Test what an AI English speaking tutor hears, teaches and does when a conversation becomes difficult. Abdelaty Tutor Assurance Studio v1.0 gives evaluators 100 structured scenarios, a formula-driven scoring workbook and a practical way to document findings across releases.

Use it to investigate problems such as interrupted planning pauses, inaccurate corrective feedback, unsuitable level adaptation and unsupported claims about pronunciation or exam scores. The evaluation is human-run: you operate the tutor, capture evidence and make the judgements.

Try the free 10-scenario sample before choosing an edition.

Designed for independent AI consultants, educational assessment professionals, AI tutoring developers, language-learning technology teams and organisations reviewing English speaking tutors. It supports internal educational quality assurance and conversational AI evaluation.

100 scenario cards across 10 modules, with learner context, stimuli, execution steps, expected behaviour, scoring criteria, critical-failure conditions and reference answers. PDF library plus CSV and JSON exports.

A 15-sheet Evaluation Workbook, supplied blank and with simulated worked-example data. It calculates coverage, module scores, repeated-run reliability, baseline-versus-candidate regression, speech-accuracy analysis and a critical-defect release gate.

55 WAV files, including two calibration signals, with reference transcripts, checksums and an audio asset register.

An Audio Implementation Protocol and editable consent template for collecting authentic second-language recordings.

A 20-section Evaluation Manual, formula test report, editable reporting templates and a complete worked example evaluation report clearly labelled as simulated.

An optional Python 3 word-error-rate helper using the standard library.

The ten modules cover speech recognition, second-language speech comprehension, environmental robustness, turn-taking, intent repair, CEFR-informed adaptation, corrective feedback, pedagogical task completion, safety/privacy/fairness, and reliability/recovery.

Professional — US$149. One authorised evaluator and one evaluation project. Includes the full core system and six core report templates. Suitable for a solo evaluator reviewing one tutor product, including its successive versions.

Team — US$449. Up to five named users and three evaluation projects within one legal entity. Adds all ten report templates, an Assessor Calibration Workbook, a Release Trend Tracker, the team workflow guide, and three defect/triage/calibration templates. Choose this when several evaluators need a shared review process.

Organisation — US$899. Employees and contractors working for one licensed legal entity across unlimited internal evaluation projects. Includes the Team materials plus a Portfolio Roll-up Workbook, onboarding and deployment pack, eight editable SOPs, evaluation plan, programme charter/RACI and portfolio report templates. Parent companies, subsidiaries and other affiliates require their own licence.

1. Open 00_START_HERE.pdf and explore the worked-example workbook.

2. Read the manual, define your tutor versions and test conditions, and check the audio route with the calibration files.

3. Run the 20-scenario core regression set, then extend coverage to the scenarios relevant to your tutor.

4. Record observations and evidence, review scores and critical defects, and use the templates to document fixes and retests.

Missing tests are not passes. N/A decisions are explicit. A critical failure can block the release gate even when an overall score looks favourable. These tools help organise decisions; they do not replace evaluator judgement.

Designed for Microsoft Excel 2010+ and LibreOffice Calc 7+. Documents are PDF and DOCX; audio is 24 kHz, 16-bit mono WAV. No workbook macros, add-ins or API keys are required. Word-compatible software is needed to edit DOCX templates. Voice tests need a compatible tutor and an audio replay/injection route; the protocol explains virtual-audio and loudspeaker methods. Python 3 is needed only for the optional WER helper.

Check the worked example in your chosen spreadsheet application before relying on results. The toolkit has no hosted evaluation service. Your tutor, spreadsheet service and audio setup determine where test inputs and results are processed.

This is a human-run methodology, not automated SaaS. Results depend on test conditions, coverage and evaluator judgements. The method is not independently certified or psychometrically validated; weights, thresholds and calibration targets are provisional. It does not certify tutors or validate learner CEFR levels.

The included synthetic audio does not establish genuine L2 accent performance. Advanced scenarios L2C-09 and L2C-10 require properly sourced customer recordings; follow the protocol and obtain the necessary permissions. The toolkit does not guarantee AI performance improvements, defect detection or commercial outcomes. It provides no legal or regulatory advice.

Each option is a one-time purchase of a perpetual licence for the version supplied, within its stated user and project limits. You may adapt the toolkit internally and share evaluation findings that do not reproduce substantial toolkit content. Redistribution, resale, sublicensing, white-label products and public training datasets are not permitted. Consultants may deliver findings to clients; clients need their own licence to use the toolkit.

Select the appropriate edition before checkout. Your purchase delivers only that edition's ZIP and licence through Contra. Download, unzip and open 00_START_HERE.pdf. Error reports and clarification of documented features are supported through Contra messages, with an aim to acknowledge within five UK working days. Consulting, training and bespoke implementation are separate.

Author and licensor: Suliman Abdelaty, FHEA, FRSA.