Everything in one product
Each capability maps directly to a paper — no marketing claim on this page is unsupported by the underlying research (papers under peer review; preprints where noted).
Defect-Driven Test Prioritisation
Run the production model gb-paper1-v4-fixed_age against any pull request. Top 10% of files capture 43.82% of defects.
LLM-Powered Test Generation
Requirements in, executable tests out. Deterministic rule verifier in the loop; 96.3% first-pass verification.
Self-Healing Automation
Tree-ensemble locator ranking recovers broken selectors at runtime with confidence-gated heal-vs-flag decisions.
Cross-Repository Risk Profiles
Leave-one-repository-out validated transfer for teams whose codebases don't look like the benchmark set.
CI/CD & Webhook Integration
GitHub Actions, advisory and gating rollouts, pre-merge risk advice, and a stable HTTP contract for IDE plugins and CLI tools.
AI Governance Layer
Payload controls, content redaction, author privacy hashing, prompt versioning, and verifiable audit trails on every call.
Trust, Augment, or Replace?
What a cross-project defect predictor is worth on 1,323 real defects from Defects4J and BugsInPy — calibration and cost-aware triage advice. Under review at PeerJ Computer Science.
Quality Analytics Dashboard
Real-time view of risk distribution, top-k coverage, and self-heal vs. flag rates across services.
How TestForge AI works
Four steps from sign-up to a pre-merge gate that catches defects before they ship.
-
01Connect your repository▶
OAuth into GitHub; TestForge AI mirrors the commit history through the Repository Analytics Engine and produces a process-metric snapshot per file.
-
02Calibrate the risk model▶
The cross-repository transfer pipeline from Paper B picks the best starting point; an in-team calibration pass uses Platt scaling to tune for your defect rate.
-
03Wire into CI/CD▶
Install the GitHub Actions workflow. Start in advisory mode (PR comments only). Promote to gating mode when the team is comfortable.
-
04Generate & verify tests▶
Drop a requirement into the RAITG service. Receive Playwright + BDD specs, the rule-verifier report, and a symbolic-mutation-indicator adequacy score.
What sets TestForge AI apart
Research-backed, not vibe-coded
Every claim on the product page maps to a paper under peer review (or a clearly-marked preprint) with reproducible artefacts.
Calibration before accuracy
Production selection criteria privilege calibration, top-k stability, and threshold sensitivity over raw AUC.
Multi-language, multi-framework
Java, JavaScript, Python, and Go on the roadmap. Frameworks: Selenium, Playwright, Cypress, PyTest, JUnit.
Verified LLM generation
Deterministic rule verifier in every loop. The LLM never gets the last word.
Governance is architecture
Payload controls and prompt receipts are not opt-in; they're how the platform talks to models.
From research to production
The exact code that produced the paper's numbers is the code running in production.