Open5e Coverage Explorer Open dashboard

MCX · DATA ANALYSIS WITH AI · CASE STUDY

Dense source material.
Clearer VTT decisions.

How I used AI and my visual design background to help independent virtual tabletop developers check D&D source coverage before adding content.

758option names compared
599SRD name matches
159names needing further review
7categories investigated

Verified project snapshot · 20 September 2026 · See the reconciled results

01 / THE PROBLEM

Building a tool shouldn’t start with a maze of sources.

I was building my own VTT and needed a clearer way to investigate which D&D options had corresponding names in the System Reference Document (SRD). I had also seen other developers raise this question on Reddit.

Manually reading dense documentation and cross-referencing lists would take substantial effort. AI helped organize that material and compare names; source evidence, reproducible checks, and my judgment were needed to make the output useful.

The decision this supports

Before adding an option, inspect its SRD name coverage and supporting source, then prioritize anything that needs further investigation.

This is a source-coverage comparison. A matching name does not establish that a VTT implementation, copied rules text, artwork, or branding is legally permitted.

02 / THE FIVE-STEP WORKFLOW

AI did the heavy sorting.
I stayed responsible for the decisions.

Each stage links to project evidence. Early planning documents describe the work at that point; the final verification report governs the corrected findings.

01

Explore

Understand the sources before trusting the comparison.

The project used my own D&D source material rather than the optional synthetic workbook. Player’s Handbook indexes defined the comparison set; SRD 5.2.1 provided the baseline for checking names. Structured JSON files hold the manifest, inventory, comparison, and derived views.

AI contribution

Organized the source inventory, identified file relationships, and surfaced extraction and matching risks: mixed editions, PDF text errors, punctuation, and aliases.

My judgment

I brought the VTT use case and game knowledge, questioned uncertain coverage such as Aasimar, and asked for traceable source references.

Stage outcome: Define a named game option as the unit of analysis and separate evidence of PHB membership from evidence of SRD presence.

02

Frame

Make the audience, question, and useful answer explicit.

The analysis brief asks which current Player’s Handbook character options have corresponding names in SRD 5.2.1, and how that distinction can support VTT product decisions.

Audience
Independent VTT builders and small teams
Comparison
Current PHB / 5.5e → SRD 5.2.1
Measures
Counts, category coverage, unmatched names & aliases
Useful answer
Find an option, inspect evidence, decide what to review next

AI contribution

Helped turn the use case into a scoped analysis brief with metrics, constraints, and required dashboard capabilities.

My judgment

I defined the problem from my own VTT work and approved scope, source traceability, and the intended developer handoff.

What is documented about the framing interview?

I recall giving ChatGPT the rubric and being interviewed before the analysis. The approved brief is available; the original chat transcript has not been recovered. This case study does not present that recollection as a verified transcript.

03

Analyze

Compare conservatively. Investigate apparent gaps.

The analysis normalized names, matched records, reconciled category counts, and applied documented aliases. It aggregated coverage across seven categories and derived the class/species matrix from its component records.

DECISION-RELEVANT FINDING

Base-class coverage does not predict subclass coverage.

All 12 of 12 base classes have SRD name matches, compared with 12 of 48 subclasses. A builder should inspect the specific subclass instead of assuming coverage from its parent class.

AI contribution

Helped extract, normalize, compare, aggregate, and implement repeatable generation and validation scripts.

My judgment

I requested checks on uncertain cases rather than treating familiarity as evidence. Aasimar is a species with no match in this SRD comparison; Bard and College of Lore do have matches.

Competing explanation: A different name may be a renamed counterpart or an extraction error, not a true coverage gap. Exact matching alone is insufficient.

04

Verify

Try to weaken the findings, then correct the work.

The review report documents three independent AI reviewers examining sources, code, and interpretation. The primary agent reproduced findings and reran validation after corrections. This was an adversarial AI review, not an external legal audit.

Material changes recorded during adversarial review
ChallengeCorrectionWhy it matters
Wrong comparison universeRemoved Energy Cell and Spiked Armor; 760 → 758 records.They came from the Basic Rules filter, not the approved PHB set.
Renamed spells looked unmatchedResolved 17 branded names to SRD counterparts.Bigby’s Hand → Arcane Hand is a documented alias, not an automatic absence.
Matching was too permissiveStrengthened equipment checks and documented exceptions.An incidental mention should not be mistaken for a listed equipment entry.
Coverage sounded like permissionNarrowed labels to “Name matched” and “Review needed.”A name comparison cannot settle legal use.

Final control: All 599 positive matches have an SRD PDF page recorded. Equipment membership remains the least certain category because there is no single official total for the chosen subset.

05

Visualize

Make the evidence understandable and usable.

I used my visual design background to guide several rounds of revisions, clarifying the context and hierarchy and paying attention to accessible color choices. The dashboard moves from summary counts to category patterns, searchable evidence, and the class/species matrix.

AI contribution

Helped build the interface and implement revisions, search, filters, source details, and the downloadable audit prompt.

