Canonical machine-readable artifact record
Exact file identity, standard model I/O, scoped findings, and hash-bound companion evidence.
ON-DEVICE Pipeline
Account
Analysis is open. Report exports require an account.
Runtime Evidence
Member Workspace
Submit access requests, feedback, and research notes. Model files and reports are not uploaded.
Privacy Agreement
This workbench runs locally in your browser. Model weights, selected images, generated reports, tensor values, and outputs are not uploaded by this app.
Select or replace the local artifact, choose the target profile, then run or re-run the static audit.
Choose a local .tflite or .onnx artifact to inspect file metadata.
A light local read starts only after a model is selected.
Model bytes are read fully only after Run Static Audit is clicked.
Switch between the main summary, delegate coverage, quantization contract, roofline pressure, and stage structure.
Review target-independent artifact identity first, then the separately labeled selected-target engineering dashboard.
Select a completed module result, open the Engineering Report, or export reusable engineering evidence.
Choose a local artifact and target device.
Pick model and target
Review graph and risks
Download evidence or request access
Run controls live here. Results and downloads appear in the report pane below after execution.
The entitlement-gated WASM module is loaded only when this account can run Advanced analysis. Browser delivery controls access, not cryptographic source secrecy.
Runs input/output drift probes and backend availability checks locally in the browser.
Run Static Audit first, then select target devices and submit.
Runs locally against the deployed TFLite runtime path and reports deploy-domain stability evidence.
Includes model structure, quantization scope, XNNPACK chains, roofline signals, benchmark evidence, and engineering action items.
Every document binds the analyzed artifact SHA-256; target-dependent requirements bind the active analyzed target.
Local deploy-artifact proxy evidence. It prioritizes follow-up validation; it is not a training-time curvature certificate.
Local input perturbation sensitivity. Model bytes and generated tensors stay in browser memory.
Local backend availability and output-drift checks.
agent.py on a real device to register it.
Runs local finite-difference probes on the deployed model function and reports research-stage stability proxies.
Run a static audit to generate the report preview.
Member accounts can download the Engineering Report. Advanced access unlocks Download Raw Data, CSV, graph, visual, ML-BOM, and bundle artifacts.
Run a static audit and sign in to download the engineering bundle.
Every completed audit is stored locally (identity, counts, contracts, quantization, delegation, hotspots — never model bytes). Re-open, re-run, or compare two versions of a model by SHA-256.
Artifact observations remain distinct from declarations, inferred lineage, conditional runtime consequences, and unassessed fields.
Exact file identity, standard model I/O, scoped findings, and hash-bound companion evidence.
CycloneDX record, interface ledger, formulation, runtime, provenance, and member-digest manifest.
Selected-target cache, ISA, roofline, delegation, and modeled engineering signals. Values in this section can change when the target profile changes.
Load a .tflite or .onnx file to generate an optimization profile.
Observed or derived conditions that deserve attention; no remediation is implied.
Recommended changes with modeled upper-bound impact and expected engineering scope where available.
Flame-style summaries for compute concentration, static bottlenecks, delegate chains, quantization risk, and target fit.
Measures local browser runtime latency with warmup separated from steady-state runs.
| Backend | Compile | First (cold) | p50 | p90 | p95 | p99 | Mean | Trimmed p50 | Jitter | Trend | Status |
|---|
Artifact findings ranked by priority; each card separates evidence, possible effects, interpretation limits, and recommended action.
Run the static audit to generate findings.
Search ops, inspect tensor metadata, and trace local producer/consumer links.
| # | Op | Out | Kernel | MACs | Est. steady µs | Bound | XNNPACK | Quant | Align | Pack | Fusion | Low-norm filters |
|---|
| T# | Name | Shape | DType | Role | Prod | Fan | Consumers | Quant |
|---|
| # | Op / path | Kernel / provider evidence | Tile / occupancy | Static assignment | Interface payload | Preparation / packing | Observed runtime | Evidence |
|---|
| Op | Count |
|---|
| # | Op | Output | MACs | Bytes |
|---|
| # | Op | Intensity | Logical traffic | Row working set | Guess | Action |
|---|