FirstMarker Phase 1 Results

Real-World Classroom Trial – Barbados Primary School

Over the past week, I conducted Phase 1 testing of FirstMarker at a primary school in Barbados.

The objective was simple:

Can AI meaningfully reduce grading time while maintaining acceptable accuracy in a real classroom setting?

This was not a lab simulation.
This was a live test using actual student scripts.

📍 Context

• Location: Primary school in Barbados
• Subject: Language Arts
• Script length: 21 pages
• Format: Booklet-style test paper
• Validation method: Multiple AI grading passes for consistency

This represents the first structured validation phase of FirstMarker in a school environment.

📱 Scanning Workflow: 21 Pages in Under 2 Minutes

Before grading begins, scripts must be digitized.

For this trial, scripts were scanned using vFlat, a mobile document scanning app optimized for books and bound materials.

Using vFlat:

• A 21-page test booklet was scanned in under 2 minutes
• Pages were automatically flattened and cropped
• Bound script curvature was corrected
• Image clarity was sufficient for AI analysis

This scanning speed makes high-volume grading feasible.

⏱ Grading Speed: Under 2 Minutes

Once scanned, FirstMarker processed the full 21-page script in under 2 minutes.

For comparison:

• Manual grading typically takes 15–25 minutes per script
• Entering marks into spreadsheets adds additional time
• Fatigue can reduce grading consistency over large batches

FirstMarker completed:

• Text extraction
• Rubric comparison
• Score calculation
• Structured feedback generation

All in under 120 seconds.

🧠 Detailed Feedback for Every Script

Each script did not simply receive a score.

FirstMarker generated:

• Structured strengths
• Areas for improvement
• Question-level feedback
• Clear breakdown aligned to the rubric

This is important.

The system does not only calculate marks — it produces actionable feedback for students.

That significantly increases its value during revision periods.

📊 Consistency Results (Language Arts – No Diagrams)

The same scripts were graded multiple times to test reliability.

Findings:

• Most results were identical across runs
• Some varied by ±1–2 marks
• Larger variations were rare

This suggests:

• Stable rubric interpretation
• Consistent scoring logic
• Minor variation primarily linked to handwriting recognition

For text-heavy Language Arts papers without diagrams, internal consistency was strong.

✍️ Limitation Identified: Poor Handwriting

The primary weakness observed was poor handwriting.

When handwriting was:

• Clear → scores were highly consistent
• Moderately messy → minor variation
• Very difficult to read → occasional misinterpretation

This is not a grading logic issue.

It is a vision/OCR limitation.

Language Arts responses are sensitive to word-level interpretation.
A single misread word can affect scoring.

This limitation must be transparently acknowledged.

⚖️ Important Positioning

FirstMarker should not be used as the final authority for official examination marks.

Instead, it is best positioned as:

• A first-pass grading assistant
• A consistency checker
• A moderation support tool
• A formative assessment accelerator

The teacher remains the final decision-maker.

Professional judgement is preserved.

🎯 Best Use Case: Exam Preparation (e.g., BSSSE)

FirstMarker is particularly well suited for:

• Exam preparation practice (e.g., BSSSE preparation)
• Mock examinations
• Large revision batches
• Rapid feedback cycles

During exam season, teachers may correct hundreds of pages in a short timeframe.
Reducing grading time from 20 minutes to 2 minutes per script is transformational.

📌 Phase 1 Summary

• ✅ Conducted at a primary school in Barbados
• ✅ 21 pages scanned in under 2 minutes
• ✅ 21 pages graded in under 2 minutes
• ✅ Detailed feedback returned for every script
• ✅ Strong internal consistency for text-based Language Arts
• ⚠️ Performance impacted by poor handwriting
• ❌ Not intended to replace final teacher judgement

What Comes Next

Phase 1 focused on speed and internal consistency.

Next phases will include:

• Direct comparison against teacher-assigned marks
• Accuracy measurement (mean absolute error, ±1 tolerance rates)
• Expanded subject testing
• Handwriting robustness improvements

Final Reflection

FirstMarker is not about replacing teachers.

It is about reducing administrative burden so teachers can focus on instruction, intervention, and student growth.

Phase 1 demonstrates that AI-assisted grading can be:

• Fast
• Structured
• Consistent
• Practically deployable

Further validation is ongoing.

More results soon.