Draft Preliminary note on identifiability and PII review · click to expand

This appendix is being circulated to the committee as a draft. It contains material that has not yet completed a final review for personally identifiable information. Before the appendix is finalized for submission, every mention of a named individual will be reviewed against the taxonomy in §B.6.5: students named in instructor-produced materials will be anonymized unless explicit written consent for educational use is documented; named guest speakers will be retained as public professional identities with their professional context attached; Luma-platform workshop feedback will be reviewed for anonymization; and external journalism is retained as already published and consented.

Index
§B.5

Analytic process

The analytic process I apply to my corpus has four components: narrative reconstruction per iteration, cross-iteration pattern matching, anchor-concept development, and member checking. Each is described below.

B.5.1 Narrative reconstruction per iteration

For each of my four iterations I produce a narrative reconstruction grounded in the artifacts specific to that iteration. The reconstruction follows the structure analytic autoethnography requires: it is chronological where chronology matters, it is sourced to specific artifacts, and it foregrounds my position as the instructor making curricular and pedagogical choices. The reconstruction also draws on Schön's (1983) framing of the reflective practitioner: the practice's choices are reconstructed from artifacts that contain "reflection-in-action" traces (the Weekly Updates Prelim Document's "Learned" entries), not retrospectively imposed on the iterations from the outside.

For Iteration 1, the reconstruction draws on the Canvas LMS export (CV-1) for week-by-week curricular structure, on the Weekly Updates Prelim Document (WU-1.W01 through WU-1.W15) for my contemporaneous reflective voice, on the ForeverGold deck (DK-1.FG) for the course-framing architecture, on the Week 1 deck (DK-1.W01) for the opening session's named student outputs, and on the Final Project Requirements (FP-1) for the capstone assignment. The reconstruction is supplemented by the Research Impact Essay (RE), which I wrote near the end of Iteration 1.

For Iteration 2, the reconstruction draws on the Canvas export (CV-2), the syllabus (SY-2), the five iteration-specific decks (DK-2.JAN13 through DK-2.MAR10), the six student teach-out presentations (SP-2.), and the administrative spreadsheets (AS-2.) that document the iteration's guest-lecturer list, teach-out schedule, and final-presentation order.

For Iteration 3, the reconstruction draws on the workshop deck (DK-3), the Luma feedback corpus (LF-3 with sub-IDs LF-3.R01 through LF-3.R29), and the participant roster (LR-3).

For Iteration 4, the reconstruction draws on the workshop deck (DK-4) and the five day-by-day YouTube transcripts (TR-4.D1 through TR-4.D5), which together provide approximately fifty-five thousand words of verbatim teaching delivery.

Each narrative reconstruction is presented in Chapter C (§C.2 through §C.5).

B.5.2 Cross-iteration pattern matching

After reconstructing each iteration individually I conduct a cross-iteration pattern-matching pass. The goal is to surface the regularities and the changes that running the same course four times across different sites and durations exposes.

The pattern-matching pass produces specific findings of the following kinds:

Tool turnover. Comparison of CV-1 against CV-2 produces a documented list of tools added (DeepSeek added in Iteration 2 Week 3 within weeks of its January 2025 public release; AI Agents as a full new module Week 9; NotebookLM moved from student-mentioned in Iteration 1 to dedicated lecture in Iteration 2; Be My AI, Sora, HeyGen, ElevenLabs, Custom GPTs with Wolfram API), tools repositioned (Hugging Face moved from Iteration 1 Week 9 to Iteration 2 Week 7), and tools dropped (Reality Editor present in Iteration 1 Week 10, absent in Iteration 2). The full list is presented in §C.6.1.

Module placement evolution. Same comparison surfaces specific structural shifts: the video module moved from Iteration 1 Week 6 to Iteration 2 Week 14, an eight-week shift; the industry theme moved from Iteration 1 Week 5 to Iteration 2 Week 8. The full list is presented in §C.6.2.

Guest-speaker turnover. Cross-comparison of the guest-speaker rosters surfaces a clean pattern: external guests turn over completely between Iterations 1 and 2 (Pinter, Sieber, Zago, Klassen, Weng all in Iteration 1 but not Iteration 2), while CU-internal speakers are stable but rotate topics (Nolan Brady returns from Iteration 1 to Iteration 2 with a different lecture). This pattern is documented in §C.6.3.

Compression evidence. Comparison of the four iterations on time-on-task produces the headline compression finding: fifteen weeks for Iterations 1 and 2, five days at one hour per day for Iterations 3 and 4. The same four-theme architecture survives both formats. The compression ratio is approximately six to one.

