SEIS 666 Track B — Self-evaluation
Mapped against the Spring 2026 project description. Cited where verified; honest where pending.
Track B deliverables — line-by-line
| Track B requirement |
Wearly's evidence |
Status |
| Working agent built with Claude Code / API / framework |
Built with Claude Code (Claude Opus 4.7). Live Streamlit Cloud + CLI. |
Exceeded |
| ≥ 2 external tools |
4 external + 5 internal tools (calendar, weather API, wardrobe, color; fit, history, shopping, graph, backup). |
2× minimum |
| Documented decision tree / process flow |
7 named steps + workflow_diagram.md (Mermaid) + interactive in-app pyvis graph + in-book vis-network.js viewer + Learning Graph DAG. |
Exceeded |
| Error handling for unexpected input |
Weather fallback, empty-pool relaxation, rejection loop, atomic JSON writes, body-positive contract refusal path. Every tool has a documented fallback. |
Met |
| Before/after comparison |
CLI python main.py --compare + UI Before/After Demo screen + quantified delta table (see §3 of demo_script.md). |
Exceeded |
| 10-min presentation |
demo_script.md with full timing, pre-baked Q&A, run sheet. |
Met |
Section 5 rubric — projected scoring
| Criteria |
Weight |
Wearly's evidence |
| Working System |
30% |
Live Streamlit Cloud + CLI + 207 passing tests + 4 external tools |
| Domain Depth |
20% |
12-chapter Intelligent Book + 8 Skill rule packs + 9 evidence categories + 10 verified primary citations + body-positive contract |
| Before/After Delta |
20% |
Quantified table (15 min → 30 sec; 0 → 7-12 reasoning lines; 0 → 10 citations) |
| Documentation |
15% |
MkDocs site, README, architecture, schemas, demo script, citation chain, business-model doc |
| Presentation |
15% |
Pending May 14 — rehearse against demo_script.md |
Section 2 "Best Practices" alignment
| # |
Principle |
Wearly |
| 1 |
Structure beats prompts |
✅✅✅ Skill rule packs + KG + Learning Graph + Evidence chain |
| 2 |
Show the delta |
✅ Two surfaces (CLI flag + UI screen) + numeric table |
| 3 |
Pick a domain you know |
✅ Personal styling — relatable, no jargon |
| 4 |
Build, don't theorize |
✅ Live app + live book — no PowerPoint hypotheticals |
| 5 |
Use AI to build the project |
✅ Built with Claude Code, documented |
| 6 |
Who would pay for this |
✅ docs/business-model.md — three buyers identified |
| 7 |
Audit trail matters |
✅✅✅ rule_refs.py registry + cite() + reasoning-line tags |
| 8 |
Start small, go deep |
✅ 7 well-scoped steps, deep on each |
Why Wearly is an agent
| Property |
Chatbot |
True Agent |
Wearly |
| Workflow has named steps in defined order |
❌ |
✅ |
✅ 7 steps |
| Uses external tools |
usually ❌ |
✅ |
✅ live weather API + 3 others |
| Branches on tool data |
❌ |
✅ |
✅ dress vs separates, gym branching, outerwear injection, gap detection |
| Pursues a goal beyond responding |
❌ |
✅ |
✅ complete appropriate outfit |
| Recovers from unexpected input |
❌ |
✅ |
✅ documented fallback per tool |
| Closed feedback loop |
❌ |
✅ |
✅ reject + regenerate |
| Persistent memory across sessions |
❌ |
✅ |
✅ wear_history.json |
| Outputs traceable to structured rules |
❌ |
✅ |
✅ [occasion-rules#R3] tags |
| Multi-tool orchestration in one task |
❌ |
✅ |
✅ 4 tools per recommendation |
Verdict: Unambiguous. Wearly satisfies every hallmark of an agent.
How Wearly compares to the "What Great Projects Look Like" examples (Section 6)
| Their A-grade example |
Wearly's parallel |
| Legal Discovery Agent: graph traversal + reasoning paths |
Wearly KG + citation chain — every reasoning line points to a graph entity + a numbered rule |
| Supply Chain Risk Graph: cascade analysis |
Wearly's User → Event → Recommendation → Items → Gap → Suggestion cascade |
| Code Review Agent: read input + check against rules + write review |
Wearly: read calendar/weather + check against 8 rule packs + write reasoned outfit |
| Healthcare Billing Optimizer: codes + payer rules + audit trail |
Wearly: occasion + season + fit + color + history + audit trail with verified citations |
| Real Estate Analyzer: scrape + analyze + memo |
Wearly: pull context + run 7 steps + produce reasoning trail + shopping memo |
Wearly maps cleanly onto every example. The audit-trail-with-citations layer goes beyond what any single example specifies.
What Wearly delivers beyond the Track B baseline
Track B did not require any of the following — Wearly delivers each anyway:
- Track-A-level Knowledge Graph (9 entity types, formal schema, interactive viewers in both app and book).
- Intelligent Textbook — 12 chapters + 28-concept Learning Graph with Bloom taxonomy + p5.js micro-sim.
- Agent Skill package conforming to agentskills.io Advanced tier (YAML frontmatter + executable validator).
- Citation chain end-to-end (evidence → rule pack → rule ID → reasoning line), tested.
- 10 verified primary citations (Springer, ACM, IEEE, ECCV, ISO, EUR-Lex).
- Body-positive language contract enforced twice (runtime + docs).
- Compact-KG export per outfit (instructor's "save the session as a compact KG" idea, delivered).
- "Today's context" free-text input (Leah's life-context pattern, applied).
- "Ask your wardrobe" pre-baked queries (Leah's KG-traversal pattern, applied).
- Auto-deployed GitHub Pages site with GH Actions.
Honest pending items
- Demo rehearsal —
demo_script.md is now complete but unrehearsed. Practice runs recommended.
- Wardrobe-management research citation (Evidence §3.7) — still
general practice; no peer-reviewed source for wear-history rotation has been located.
- CCPA/CPRA references (Evidence §3.9) — only relevant if jurisdiction expands.
- Fashion-specific XAI user study — still pending across §3.1 and §3.5.
Repo + live links