❶ Data Collection Strategy
This study employed a web-based conversational survey (chatbot) as its primary data collection method. The survey was designed as a structured conversational experience delivered through Nesolagus LLC’s proprietary platform, Warren, and tailored for a youth population (high school students, grades 9–12). The instrument collected primarily qualitative data through guided open-ended prompts, supplemented by structured response formats including 3-point and 7-point scales, single-choice selections, multi-select options, yes/no gates, and a ranking exercise. The survey explored student experiences across eleven thematic areas: belonging and identity, safety and trust, engagement and agency, relationships and support, growth and expectations, voice and influence, learning content and curriculum, physical space and ancillary services, school culture, residential life, and policies and transparency.
❷ Who Sponsored the Research and Who Conducted It
Sponsor / Commissioners
This survey was commissioned by Derek Hall and Suri Lynne Seymour, independent consultants specializing in Diversity, Equity, Inclusion, and Belonging (DEIB) in K–12 education settings. The survey was conducted in preparation for a faculty and staff professional development workshop at Westover School scheduled for January 9, 2026.
Conducted By
Nesolagus, LLC (nesolagus.com). Aaron Lyles, Founder and CEO, served as project lead. Survey design, data collection, and analysis were conducted by Nesolagus. The Warren survey tool used for data collection in this study was developed in collaboration with a technical partner under a prior arrangement. Nesolagus has since independently rebuilt the platform. The research methodology, instrument design, analytical framework, and all client-facing deliverables were created solely by Nesolagus.
❸ Measurement Tools / Instruments
The survey instrument was a multi-block conversational script developed through a collaborative design process involving the commissioning consultants and informed by two established educational equity frameworks: the Culturally Responsive-Sustaining (CR-S) Education Framework and Dr. Gholdy Muhammad’s Five Pursuits of Equity (Identity, Skills, Intellect, Criticality, Joy). The instrument achieved a final internal QA score of 86/100 across six weighted dimensions: objective alignment (85), question neutrality (88), conversational flow (90), trust building (90), question type balance (78), and methodology compliance (88). The full conversational script is available upon request.
AI-Assisted Instrument Development
The survey instrument was initially drafted using Nesolagus’s AI-assisted design workflow, in which an AI language model (Anthropic Claude Sonnet 4) generated an initial conversational script from the approved project specification and methodology framework. The instrument was then substantially rewritten by the project lead and commissioning consultants (Derek Hall and Suri Lynne Seymour) before deployment. All AI involvement occurred during instrument design before any respondent data existed. No respondent data is transmitted to any AI API at any stage.
Opening Sequence
A multi-message introduction from the survey facilitators (Derek and Suri), establishing purpose, anonymity (“Everything you share is anonymous. No one will know which answers are yours.”), and a non-pressuring consent gate. Students who declined were shown a friendly exit message.
Question Types and Distribution
Open-ended text (16 questions, 36%), single-choice (7, 16%), multi-select (4, 9%), 3-point and 7-point scales (10, 23%), yes/no (3, 7%), and ranking (1, 2%). Demographics comprised approximately 11%. Conditional branching directed students through different paths based on input.
Bias Prevention
All questions were reviewed for six bias dimensions: leading language, presupposed experience, abstract labeling, compound constructions, deficit framing, and double-barreled phrasing. The instrument passed internal review on all six dimensions.
Demographic Collection
Demographics placed at end with explicit permission to skip. Race/ethnicity via open text self-identification. Gender identity and sexual orientation via multi-select with write-in options. Neurodivergence and disability questions included “prefer not to answer” with inclusive framing.
❹ Population Under Study
All currently enrolled students at Westover School, an independent all-girls boarding and day school in Middlebury, Connecticut, grades 9–12. Approximately 200 students. No eligibility criteria beyond enrollment.
❺ Method Used to Generate and Recruit the Sample
Non-probability sample. Census approach: all enrolled students (~200) invited via school communication channels. Survey link distributed one day before winter break. Voluntary participation. The commissioning consultants had met with some students in person prior to distribution, establishing familiarity. No incentives offered.
❻ Method(s) and Mode(s) of Data Collection
Self-administered web-based conversational survey via Warren platform. Chat-style interface, one question at a time. English only. Approximately 12-minute completion time (not displayed to participants in the final deployed version, noted as outstanding QA item).
❼ Dates of Data Collection Used to Generate and Recruit the Sample
Approximately December 18, 2025 through early January 2026 (approximately 2 weeks). Deployed one day before winter break.
❽ Sample Sizes and Discussion of the Precision of the Results
Population Invited | ~200
——————————————
Completed | 28
——————————————
Response Rate | ~14%
——————————————
Non-probability sample. Traditional margins of sampling error not applicable. Small sample size (n=28) limits precision. Results represent those who participated, not generalizable to the full student body. Primary value lies in qualitative narrative data used to inform faculty professional development programming.
