Fetches current listings and uses an AI-arm to flag listings that represent notably good value for their section. Goes beyond raw threshold checks.

Good-Seat AI Watch

Fetches the cheapest current listings for an event and uses an ai: arm to classify whether any of them is notably good value FOR ITS SECTION. Goes beyond raw price-vs-threshold checks — a $40 upper-corner seat isn't a deal, but a $90 lower-bowl seat (typical floor $150+) is.

What it does

On each fire:

The check calls a ticket-pricing endpoint that returns the cheapest 20 listings as a JSON array, published as {{check.listingsJson}}.
The when: block has ONE arm — an ai: arm whose prompt: - Describes typical price floors per section (lower bowl, mezz, upper). - States the value criteria (≥ 30% below typical floor, decent section, no obstruction tags). - Asks the AI to return triggered=true only if at least one listing matches ALL criteria.
If the AI says yes, notify with the AI's reasoning + the full listings array. Otherwise the else: arm matches and nothing fires.

Why an AI arm here

The "is this seat a deal?" decision can't be captured cleanly in a condition: expression because:

Price-per-section matters, not absolute price.
"Limited view" / "partial view" tags should disqualify even cheap listings.
Different venues have different price floors — a New York lower bowl is not a Cleveland lower bowl.

A literal expression covering all that becomes a 5-clause chain that misses edge cases. The AI handles it in one prompt at the cost of ONE LLM call per fire.

Data-source setup

Same scraper / endpoint as event-price-watch, with ONE additional requirement: the response must include listingsJson — either a JSON-array string or a structured array — containing the cheapest N listings. The AI reads this directly via Mustache interpolation in the prompt.

A minimal listing record looks like:

{ "section": "108", "row": "12", "price": 95, "note": "" }

Add note for obstruction tags ("limited view"), delivery for ticket delivery method, etc. — the AI consumes whatever you give it.

Expected published bindings:

Binding	Expected value
`{{check.eventName}}`	Display name of the event.
`{{check.eventDate}}`	Event date for the AI prompt.
`{{check.listingsJson}}`	JSON-encoded array of the cheapest N listings.
`{{ai_reasoning}}`	Set by the AI-arm; explains the match.

Edit before installing

In monitor.yaml:

check.args.url — replace with your scraper / API endpoint. Substitute REPLACE_EVENT_ID.
The AI prompt — adjust the typical-price-floor table for your venue category (NHL > MLB > MLS for floors; arena vs stadium also matters). Be specific about what counts and what doesn't.
when[0].do[0].body — adjust copy.

Schedule guidance: a few times per day. Each fire is one LLM call so cost adds up at high cadence. Daily + within 24h of event = cheaper rule-of- thumb.

Requirements

AI assistant configured (Server Settings → AI Assistant). If no AI is configured, the ai: arm returns false safely (no false fires).

Tools used

http — core InTouch HTTP task type.
AI assistant (any provider — Anthropic, OpenAI, Gemini, Ollama).

Pattern showcase

This is the AI-judged value pattern. Use it whenever the "is this a deal?" judgment is fuzzy enough that a literal condition can't capture it, but the data fits in a small JSON blob the AI can read. Same shape works for: cheap flights in your routes, RFQs that look worth pursuing, support tickets that look like real fires, etc.

Related monitors

event-price-watch — raw threshold, no AI.
gameday-firesale-watch — day-of relative discount.
hn-keyword-watch — different shape, same ai-arm pattern.