Page MenuHomePhabricator

Instrumentation Needs Assessment & Event/Property Rewrite for Hybrid Search MVP
Open, Needs TriagePublic

Description

Background

This task evaluates all instrumentation requirements described in the Semantic Search MVP Measurement Plan and rewrites placeholder (green) events/properties so they align with the Android Apps analytics system.

Scope of Work

1. Review Measurement Requirements (Pages 1–7)

Identify instrumentation required to evaluate:

  • Search session initiation and exit
  • Query submission and reformulation
  • Result impressions and clicks
  • Exploration and reformulation labeling
  • Good abandonment
  • In-app satisfaction micro-surveys
  • Seven-day search return behaviors
  • Latency and performance guardrails
  • Per-result metadata (rank, provenance, snippet type, retrieval mode)

2. Map PDF “Placeholder” Events to Apps System

Review the Measurement Plan’s core events:

  • search_session_start
  • query_submitted
  • search_results_impression
  • search_result_click
  • search_query_reformulated
  • search_exit
  • search_good_abandonment_candidate

For each:

  • Determine whether an equivalent Apps event already exists
  • Identify naming updates needed
  • Identify if it must be emitted client-side, server-side, or both

3. Rewrite Placeholder Event Properties to Apps Schema

Review all Required Event Properties (Pages 4–5) and rewrite them to match Apps logging conventions.

Property groups to rewrite:

  • '''Session metadata''' (session_id, user_id_hashed, entry_point, etc.)
  • '''Query metadata''' (query_string_hashed, tokens, language, similarity metrics)
  • '''Retrieval/API metadata''' (retrieval_mode, latency_ms, fallback_triggered)
  • '''Results list metadata''' (snippet_type, provenance_label_shown, rank, snippet_length_chars)
  • '''Click metadata''' (clicked_result_id, scroll_depth_at_click, dwell_time_on_article_ms)
  • '''Exploration vs reformulation''' signals
  • '''Exit metadata''' (exit_reason, queries_in_session_count, session_length_ms)
  • '''Performance/guardrail metadata''' (api_timeout_occurred, network_type, cost_bucket)

For each property:

  • Indicate whether it is an existing field, renamed field, new addition, or not feasible
  • Propose final property names consistent with Android logging standards

4. Evaluate Instrumentation Needed for all Metrics

Map the Measurement Plan’s metrics to required events and properties, including:

  • Search sessions initiated per unique user
  • Avg. session length
  • Time-to-click
  • Reformulation rate
  • Good abandonment rate
  • Top-results CTR
  • Satisfaction and perceived relevance
  • Seven-day search return rate
  • Latency, fallback, and error tracking

6. Identify Missing Instrumentation

Highlight missing fields needed for:

  • Ranking mode (lexical / semantic / hybrid)
  • Provenance indicator exposure
  • Snippet-length variants
  • Semantic → lexical fallback detection
  • Per-result metadata arrays
  • Scroll depth tracking
  • Observation window compliance for “good abandonment”
  • Other identified metrics

7. Produce Final Instrumentation Specification (Deliverable)

The final document should include:

  • Revised event and property names
  • Event triggers and firing conditions
  • Required timestamps
  • Cross-join schema
  • Open questions for Engineering
  • A final logging schema proposal for the MVP
  • Recommendations for placeholder goals (e.g. X%)

Deliverables

  • Annotated review of Measurement Plan instrumentation requirements
  • Full event & property rewrite table (placeholder → App final name)
  • Gap analysis (missing fields vs. available fields)
  • Proposed Android-side logging schema
  • Proposed server-side logging schema updates
  • Metric-to-event/property mapping table
  • Recommended goals for placeholder metrics based on our existing growth rates

Event Timeline