My judgment

I directed visual refinement and the clarity of the data story. Explicit status labels and symbols support meaning alongside color.

A subsequent WCAG 2.2 AA review checked contrast, keyboard access, tables, dialogs, and reflow. Read the checks and their limits. A screen-reader walkthrough and user study have not been completed. No before/after screenshots are included as evidence of the revision history.

03 / OUTCOME & INTENDED IMPACT

From a reference dashboard to a development habit.

The dashboard gives developers a searchable starting point for source review. The downloadable prompt extends the comparison into their own AI workspace: inventory VTT names, cite matching records and SRD pages, and produce a queue of unmatched or ambiguous items.

I tried the prompt on my own early VTT implementation, which currently includes one class. It flagged nothing in that limited trial. My intended practice is to repeat the check before adding classes, subclasses, or other content.

This is a creator-reported trial, not evidence of broad audit accuracy. Time savings and developer adoption have not been measured. Anonymous search analytics were proposed in the brief but are not active.

04 / ABOVE & BEYOND

Extra depth, with evidence attached.

These extensions go beyond a dashboard-only deliverable. Adversarial review itself was required by the assignment; the specific corrections show how thoroughly it was applied.

01 / TRACEABILITY

Inspect the evidence

Every positive match has a recorded SRD page. Developers can trace the answer instead of trusting an unexplained status.

View the comparison ↗
02 / REPRODUCIBILITY

Rebuild the analysis

Versioned inputs, source snapshots, generation scripts, and validation records make the reasoning inspectable.

Read the refresh method ↗
03 / PRACTICAL HANDOFF

Audit your own VTT

A reusable prompt transfers the method into a developer’s workflow and asks for a reproducible exception report.

Find the audit download ↗

05 / RECORDING GUIDE

One story. Four minutes. Your voice.

Use these suggested talking points in your own words. Show the actual work on screen and keep the finished dashboard visible during the demo. Camera optional; record your own voice.

  1. 0:00–0:30

    Start with the problem

    “I’m building a VTT. Other independent developers face the same question: where do I check source coverage before adding D&D content? I wanted to make that research easier.”

    Show: Problem and audience, then dashboard overview.

  2. 0:30–1:10

    Explore → Frame

    “AI helped organize dense reference material. We separated the PHB comparison list from the SRD baseline and defined the audience, scope, and success criteria in an analysis brief.”

    Show: Source inventory and the brief summary.

  3. 1:10–1:50

    Analyze → Verify

    “We compared 758 names. All 12 base classes matched, but only 12 of 48 subclasses did. Adversarial review removed two incorrect equipment entries, resolved 17 renamed spells, and tightened the wording.”

    Show: Correction table and verification evidence. Explain that coverage is not legal clearance.

  4. 1:50–3:05

    Visualize: demonstrate the finished tool

    “My design background guided revisions to context, hierarchy, and color choices. A developer can search a name, see its status, and inspect the source evidence.”

    Show live: Search Aasimar, read its review-needed status, clear the search, then search Bigby’s Hand and show its SRD alias and page. Finish with the matrix and audit prompt download.

  5. 3:05–3:40

    Make your judgment and extra work visible

    “I questioned uncertain cases, asked for source references, and directed the visual revisions. Beyond the dashboard, the project includes page-level evidence, reproducible scripts, and a reusable prompt for another developer’s VTT.”

    Show: The three evidence-backed extensions.

  6. 3:40–4:00

    Close the loop

    “My early one-class trial flagged nothing, so this isn’t proof of a comprehensive audit. The intended impact is a repeatable source check before adding content, with clear evidence and a review queue.”

    End on: The finished dashboard. Tie the result back to independent VTT developers.

06 / SUBMISSION READINESS

Check the rubric against what you actually show.

This is a rehearsal checklist, not a grading prediction. Your checkmarks save only in this browser.

Rubric coverage and evidence to show
CriterionProject evidenceBefore submitting
Problem / Application FitSpecific VTT audience, source-review decision, dense-document use case for AI.Explain why AI helps and state the scope boundary.
Meaningful Use of AIAI organization and comparison; your uncertainty checks, source requests, and design direction.Explain at least one judgment call in your own words.
Artifact QualityWorking dashboard, seven-category dataset, source evidence, audit prompt, validation reports.Rehearse search, evidence links, and the download on screen.
Presentation QualityProblem → five steps → deliverable → intended impact.Watch the final recording: 3–5 minutes, your voice, deliverable visible; then upload it for review.
My final recording check

0 of 7 checked

Evidence boundaries and remaining opportunities
  • The framing interview is recalled by the creator; its original transcript has not been recovered.
  • The one-class VTT trial and design revision rounds are creator-reported. No trial report or before/after design archive is attached.
  • A WCAG 2.2 AA-focused review and fixes are documented in the accessibility report. Real screen-reader validation, cross-browser testing, and user testing remain opportunities.
  • No measured time savings, adoption results, active search analytics, or legal approval are claimed.
  • The project uses its own source-derived dataset stored in JSON. The class’s synthetic workbook was optional; no separate CSV/XLSX submission requirement was supplied.