Stable elements. Cross-comparison also surfaces what does not change. The Week 1 prompt-engineering opener is identical in structure across all four iterations. The four-theme architecture is present in every iteration's framing. The Midjourney Self-Portrait assignment recurs.

B.5.3 Anchor-concept development

From the cross-iteration patterns and from my reflective material I develop three anchor concepts, each of which becomes a theoretical claim in Chapter D. The development process is iterative and reflexive: I name a candidate concept, I test it against the artifact corpus, I refine the concept's wording, and I retain it as an anchor only if it withstands sourcing against multiple independent artifacts.

Hallucination-as-pedagogy anchors my first finding. The concept names a reframing of generative-AI hallucination from a system limitation to a teachable pedagogical moment. I tested the concept against four independent sources: my own Iteration 1 Week 1 reflection (WU-1.W01-Q1, "I learned that some of the multiple choice quizzes generated by ChatGPT were not correct and had hallucinations"), the UW KidsTeam children's independent surfacing of "images produced with a third arm" as a concern (KT-THEMES-C5), my public-facing Keep Up Newsletter framing of hallucination as expected behavior (KN-EP1-Q1, "Expect variable results and occasional hallucinations"), and the live workshop delivery in Iteration 4 (TR-4.D1 carries hallucination as a teaching topic). The concept survived testing and is retained.

Compression-as-curriculum-maturation anchors my second finding. The concept names a pattern in which the same content architecture survives a six-fold compression from a fifteen-week semester to a five-day workshop, and in which the compression represents maturation by distillation rather than content loss. I tested the concept against the four iteration artifact sets and against my reflective voice across the cross-iteration material. The concept survived testing and is retained.

Multi-channel teaching practice anchors my third finding. The concept names a configuration in which pioneer instructor practice unfolds simultaneously across multiple delivery channels rather than within a single classroom. I tested the concept against the documented channels: two undergraduate semester courses (CV-1, CV-2), two online workshops (DK-3 with LF-3, DK-4 with TR-4.D5), an HCI grad guest-lecture series (HC series), K-12 outreach (ST-MURAL, KT-DECK), a LinkedIn newsletter (KN-EP series), a podcast (KP-EP series), and a federal research webinar (WB-2026-03-03). All eight channels are documented; the concept survived testing and is retained.

Section D.2, 4.3, and 4.4 develop each anchor concept into a theoretical claim.

B.5.4 Member checking by chair and committee

Analytic autoethnography permits and encourages member checking with collaborators and advisors who have direct knowledge of the setting. My chair has been present in the setting in two ways. He is co-PI on the AI-IRT Seed Grant that funded the research arc (AI-PROPOSAL), and he delivered the DeepSeek guest lecture in Iteration 2 (named in CV-2 and on the Iteration 2 guest-lecturer spreadsheet AS-2.GUESTS, and appearing live in the Iteration 4 Day 1 transcript TR-4.D1 as a guest from CU Boulder). His member-checking role is not extra-textual; it is partially documented in the corpus itself.

My second committee member, Diane Sieber, is also co-PI on the AI-IRT Seed Grant and delivered the writing-with-GenAI guest lecture in Iteration 1 (named in CV-1). Her member-checking role is similarly partially documented.

I treat the supervisory and committee feedback I have received during proposal review and during the dissertation submission process as part of the analytic process rather than as external editorial input. This is consistent with Anderson's framework, which positions the researcher's analytic position as informed by but not subordinate to the surrounding intellectual community.

B.5.5 Artifact provenance and dating

Every artifact cited in this appendix carries provenance: a date or date range and a position within the iteration framework or cross-iteration timeline. The full per-artifact catalog appears as Appendix E, organized by iteration and grouped by artifact category, with brief descriptive entries for every ID the chapters cite.

I use three dating conventions throughout. Exact dates when the artifact is precisely dated (the Aspen Public Radio article AP-2024-05-16 is dated May 16, 2024; the CU RMACC webinar WB-2026-03-03 was delivered March 3, 2026). Approximate dates marked with c. when an internal cue locates the artifact within a season but not a specific day (the Research Impact Essay is dated c. Spring 2024 from its "during this spring semester" language). Undated when no internal cue is available (the DLS Prompt Engineering deck DK-DLS carries no internal date marker; Appendix E flags it as undated).

Provenance integrity is the basis for the trustworthiness claims developed in §B.6. A reader can audit any in-text citation against Appendix E to confirm the artifact's date, iteration position, and descriptive content.

Index