How Matflow compares to adjacent platforms
This is an evaluation guide, not a vendor scorecard. We publish what Matflow verifiably does — with the file that backs each claim — and the questions to ask every platform, including ours. Vendor capabilities change; we do not assert them. As of 2026-10-05.
Why there is no competitor feature matrix here
Two reasons, both about trust.
Six dimensions we can point to in the code
These are the axes we would use to compare any research platform — and the ones we hold ourselves to.
A free tier and published prices that anyone can reach without a sales call. The pricing page renders names, prices and limits live from the same billing catalog the server enforces (client/src/pages/StudioPricing.jsx → GET /api/billing/plans, backend/core/pricing.py).
Every scientific response carries one of six evidence classes — MEASURED, COMPUTED, PREDICTED, EXTRACTED, HYPOTHESIS, DEMO — with a level and an engine card naming what produced it. Surrogates are relabelled DEMO rather than printed as facts (backend/core/engines.py:98-116, 303-330).
Benchmark reports ship downloadable reproducibility bundles, and saved datasets export as RO-Crate v1.1 ZIPs with provenance and engine cards. Training files and dataset versions are SHA-256 hashed (client/src/pages/StudioBenchmarks.jsx:565, backend/routes/rocrate_routes.py:3-10, backend/core/provenance.py).
Files live under a per-account storage directory, and every served path is validated against that directory — another account cannot reach your files (backend/core/paths.py:71-105). Models are per-owner rows in the model registry, and the RO-Crate export gives you a portable copy of your data.
A REST API with an OpenAPI 3.0 spec and interactive reference, a Python SDK under tools/sdk, per-user API keys stored as SHA-256 hashes, and HMAC-signed webhooks with retries for run completion and failure (backend/core/api_auth.py:24, backend/core/webhooks.py:1-14).
The product is a web application with no desktop install. The repo ships a single-tenant Docker Compose overlay for dedicated deployments (deploy/production/docker-compose.single-tenant.yml), and the deployment-readiness endpoint reports tenancy, region gating and BYOK mounting flags without exposing secrets (backend/core/soc2.py:367-411).
Nine questions to ask every platform, including Matflow
Take these to any evaluation. None of them requires believing a comparison page — ours included.
- Can I try it before talking to sales? — Look for a self-serve signup, a free or trial tier, and published prices — not a "request a demo" wall.
- Is pricing public and itemized? — Ask what is metered, what a unit means, what resets when, and what happens at the cap.
- Are results labelled by how they were produced? — Measured, computed, predicted, extracted and heuristic outputs should be distinguishable at the result, not in a footnote.
- Can I export my data and my model? — Ask for a portable, documented export format and deletion path — not just a CSV download button.
- Can a reviewer reproduce a result? — Ask whether the exact dataset version, configuration, seed and metrics travel with the result.
- How is my data isolated and used? — Ask where files live, who can access them, and whether your data trains anything shared.
- What runs when an engine is unavailable? — The honest answer is a labelled fallback or a refusal — never a fabricated number.
- What is the programmatic surface? — Ask for an API spec, SDKs, webhook events and how keys are stored and revoked.
- What compliance claims are certified vs. aspirational? — Ask for the certificate, its scope and its date before treating any compliance statement as real.
What Matflow does not claim
Honesty cuts both ways. These are the boundaries of what this platform is.
- We do not claim numerical superiority over any other platform. No head-to-head benchmark scores are published because we cannot independently run and verify another vendor's product from here.
- We do not physically run experiments. The platform is software: wet-lab execution, robotic synthesis and instrument operation stay with you or a hardware partner.
- We do not hold SOC 2 or ISO certifications. The repository contains SOC 2 evidence scaffolding (audit CSV export, optional access logs, header configuration, dependency scans) but no certification is claimed.
- Predicted, extracted and screening outputs require validation. Every result carries its evidence class so a hypothesis is never mistaken for a measurement.
Where each claim lives in the repository
For technical evaluators: the claim, and the module that makes it true.
| Claim | Enforced / recorded by |
|---|---|
| Prices, limits, seats and credit packs | backend/core/pricing.py + GET /api/billing/plans |
| Tier quotas enforced per request | backend/core/constants.py (USER_TIERS) + core/runtime.py |
| Compute metering and quota windows | backend/core/credits.py + core/compute_quota.py |
| Evidence classes, levels and engine cards | backend/core/engines.py + core/evidence_middleware.py |
| Dataset versions and file hashes | backend/core/provenance.py |
| Per-account storage path validation | backend/core/paths.py |
| API key storage and session checks | backend/core/api_auth.py + utils/auth_utils.py |
| Webhook signing and retries | backend/core/webhooks.py |
Experience the self-serve difference
No waiting for demo approvals. Sign up, load a bundled dataset or upload your own, and inspect every result's evidence class before you decide.