# Go Do Stuff: Complete AI & Agent Reference (`llms-full.txt`)

> Go Do Stuff is an agentic local event discovery and taste-modeling web application. Its core thesis is that the product is not a static event directory with category filters; the core product is a continually improving, human-readable model (`preferences.md`) of what kinds of real-world experiences a specific person finds worthwhile.

---

## 1. The Problem Go Do Stuff Solves

Mainstream event platforms (Meetup, Eventbrite, generic city guides) rank by popularity and broad tags, surfacing the same repetitive activities: running clubs, rooftop cocktail parties, park yoga, corporate networking mixers, and massive commercial festivals. Users looking for genuine community, hands-on workshops, niche subcultures, or low-pressure small-group social activities struggle to find them, even in vibrant cities full of independent venues, maker spaces, museums, tool libraries, and neighborhood collectives.

## 2. How Go Do Stuff Works

1. **Taste Profile Onboarding (`/preferences`)**:
   - Users chat (via text or voice dictation) with the Taste Profile Coach to establish their city/radius, hard exclusions (semantic category filters), and underlying social/experiential goals.
   - When a user mentions a specific local venue or past event they enjoyed (e.g., *Free Oakland Up* or *The Crucible*), the coach uses Google Search + URL Context grounding to look up the venue/event and extract the underlying experiential dimensions.
2. **Multi-Bucket Grounded Search (`/discover`)**:
   - Given the user's active `preferences.md` and search prompt, Go Do Stuff resolves relative dates (e.g., `"this weekend"`) into concrete calendar dates and generates targeted search queries across three buckets:
     - **60-70% Exploitation**: High-confidence matches aligned with established preferences.
     - **20-30% Adjacent Exploration**: Neighboring domains sharing the same experiential/social mechanics (e.g., if a user likes DIY creative reuse -> community printmaking, ceramics firing, repair cafes, or tabletop RPG drop-ins).
     - **10% Plausible Wildcards**: Unexpected picks with a defensible hypothesis.
   - Executes live Google Search + URL Context deep-reading of local aggregator calendars, independent venue schedules, maker studios, libraries, and community organizations.
3. **Single-Card Stack Triage & Feedback Synthesis**:
   - Candidate events are verified (location, date/time, exclusion compliance) and presented one card at a time in a focused stack.
   - Each card provides a direct venue link, date/time, price, a concrete non-marketing description of what actually happens there, an explicit *"Why it fits you"* explanation, and verification caveats.
   - Users react with:
     - `Going` (adds to Upcoming Schedule)
     - `Interested` (saves to Candidate Backlog for future weeks)
     - `Not for Me` (aesthetic or social mismatch; refines exclusions and latent boundaries)
     - `Bad Timing / Hold` (logistical or seasonal pass; suppresses for the current window without penalizing the underlying activity type)
4. **Post-Event Outcome Loop (`/schedule`)**:
   - After an event's date passes, users record whether they `Attended & Enjoyed`, `Attended (Mixed)`, or `Skipped`, along with notes on how easy it was to talk to people and whether they would go again as a regular. Real-world attendance outcomes carry the highest weight in evolving `preferences.md`.

---

## 3. The 7-Section `preferences.md` Schema

Every user profile is stored and exportable as a 7-section Markdown document:
1. `## 1. Current Reframed Goal`: A one-sentence thesis statement of what the user is looking for right now.
2. `## 2. Defaults & Hard Constraints`: Default location, travel radius, primary goal, and demographic filters.
3. `## 3. Hard Exclusions (Semantic Category Filters)`: Categories and formats to filter out semantically.
4. `## 4. Latent & Experiential Preferences (The "Why")`: Table of cross-cutting dimensions (e.g., *Participatory doing > watching*, *Small group*, *Shared task*, *Organic weirdness > manufactured fun*) with confidence, origin, and supporting quotes.
5. `## 5. Positive Interests & Anchors`: Specific activities and real-world anchor events/venues.
6. `## 6. Uncertain Hypotheses & Untried Candidates`: Planned upcoming events and a backlog of recurring local candidates to try later.
7. `## 7. Event & Feedback History`: Verified real-world attendance outcomes, logistical holds, and rejected items.