service:next-prototypes, env:prod) and Amplitude (source = labs) numbers behind the 002 analysis.amplitude_id.@event:trips.searched env:prod — the MCP search-trips tool, the path the chat flow uses.
| Bucket | count |
|---|---|
Total prod trips.searched | 458 |
Returned trips (@trip_count:>0) | 323 |
Zero trips (@trip_count:0) | 135 |
| Zero-result rate | 29.5% |
Compare 001: 168/169 = 99.4% zero. Remaining zeros are largely real no-service corridors (Tampa→Pittsburgh, Minneapolis→Portland ME, Texarkana→Dallas, Waterloo IN→Chicago, a Coron/Palawan Philippines origin).
@event:search.completed env:prod returns only the once-daily synthetic Boston→NYC canary.
search.completed: BOSSST → NYCPAS (0 trips) custom.failed_carrier_count: 7 (GLI, MGL, PPP, BLT, AMT, LML, TDS) custom.ok_carrier_count: 0 custom.total_trips: 0 env: prod | pilot | dev — same signature across all prod occurrences this window: 9 (≈1/day)
Real-user classic handoff (Search Submitted) resolves depart_wcityid on only 2 / 101, so users barely use this path; the win came from the chat path resolving origins via nearest hub instead.
@event:llm.prompted env:prod — 837 events; sampled cost_usd ~$0.0078–0.0082/call.
| Metric | value |
|---|---|
| Prod prompts | 837 |
| Model | claude-haiku-4-5-20251001 |
| Cost / call (sampled) | ~$0.008 |
| Est. window spend | ~$6.70 |
Hallucinations (@hallucination_count:>0) | 1 (Jun 8, old build) |
| Latency >3s / >5s | 56 / 20 |
| Worst latency | 8,899 ms |
| Pattern | count | note |
|---|---|---|
[Experiment] Fetch failed: Socket connection timed out | 7 | Amplitude variant fetch; intermittent |
Fetch error response: status=503 | ~5 | Jun 10 blip, self-resolved |
mcp.no_structured_content | 3 | tool returned empty |
wapi.request_failed (predictions.json) | 2 | autocomplete; no auth recurrence |
| event | n | note |
|---|---|---|
| Explore Page Interacted | 1,472 | umbrella |
| [Amplitude] Page Viewed | 1,298 | auto |
| LLM Prompted | 766 | ≈ Datadog llm.prompted (837 prod) |
| session_start | 108 | auto |
| Search Submitted | 101 | classic handoff |
| Trip Selected | 73 | NEW — trips[] + batch_id + results_count, 100% fill; trip_id 0/73 |
| session_end | 49 | auto |
Legacy events (Chiku Message Sent, Chiku Destinations Lit, Inspiration Origin Detect Failed): 0 firings — retired confirmed.
| signal | value |
|---|---|
| Explore Page Interacted by type | destinations_suggested 645 · destination_selected 445 · trip_understanding_opened 173 · origin_set_from_autocomplete 125 · use_my_location 48 · date_edited 24 · constraint_edited 12 |
| Trip Selected results_count | all 73 > 0 (mostly 5; none zero) |
| destination_selected origin resolution | origin_wcityid 24/445 · nearest_hub 102/445 |
| Search Submitted depart_wcityid | 2/101 (arrive 101/101) |
| date_source on picks | "this weekend" 213 · default-2wk 117 · "your pick"/high-conf 52 · explicit dates (June 15th, Wed Jul 15, …) ~80 total high-conf |
| pick_source | chat_card 445 / 445 (map 0) |
| distinct cities suggested | 63 (was 16) |
| conversation depth | turn1 424 · turn2 183 · turn3 82 · turn4 35 · turn5 18 · turn6+ 24 (45% follow-ups) |
| device | phone 518 · desktop 233 · tablet 15 |
| utm_source | direct 973 · google 271 · bing 32 · blueshift 18 · chatgpt.com 4 (NEW) |
| city | suggested | selected | conv |
|---|---|---|---|
| Miami | 147 | 40 | 27% |
| Boston | 140 | 28 | 20% |
| New York | 134 | 39 | 29% |
| Chicago | 131 | 36 | 27% |
| Montreal | 125 | 19 | 15% |
| Philadelphia | 120 | 31 | 26% |
| Atlanta | 78 | 35 | 45% |
| New Orleans | 76 | 21 | 28% |
| Denver | 43 | 17 | 40% |
| Nashville | 28 | 15 | 54% |
Newly-unlocked cities (Nashville, Atlanta, Denver, New Orleans) convert highest once shown; Boston / Montreal over-suggested.
| theme | matches |
|---|---|
| Vibe / inspiration | 93 |
| Budget-led | 53 |
| Date / timing | 41 |
| Train / Amtrak / rail | 34 |
| Nearest stop / station | 29 |
| Off-network / international | 24 |
| Support / refund / policy | 6 |
Themes overlap (a prompt can match several); counts are directional, not a partition.