# A09 · From Presence to Revisable Explanation

Original primary-study protocol, NASA stimulus-feasibility audit and reproducible design sensitivity. First author: Agathe Bertin (Hasso Plattner Institute of Design; Santa Clara University). Second author and Principal Investigator: Dr. Freedom Cheteni. Publisher: The Research University. Journal: Journal for School Superintendents. Version 1.1 · Prepared2026-10-16UTC (October15 in California).

## What is completed

- Actual NASA Exoplanet Archive Planetary Systems response retrieved through public TAP, with exact query, timestamp and SHA-256 in retrieval.json.
- Aggregate feasibility audit:6,451 default parameter rows;1,692 core stimulus candidates;1,220 under a stricter two-sided-uncertainty rule. Solution type, projected-separation semantics, missing eccentricity, limit flags and reference identity are checked explicitly.
- Nine artificial-data factorial-design scenarios,10,000 replicates each, with seed, assumptions, null rejection rates and Monte Carlo uncertainty.
- Geometry and estimator code plus six meaningful test cases using synthetic educational fixtures and a tiny synthetic database fixture.
- A prospective32-school/32-instructor blocked2×2 study,128 course sections and3,072 planned learners, with delayed blinded scoring and original draft instruments.

**There are zero actual trial participants.** No educational treatment effect, ethics approval, registration, human review, author approval, recruitment or acceptance is asserted. The astronomical database provides task parameters, not student learning outcomes. The task bank, curriculum validation, full missing-data implementation and study authorization remain prospective.

## Files

- article.md: original scholarly article and complete bibliography.
- study-specification.json: prospective design and governance boundaries.
- instruments.md: original draft prompts, model conventions,0–20 rubric and human scoring process.
- analysis-specification.md: estimand, contrasts, clustering, planned imputation, sensitivity and multiplicity.
- retrieval.json: exact public query, URL, source-response byte count, timestamp and SHA-256.
- stimulus-feasibility-audit.json: executed aggregate counts and selection stages.
- design-analysis.json: actual artificial-data simulations, with Monte Carlo uncertainty.
- research_tools.py: executable local audit, simulation, geometry, allocation and complete-cohort estimator utilities.
- test_research_tools.py: tests; educational fixtures are artificial.
- requirements.txt: exact NumPy/SciPy versions used.
- source-audit.json: item-specific reference evidence levels and rights boundary.

## Reproduce calculations locally

Install requirements in your own isolated Python environment. No participant records are transmitted by the tools. Only the explicitly requested fetch command makes a network call.

```sh
python3 -m pip install -r requirements.txt
python3 research_tools.py design --seed 20261016 --replicates 10000 --output design-reproduced.json
PYTHONDONTWRITEBYTECODE=1 python3 test_research_tools.py
python3 research_tools.py fetch --output nasa-ps-default.csv
python3 research_tools.py audit --input nasa-ps-default.csv --output audit-reproduced.json
```

The live archive changes. The frozen response analyzed in this article has SHA-256 `034f619a7366eea09fdf2a169d9c7c9ca61f926731702f29fe9ff9e61b4b56cd` and1,873,252bytes, retrieved2026-10-16T00:25:01.360504+00:00. Check your locally fetched bytes before claiming an exact snapshot reproduction. A different hash means a different snapshot, even if some counts agree. The raw original response is deliberately not distributed here; aggregate results and its provenance are.

## Optional prospective utilities

A synthetic or authorized local roster CSV requires school_id,instructor_id,section_id. Every instructor/school block must have exactly four globally unique sections. This design permits exactly one instructor per school.

```sh
python3 research_tools.py randomize --input roster.csv --data-kind synthetic --seed 20261016 --output assignment.json
```

Use `--data-kind study` only for separately authorized real data under this32-block protocol; the utility refuses another block count. Its label is an investigator declaration, not a certification of ethics, consent or study conduct. Real allocation requires an independent custodian and concealed seed after baseline roster lock.

The complete-cohort CSV requires school_id,instructor_id,section_id,student_id,immersion,revision,delayed_score. Student IDs must be globally unique pseudonyms; each section has24baseline-cohort scores, each0–20; each instructor/school has one section in every cell. No participant fixture is provided because none has been collected.

```sh
python3 research_tools.py analyze-complete --input synthetic-scores.csv --data-kind synthetic --output estimates.json
```

The estimator rejects missing, nonfinite, duplicate or out-of-range scores and incomplete blocks. It must not be used as a silent complete-case alternative to the declared intention-to-treat missing-data procedure. Imputation, baseline-adjusted models and curriculum validation are specified in the methods document, not implemented or claimed as fitted.

## Rights and credit

The NASA service is publicly queryable and its acknowledgment/citation instructions were inspected. Public access is not represented as an article-wide Creative Commons license. No raw source CSV, archive image, NASA/Caltech mark, original article text or third-party figure is included. Each future selected case must preserve its original parameter-paper attribution and review material-specific reuse conditions. The original code, prose and instruments have distinct authorship from their cited scientific sources; final new-work licensing remains the publisher's separate record.

NASA/Caltech are source-service providers, not coauthors, endorsers or study sponsors. The predecessor conceptual article remains a separate scholarly work; its review does not transfer to this new protocol.

Authorship updated by publisher instruction on 16 October 2026. The study design and executed numerical analyses are unchanged.