Claude-Assisted Evaluation Analysis
Move from evidence to analysis, triangulation, findings, verification and improvement.
Build a Claude-Assisted Evaluation Analysis
Practice Case: Youth Employment Pathways Programme
The programme supports young people aged 18–29 through:
- Employment-readiness training
- Technical skills development
- Career counselling
- Employer connections
- Employment placement support
Illustrative monitoring information indicates that employment among participants increased from 42% at baseline to 57% six months after programme completion.
This change should not automatically be interpreted as evidence that the programme caused the improvement.
Evaluation Questions
EQ1: What changes in employment outcomes are observed among programme participants?
EQ2: What factors appear to support or constrain participants' progress?
EQ3: What changes in employment readiness and opportunities are reported?
EQ4: What alternative explanations or differences in participant experience should be considered?
What this interactive lab does
This simulation allows you to practise the reasoning process used in a Claude-assisted evaluation analysis. It does not connect to Claude or another external AI service.
You will demonstrate what you would ask Claude to do, what you would examine, and how you would verify the resulting analysis.
Build the Evidence Matrix
Your first task is to establish what evidence exists before asking Claude to synthesize findings.
Your evidence package may include:
- Terms of Reference or evaluation assignment
- Programme description
- Theory of Change
- Results framework or logframe
- Monitoring report
- Participant dataset
- Anonymized interview transcripts
- Focus-group evidence
- Open-ended survey responses
- Employer or stakeholder evidence
- Previous evaluation or programme review
Ask Claude to create an evidence inventory showing source identifier, source type, date, population or perspective, evaluation question addressed, key evidence, limitations and whether the source is primary or secondary.
Then ask Claude to create an evidence matrix separating evidence claims, supporting evidence, contradictory or qualifying evidence, limitations and verification status.
Analyze Qualitative Evidence
Use interview transcripts, focus groups and open-ended responses to explore participant experiences while preserving differences, contradictions and negative cases.
Ask Claude to identify potential codes and themes, create a draft codebook, distinguish expected from emerging codes, preserve participant variation, identify unusual or negative cases and challenge possible overgeneralization.
Qualitative analysis should examine
- Recurring experiences
- Participant variation
- Supporting evidence
- Contradictory evidence
- Negative cases
- Alternative explanations
- Context
- Evidence limitations
Analyze the Quantitative Evidence
Before interpreting the participant dataset, establish whether the data are suitable for the planned analysis.
Ask Claude to check missing values, duplicates, inconsistent categories, impossible or unusual values, variable types, coding issues and denominator problems.
Then request descriptive statistics, baseline and follow-up employment analysis and only those subgroup comparisons supported by the dataset.
The 42% → 57% change
The observed change is 15 percentage points.
The analytical task is not simply to report that the number increased. Examine denominators, missing data, measurement, subgroup patterns and what the evaluation design permits you to conclude about programme contribution or causality.
Compare and Triangulate Evidence
Now compare quantitative, qualitative and documentary evidence. Do not automatically reconcile contradictions.
Ask Claude to compare evidence for each evaluation question and identify:
- Convergence
- Divergence
- Where one source qualifies another
- Missing evidence
- Whether sources are genuinely comparable
- Methodological or contextual explanations for differences
Develop Evidence-Supported Findings
Move from verified evidence toward candidate findings without skipping the reasoning process.
Do not skip the intermediate reasoning.
Ask Claude to draft candidate findings for each evaluation question.
For every finding, require:
- Finding statement
- Supporting evidence
- Contradictory or qualifying evidence
- Sources
- Interpretation
- Confidence supported by the evidence
- Important limitations
- Alternative explanations
Do not ask Claude to make recommendations yet.
Challenge and Verify the Analysis
Human verification is a required component of the project.
Important outputs must be checked independently against the original evidence before they are treated as evaluation evidence or findings.
Evidence claims: Does the original source support the claim?
Quantitative results: Are calculations, denominators and transformations correct?
Qualitative coding: Does the coding preserve meaning and context?
Themes: Are themes supported and challenged by negative cases?
Triangulation: Are sources genuinely comparable?
Findings: Do findings remain within what the evaluation design can support?
Improve the Evaluation Analysis System
The project should not end when the analysis is complete.
Use what you learned from verification to improve the workflow before reusing it in another evaluation.
Review where Claude required correction, where evidence was unclear, where prompts were insufficient and where verification revealed risks.
Convert those lessons into improvements to your prompts, evidence structure, verification process or project instructions.
Claude-Assisted Evaluation Analysis — Results
Human–AI Verification Record
For the real project submission, document:
- What Claude did
- What Claude got wrong
- What evidence you verified
- What you changed
- What required human judgment
- What risks you identified
- What safeguards you applied
Final Quality Standard
Your analysis should make evidence traceable, distinguish evidence from interpretation, examine contradictory evidence, consider alternative explanations, respect methodological limitations and document human verification.