preserves those limits. Existing G0/G1 diagnostics retain their released boundaries; G2–G7 stay registered or locked.

Model Adapter registry

Five adapters, zero execution authorizations.

An adapter describes the interface and its released blocking state. It is not permission to rerun a model or inherit a nearby task result.

Open source status →
REJECTADAPTER:TXPERT:V1

TxPert

HISTORICAL RUN ONLY

Released state
HISTORICAL LOCAL GATE FAIL NO REPEAT
Gateway decision
REJECT UNTIL NEW PREREGISTRATION OR UNTOUCHED TRUTH
Compatible task cells
G0
Receipts
0
Claim ceiling
E0_EXECUTION_ONLY
REJECTADAPTER:STATE:V1

STATE

HISTORICAL RECEIPTS ONLY

Released state
HISTORICAL RECEIPTS MIGRATED
Gateway decision
REJECT NO ADVANCEMENT
Compatible task cells
G0
Receipts
8
Claim ceiling
E0_EXECUTION_ONLY
REJECTADAPTER:STACK:V1

Stack

FORWARD PATH SMOKE ONLY

Released state
LOCAL NVME CORE BASE FORWARD PASS NO CHECKPOINT
Gateway decision
REJECT NO TASK COMPATIBLE CHECKPOINT
Compatible task cells
None
Receipts
2
Claim ceiling
E0_EXECUTION_ONLY
REJECTADAPTER:TAHOE-X1:V1

Tahoe-x1

INPUT CONTRACT FAILED

Released state
SOURCE PINNED INPUT CONTRACT FAIL
Gateway decision
REJECT RAW COUNTS REQUIRED
Compatible task cells
None
Receipts
2
Claim ceiling
E0_SOURCE_ONLY
REJECTADAPTER:SLIM:V1

SLIM

PRIMARY SOURCE PENDING

Released state
PRIMARY SOURCE PENDING
Gateway decision
REJECT SOURCE NOT VERIFIED
Compatible task cells
None
Receipts
0
Claim ceiling
E0_SOURCE_ONLY

Model Gateway

No hosted inference endpoint; every unregistered request is rejected.

ABSTAIN_NO_CALIBRATED_DMD_MODEL_RUN. All eight admission gates must resolve inside one exact task and ModelRun identity before execution can be considered.

Inspect ModelRunsInspect tasks
  1. 01registered task contract
  2. 02source revision and licence identity
  3. 03exact input contract pass
  4. 04task-compatible weights or checkpoint
  5. 05immutable split and feature mapping
  6. 06permanent baseline comparison
  7. 07metric applicability and independent-unit contract
  8. 08receipt, uncertainty and claim-ceiling record

Metric Lab

Metrics do not collapse into one score.

A metric is comparable only within the same task, split, feature space, endpoint and independent-unit contract. Missing metrics stay locked and no global score is computed.

Open diagnostic source →
EXECUTED BOUNDEDAGGREGATE_ERROR

Aggregate error

RMSEMAE

Aggregate response vectors only; lower error does not establish direction, disease transfer or function.

Eligible task cells: G0 · G1
EXECUTED BOUNDEDRESPONSE_DIRECTION

Response direction

raw cosineresidual cosinedelta Pearsondelta Spearman

Same-context and external aggregate diagnostics remain separate and do not establish DMD transport.

Eligible task cells: G0 · G1
DIAGNOSTIC SENSITIVITY ONLYPERTURBATION_RETRIEVAL

Perturbation retrieval

PDS L1 raw

Formula-aligned aggregate diagnostic; not an official cell-eval or Virtual Cell Challenge score.

Eligible task cells: G1
EXECUTED BOUNDEDSAMPLING_UNCERTAINTY

Sampling uncertainty

paired bootstrap intervalpaired sign-flip testtarget bootstrap interval

Intervals quantify the registered benchmark unit only, not donor, disease-stage or clinical uncertainty.

Eligible task cells: G0 · G1
LOCKED CELL LEVEL OR DE TRUTH MISSINGCELL_LEVEL_DE_DISTRIBUTION

Cell-level and DE distribution

DESDEG AUPRCsingle-cell distribution distancecell-state proportion recovery

Aggregate predictions cannot be relabelled as cell-level or differential-expression truth.

Eligible task cells: none in the current release
PARTIAL DIAGNOSTIC NOT DMD CALIBRATIONCALIBRATION_ABSTENTION

Calibration and abstention

calibrationcoverageselective riskabstention quality

Historical selective-risk diagnostics do not constitute a calibrated DMD predictor.

Eligible task cells: G1
LOCKED MATCHED OUTCOME MISSINGFUNCTIONAL_OUTCOME

Functional and safety outcome

functional hit ratetoxicity miss ratereplicationsynergy classification

No directly measured candidate-conditioned DMD functional outcome exists.

Eligible task cells: none in the current release

Generalization Cube

2 executed · 1 registered · 5 locked.

This is a sparse set of registered task cells across six novelty axes, not a Cartesian leaderboard or a single generalization number.

Open task registry →
G0EXECUTED LIMITED

Same-context target holdout

perturbation novelty · SAME CONTEXT HELD OUT TARGET

Current result
Small average-error gain; strict gate 0/16 and direction unreliable
Decision blocker
No external cell-level or disease-context truth
Open task contract →
G1EXECUTED PARTIAL UNSUPPORTED

External unseen perturbation

dataset or laboratory · EXTERNAL UNSEEN PERTURBATION

Current result
No directional-transfer support
Decision blocker
No harmonized cell-population outcome or matched feature space
Open task contract →
G2REGISTERED NOT EXECUTED

Unseen gene family or pathway

perturbation novelty · UNSEEN GENE FAMILY OR PATHWAY

Current result
No result
Decision blocker
Current candidate set is too small for a powered family holdout
Open task contract →
G3LOCKED NO TRUTH

Unseen donor

donor or line · UNSEEN DONOR

Current result
No result
Decision blocker
No qualifying donor-resolved DMD perturbation matrix
Open task contract →
G4LOCKED NO TRUTH

Unseen cell state

biological context · UNSEEN CELL STATE

Current result
No result
Decision blocker
Current myogenic trajectory is real but unperturbed
Open task contract →
G5LOCKED NO TRUTH

Future batch or external laboratory

dataset or laboratory · FUTURE BATCH OR EXTERNAL LAB

Current result
No result
Decision blocker
No registered prediction or future outcome
Open task contract →
G6LOCKED NO TRUTH

Independent DMD disease line

disease background · INDEPENDENT DMD DISEASE LINE

Current result
No result
Decision blocker
0 independent DMD candidate-perturbation outcomes
Open task contract →
G7LOCKED NO TRUTH

New modality and combination

modality or combination · NEW MODALITY OR COMBINATION

Current result
Only an additive no-interaction preview is available
Decision blocker
No observed candidate doubles or matched DMD drug screens
Open task contract →

Claim boundary

Zero new runs · zero new scientific claims.

The Model Evaluation Gateway reports admission, metric and truth availability from released objects. It creates no inference endpoint, new benchmark result, DMD prediction, experiment authorization or biological claim.

Does not imply: a runnable hosted model · a newly eligible adapter · a global model score · DMD generalization · clinical or therapeutic validity.