Event Scouting

Event Scouting — discovery + tiering

Do BOTH discovery (what exists) and selection (where the target audience actually shows up), producing a tiered spreadsheet + a short summary, and accumulating evaluations in a knowledge base so each run starts ahead of the last.

Core asymmetry to respect: public calendars skew toward large retail/developer events; the valuable small, closed gatherings barely surface in web search. Low web presence is not low value — often the opposite.

Two modes (auto-detect; ask only if genuinely ambiguous)

1. Period scan — a period is given ("Q4", "September"). Default geography = major hubs; narrow if a region is given. Frame results event-first: "this week in this city = this density."
2. Trip mode — a city + date window is given. Parse fuzzy windows ("early / mid / late month") and state the assumed window. Cover the window AND flag near-miss events within ±3 days ("extend the stay one day to catch X"). Never silently widen the window — mark each event in/out and flag near-misses explicitly.

Workflow

1. Scope — detect mode, period/region/window, and the goal (which shapes tiering weight, not the sweep). Don't block on non-essential questions: state assumptions, proceed, append open questions at the end.
2. Load the knowledge base first — reuse recent evaluations; re-verify only this year's dates/venue for known events. Create a KB entry for anything new.
3. Discovery fan-out — parallel searches, one per event category × region, so no single category crowds out the others. Also sweep aggregators, invite-only side events around anchor conferences (this layer is often the primary target, not a follow-up), and non-conference clusters overlapping the window.
4. Check internal records (non-skippable, before tiering) — search your team's history for candidate events + organizers: past attendance and verdicts, invites already in hand, organizer connections. This flips tiers in both directions.
5. Verify + tier — every density claim needs evidence (prior-year speaker/sponsor/attendee lists). Dates MUST come from this year's official source — never recycle last year's; confirm the event actually runs this year (cancelled/stale pages are common). Mark FACT vs. inference vs. not-found — an honest "unverified" beats a wrong claim.
6. Outputs + KB update — spreadsheet + summary, then update the knowledge base.

Tier criteria (two axes: audience density × closed format)

• Tier 1: density verified as FACT (the right people demonstrably attend) AND a 1:1-friendly format (small, invite-only, or a VIP track) — or proven by your own past results.
• Tier 2: density signals but unverified, OR a large event with a separate VIP track.
• Tier 3: topical relevance only (broad/general crowd) — spreadsheet only, omit from summary.
• Boundary rule: a large event's VIP track with FACT-verified density → Tier 1; track exists but density unverified → Tier 2.

Outputs (both required — a chat-only answer is not a valid deliverable)

Spreadsheet

Columns: No. | Event | Tier | Dates (this year, verified) | City/Country | Category | Format (size · invite-only? · VIP track) | Attendee profile (FACT) | Why us (why · who · which goal) | Entry (ticket price · how to apply · deadline) | Internal contact | Website | Source. Rows grouped by tier with colored group-headers (Tier 1 green / 2 yellow / 3 red); unverified cells flagged; a separate legend sheet. Trip mode: an in/out-of-window column.

Summary note (deliver in chat to paste; never auto-send)

• Title + 1-2 sentence scope + verdict ("Tier 1 = N events").
• Per Tier-1 event, 4-5 lines: what it is / attendee profile (FACT) / why it fits the goal / how to attend + deadline. Tier 2 one line each. Tier 3 omitted.
• Trip mode: near-miss flags at the end.
• Close with honest limits (invite-only events don't surface on the web), open questions, next steps.

Common failure modes (each observed in testing — don't repeat)

• Anchoring on famous conferences and deferring the invite-only/side-event layer to "a follow-up round" — that layer is the point.
• One category's queries crowding out the others.
• Treating internal-record search as an optional afterthought — it decides tiers.
• Delivering a chat list only — the spreadsheet + summary are the deliverable.
• Trusting last year's dates or dead events — verify this year's official source.
• Silently widening the requested window instead of near-miss flagging.