❾ How the Data Were Weighted
No weighting applied. Qualitative-dominant inquiry. Weighting neither appropriate nor attempted given non-probability sampling, small sample size, and qualitative emphasis.
❿ How the Data Were Processed and Procedures to Ensure Data Quality
Data processing was handled through the Warren platform's configurable data cleaning pipeline, which operates at the point of data import and produces both cleaned data and a complete audit trail documenting all transformations. No 'save and resume' feature was available at deployment; each session was a unique attempt. Race/ethnicity normalization rules were developed specifically for this project's open-text responses. The pipeline consists of five stages:
Stage 1: Row-Level Validation
Empty rows with no substantive data were detected and removed. Duplicate responses were identified by matching session IDs and near-identical response content. Timestamps were validated for impossible dates and out-of-range durations.
Stage 2: Field-Level Cleaning
Whitespace was normalized (trimmed, multiple spaces collapsed). Case normalization was applied selectively: categorical response fields were normalized for consistency, while open-ended narrative text was preserved as entered. UTF-8 encoding issues, smart quote inconsistencies, and character encoding artifacts were repaired.
Stage 3: Spam and Low-Effort Detection
A configurable blocklist was applied to flag known spam patterns. Repetition detection flagged responses consisting of repeated characters or phrases. A minimum effort threshold was enforced per field type based on character count and word count. Responses falling below quality thresholds were flagged for review and excluded from analysis.
Stage 4: Response Normalization
For categorical fields with open text entry (e.g., race/ethnicity), semantically equivalent responses were grouped under canonical labels using configurable normalization rules. Multi-select values stored as semicolon-separated strings were parsed into structured arrays for analysis.
Stage 5: Quality Scoring
Each response received a per-response quality score (0–100) based on response length, apparent effort, and coherence. Completion status was tracked (partial vs. full), and completion scoring recorded how far each respondent progressed through the survey. Analysis was conducted on completed surveys (n=28).
The pipeline produced a structured audit log documenting every action taken, including filtering, normalization, modification, and flagging, with the original value, new value, reason, confidence score, and an indication of whether each action was overridable by a human reviewer.
Prior to deployment, the instrument achieved an internal QA score of 86/100 with all critical issues resolved and launch readiness confirmed across nine criteria.
Post-Collection Analysis Methods and AI Disclosure
No AI, machine learning, or large language model was used in any stage of data processing, qualitative coding, or quantitative analysis. Verified through code-level audit, February 18, 2026.All analysis methods are deterministic and rule-based: keyword matching for themes, word-count sentiment scoring, rule-based archetype classification, configurable quality scoring, and blocklist-based spam detection.
Theme detection uses deterministic keyword matching against predefined lexicons configured for each project's thematic areas. Not machine learning classification.
Sentiment scoring uses positive/negative word-count methods against predefined word lists. Not a machine learning sentiment model.
Respondent segmentation (archetype classification) uses rule-based keyword pattern matching in JSON configuration. Not machine learning clustering or classification.
Quality scoring uses configurable deterministic rules (response length, word count, character count thresholds). Not machine learning prediction.
Spam and gibberish detection uses blocklist terms, regex pattern matching, and repetition detection. Not machine learning content filtering.
Report narratives, executive summaries, and strategic recommendations were drafted collaboratively by the project lead with AI assistance (Claude Code) during dashboard development, then reviewed and approved by the project lead. No AI processes respondent data at runtime to generate text.
All analysis code and methodology documentation is available for audit upon request.
Qualitative responses were reviewed thematically and mapped to the CR-S and Muhammad equity frameworks by the project lead in collaboration with the commissioning consultants. No data imputation was performed.
⓫ Limitations of the Design and Data Collection
As with all research, this study has limitations that should be considered when interpreting the results:
Small sample size
28 completed from ~200 (14% response rate). Quantitative patterns directional only.
Non-probability sample
Voluntary participation. Self-selection bias likely present.
Timing of deployment
Deployed one day before winter break. Narrow completion window likely reduced response rate.
No save-and-resume functionality
Students interrupted mid-survey could not return to complete it.
Open-ended density
16 open-ended questions (36%) exceeds the 25–35% youth guideline. Intentional for depth but may have increased fatigue.
Population composition
All-girls school. Findings should not be extrapolated to co-educational settings.
Mode effects
Novel survey modality for educational research. Response patterns may differ from traditional formats.
Engagement Parameters
Scope, timeline, and logistics reflected exploratory nature rather than a fully resourced project.
Time estimate not displayed
12-minute estimate existed in metadata but was not shown to participants.
Qualitative analysis review
Conducted by project lead with commissioning consultants. Formal independent analyst verification not performed. Protocol being implemented for future engagements.