Final demonstration and presentation plan¶
How to use: this is the plan for the 20% final demonstration on Fri 2026-12-04. The team confirms the speaker rotation at Sprint 3 planning (Thu 2026-10-29) and finalizes timings by the Sprint 3 check-in on Thu 2026-11-12 (rehearsal 1). The rubric asks the demonstration to show the research question, pipeline, implementation, results, and limitations.
Target length: 12 to 15 minutes, then questions. Every member presents at least one segment.
Structure, timing, and speakers (proposed)¶
| # | Segment | Time | Content | Speaker | Backup |
|---|---|---|---|---|---|
| 1 | Question and motivation | 1.5 min | The one-sentence P-01 research question, why copied agent skills matter for AI-native software engineering, the competing explanation tested | Jerad Dunne | Leticia Aderhold |
| 2 | Data | 2 min | GitSkills and the exact snapshot (release, DOI, MANIFEST hash), the sample and why it is defensible, unit of analysis, key variables | Leticia Aderhold | Hina Kramer |
| 3 | Pipeline, live | 3.5 min | Fresh clone in a clean environment, make pipeline, stage outputs and passing tests, the scanner and variant-linking method |
Hina Kramer | Jerad Dunne |
| 4 | Results | 3 min | RQ1 prevalence, RQ2 reach, RQ3 variant drift, and one robustness check; observations only, numbers read from generated artifacts | TBD (D-025) | Leticia Aderhold |
| 5 | Interpretation and limitations | 2 min | What the evidence supports and what it does not; top three threats to validity; inter-rater agreement and error analysis examples | Leticia Aderhold | TBD (D-025) |
| 6 | Process story with Lean Six Sigma metrics | 2 min | Velocity, cycle time, and contribution balance across sprints; the most important process improvement with before and after numbers | TBD (D-025) | Hina Kramer |
| 7 | Close and questions | 0.5 min | One-sentence answer, one-sentence next step, release tag and repository URL; the Product Owner routes questions to the member closest to the topic | Hina Kramer (Finalization Product Owner) | Jerad Dunne |
Reproducibility is shown by running the demo from a fresh clone rather than a working copy.
Live demo checklist¶
- Laptop with the tagged release cloned fresh into an empty folder the morning of the talk.
- Dependencies installed and
make testgreen before the session. - Sample data present at the documented path; no network needed during the demo.
- Terminal font enlarged; commands typed from a visible cheat sheet, not memory.
- Output files opened directly from
results/andfigures/after the run. - Timer visible to the speaker.
Backup plan if the demo fails¶
- Pre-recorded run: a screen recording of
make pipelineandmake figureson the tagged release, stored atreport/demo-recording.mp4(or linked in the release notes). Recorded at rehearsal 3 on Thu 2026-12-03. - Static figures: the committed
figures/PNGs andresults/tables in the slides, with the producing script name on each slide. - Decision rule: if the live run has not produced the first output within 90 seconds, the speaker says "switching to the recorded run" and continues. No debugging on stage.
Rehearsals¶
| Rehearsal | Date | Goal | Timed length | Fixes |
|---|---|---|---|---|
| 1 | Thu 2026-11-12 | Structure, timing, speaker hand-offs; slides draft | ||
| 2 | Mon 2026-11-30 | Full run with live demo on a fresh clone | ||
| 3 | Thu 2026-12-03 | Final run; test the backup path; record the demo |
Slide checklist¶
- Title slide: project title, the four member names, Group 9, course, date, repository URL and release tag.
- Research question slide with the competing explanation.
- Data slide with snapshot statement and sample description.
- Pipeline diagram (stages, inputs, outputs, tests).
- Results slides: each figure or table captioned with the script that produced it.
- Robustness or subgroup slide.
- Limitations and threats slide, including rater agreement.
- Process slide: KPI trend chart from
docs/lean-six-sigma/metrics/. - Most important process improvement slide with before and after numbers.
- AI-use disclosure line on the closing slide.
- Every number on a slide can be traced to a file in the repository.
- Slides exported to PDF and attached to the GitHub release.
Likely questions to prepare¶
- Why this sample and not the full dataset? What would change with the full dataset?
- How do you know the scanner is correct? What was the precision per rule and the inter-rater agreement?
- What is the simplest alternative explanation for the main result?
- How was the work divided, and how did reviews work across the team?
- Which decision would you change if you started again?