Guided Scholar · AI Transparency

How AI works inside Guided Scholar.

What the AI does at each step, what it cannot do, and how the scoring model was validated — explained plainly for educators, administrators, and district technology teams.

The Workflow

What happens when a student submits.

Every step of Guided Scholar's feedback process is deliberate and traceable. Here is what the AI does — and what it does not do — at each stage of a student submission.

01
Student submits their work.

The student enters their own content — an introduction, a body paragraph, a complete essay, or a full ACT response. Guided Scholar also supports spreadsheet and presentation assignments. For those formats, students receive step-by-step construction guidance as they build their document in Sheets or Excel, or their slides in PowerPoint or Google Slides. In all cases, the AI receives exactly what the student produced, nothing more.

02
The AI evaluates against the rubric.

The AI scores the submission across rubric categories — Organization, Clarity of Ideas, Supporting Details and Evidence, Grammar and Mechanics, and Overall Effectiveness. This rubric applies to written assignments, spreadsheets, and presentations alike, with evaluation adapted to what each format requires. ACT mode uses the four official ACT domains instead. Each score is accompanied by a written explanation tied to what the student actually produced.

03
Feedback is structured and specific.

The system identifies specific strengths in the submission, specific areas for improvement, and generates an example revision suggestion — showing one of the student's actual sentences alongside a stronger version. This is not generic advice; it is tied to the student's specific draft.

04
Next steps are prioritized.

The feedback closes with a prioritized next step — the single most impactful revision the student can make — so they are not overwhelmed by a list of everything that could be better.

05
The student revises and resubmits.

The student reads the feedback, decides how to revise, writes the revision themselves, and resubmits. The AI evaluates the new draft. This loop repeats as many times as the student needs. The teacher dashboard records the progression across every draft.

Capabilities & Limits

What the AI does and does not do.

The AI does
  • Evaluate student-submitted writing against defined rubric criteria
  • Provide rubric scores with written explanations
  • Identify specific strengths in the submitted text
  • Identify specific areas for improvement
  • Generate a revision example using one of the student's actual sentences
  • Provide prioritized next steps for revision
  • Score ACT responses against the four official ACT domains
  • Run an ACT structure check identifying what elements are present, partial, or missing
  • Classify revision depth between draft submissions
  • Flag students who have not revised after receiving feedback
The AI does not
  • Generate text on behalf of the student
  • Write or complete any part of a student assignment
  • Provide grades that replace teacher assessment
  • Detect or identify AI-generated content
  • Access or transmit data to third-party AI providers
  • Use student submissions to train or retrain any model
  • Make instructional decisions — those remain with the teacher
  • Override teacher judgment about student performance
ACT Scoring Validation

How ACT scoring accuracy was validated.

Educators and administrators evaluating Guided Scholar for ACT preparation have a reasonable question: how closely does the system's scoring align with actual ACT scoring? This is a meaningful question because a tool that scores inconsistently with the real test creates false confidence rather than genuine preparation.

66 / 66
Essays scored within one point of human-graded score

The ACT scoring model was validated against six sets of essays graded by humans using official ACT rubric criteria. The model's scores were compared against the expected human scores across all 66 essays. Every essay landed within one point of the human-graded score.

What this validation means — and what it doesn't.

This validation confirms that Guided Scholar's ACT scoring model produces results meaningfully aligned with human scoring on the official ACT rubric. Students who use Guided Scholar to practice ACT writing are receiving feedback calibrated to the same standards the actual test applies.

What this validation does not mean: Guided Scholar's scores will match the actual ACT score a student receives on test day. The actual ACT is scored by trained human scorers under specific conditions. Guided Scholar's scores are a practice-context approximation — useful for identifying strengths and weaknesses and improving performance, but not a prediction of official test results.

Ongoing calibration: The ACT scoring model continues to be calibrated as additional human-graded essays become available. Validation results will be updated as the dataset grows. Schools or districts that have access to official ACT writing samples and are willing to support calibration efforts are encouraged to contact us.
For District Technology Teams

Technical questions from administrators.

Guided Scholar is currently a Streamlit-based application in a controlled beta phase, transitioning to a cloud platform. The system processes student submissions through an AI model that is hosted and operated by Brau Consulting LLC. No student data is transmitted to external AI providers.

The teacher dashboard is built on the same platform and provides real-time access to student draft histories, revision depth classifications, and flagged follow-up items. All data visible in the teacher dashboard reflects actual student submission records — not estimates or inferences.

District technology teams with specific questions about infrastructure, data handling, security architecture, or integration requirements are encouraged to contact us directly. We will provide whatever documentation is available and be transparent about what is still in development.

Technical questions about how Guided Scholar's AI works?

peter@guidedscholar.ai