TRL for software: what each technology readiness level means for an app, platform or AI product
TRL for software explained: what TRL 1 to 9 mean for an app, SaaS or AI product, relevant vs operational environments, and the evidence each level needs.
Part of the guide TRL levels explained: what TRL 1 to 9 really mean, and how to place your technology
For software, TRL 4 means key components work together and are validated in a development setting, TRL 6 means a prototype is demonstrated on full-scale realistic problems in a relevant end-to-end environment, and TRL 9 means the system is operating in its real environment with ongoing support. India's draft NTRAF has a separate software questionnaire that ties the higher levels to alpha and beta testing, third-party validation, security audits and a market launch.
Grant forms and investor templates in India now ask for a technology readiness level. For a hardware team, the scale reads naturally: a lab bench, a prototype, a field trial. For a software or AI team it feels awkward. Is a working demo TRL 6? Is a beta with ten users TRL 7? Does "operational environment" mean production servers?
The confusion is real and well documented. The US Department of Defense's Technology Readiness Assessment Deskbook says plainly that the original set of definitions suited hardware technologies but was inadequate for software. Software-specific definitions were written to fix that, and India's draft framework now has its own software questionnaire. This guide puts them together into plain language.
The short answer
For software, readiness is about how complete and integrated the code is, and how realistic the data, users and infrastructure it has been tested with are. Lower levels are about components and synthetic data. Middle levels are about an end-to-end system tested on realistic problems. Higher levels are about integration with real systems, independent testing, security and actual use in the market.
Where the software definitions come from
Three sources are worth knowing:
- NASA's TRL scale has a hardware column and a separate software column for each level. Its software wording is the most widely quoted.
- The US DoD Technology Readiness Assessment Deskbook (2009) has a software TRL table and a useful definition of relevant and operational environments.
- India's draft National Technology Readiness Assessment Framework (NTRAF), from the Office of the Principal Scientific Adviser, developed with CII. It was unveiled for public consultation at the end of December 2025 and is still described as a draft. Its Annexure B is a software questionnaire with checkpoints for each level.
If your proposal goes to BIRAC, check its sector document too. BIRAC publishes sector-specific TRL definitions, including one for artificial intelligence, big data analysis, IoT, software development and bioinformatics, and says it uses TRLs for selection and monitoring. Use the wording of the scheme you are applying to.
TRL 1 to 9 for software, in plain words
- TRL 1: The scientific basis is known: the algorithm, the mathematical formulation or the architecture principle.
- TRL 2: The application is still speculative. Basic principles are coded and tried on synthetic data.
- TRL 3: Limited functionality is built to test the critical properties, using components that are not yet integrated.
- TRL 4: The key, functionally critical components are integrated and validated in a development setting. NTRAF's TRL 4 checkpoints refer to software components tested with partially representative data. The relevant environment is defined, and performance in it is predicted.
- TRL 5: An end-to-end system is tested in a relevant environment and meets predicted performance. NTRAF asks for conformity with the target environment and interfaces, realistic problems and rigorous alpha testing.
- TRL 6: A prototype is demonstrated on full-scale realistic problems and partially integrated with existing hardware and software systems. NTRAF adds beta testing, scalability in a real-world context and a plan for user feedback.
- TRL 7: Prototype software has all key functionality and is well integrated with operational systems. NTRAF expects third-party testing, independent verification and validation, and third-party security assessments and audits. The DoD Deskbook notes that for software, TRL 7 means all source code has been written and tested.
- TRL 8: The software is thoroughly debugged and fully integrated with operational systems, documentation is complete and verification and validation are done. NTRAF adds quality to ISO/IEC 9126 or an equivalent standard, data privacy compliance, and launch in the intended market or user environment.
- TRL 9: The system has been operated successfully in its operational environment, with sustaining engineering support in place. NTRAF looks for continuous improvement through new versions based on user demand.
Software TRL, from algorithm to product in use
Relevant environment versus operational environment
These two phrases decide most arguments about software TRL, so be precise.
NTRAF defines a relevant environment as one that simulates the key aspects of the operational environment. The DoD Deskbook goes further: it should simulate both the most important and the most stressing aspects, and it groups environments into physical, logical, data, security and user categories. For software, that means realistic data volumes and quality, realistic users, the interfaces you will actually connect to, and the security conditions you will face.
An operational environment, in NTRAF's words, addresses all the operational requirements and specifications of the final system. For software, that is the real deployment: real infrastructure, real integrations, real users and real data.
Relevant or operational environment for software?
A polished demo on clean sample data is not a relevant environment.
AI and machine learning products need extra care
A model that works on a benchmark dataset may still be far from ready. ML systems learn from data, and that brings problems hardware scales never had: data drift, bias, and performance that changes when real data arrives. A 2022 paper in Nature Communications by Lavin and colleagues argued that NASA's spacecraft process is overkill for ML and proposed a separate Machine Learning Technology Readiness Levels framework.
You do not need a new scale for an Indian grant. But in an AI proposal, show:
- what data the model was trained and tested on, and how close it is to real data,
- performance on data the model has not seen, from the target setting,
- how you monitor and retrain once it is deployed,
- privacy and security checks, especially with personal or health data.
The evidence each band needs
- TRL 3 to 4: architecture document, component test results, code repository history, synthetic or sample data results.
- TRL 5 to 6: end-to-end test reports on realistic data, alpha and beta test logs, performance and scalability results, user feedback records.
- TRL 7 to 8: third-party test or audit reports, security assessment, integration records with real systems, user and maintenance documentation, compliance evidence.
- TRL 9: usage data from real deployment, support records and release history.
Our guide on how to prove your TRL explains how to build an evidence pack reviewers trust, and TRL levels explained covers the general scale.
Common mistakes
- Calling a demo TRL 6. A demo on clean data with no real users is usually TRL 4 or 5.
- Confusing "deployed on the cloud" with operational. Hosting is not the same as real users and real integrations.
- Skipping security and privacy until TRL 8. NTRAF asks for third-party security assessment at TRL 7.
- Using hardware wording in a software proposal. Use the software definitions and say which source you used.
- Overclaiming for an AI model based on benchmark results alone.
Getting a structured assessment
If you want a second view before a grant or investor meeting, TRL Pro walks you through a structured readiness assessment for your technology and gives you a TRL from 1 to 9 that you can check your evidence against. It uses Zernyx's own copyrighted framework (ZTRAF), covering more than 60 technology domains and subdomains. You can run it at TRL Pro. For the next steps up the scale, read moving from TRL 4 to TRL 6.
FAQ
What TRL is a software MVP?
It depends on what the MVP has been tested with. An MVP used by real early customers on real data, integrated with their systems, can be TRL 7 or above. An MVP shown only as a demo is usually TRL 4 or 5.
Is beta testing TRL 6 or TRL 7?
NTRAF's software questionnaire places beta testing at TRL 6. TRL 7 adds third-party testing, independent validation and security audits, with integration into operational systems.
Does India have its own TRL framework for software?
India's draft NTRAF includes a software annexure with checkpoints for each level. It is still a draft, so check the scheme you are applying to for the definitions it uses.
Can an AI model be TRL 9?
Yes, when the system is operating successfully in its real environment, with support, monitoring and updates in place. A model that only performs well on a benchmark is not.