Recursively generate a dictionary representation of the model, optionally specifying which fields to include or exclude.
count is how many root traces (conversation turns) fall in this thread for the
query time range.
feedback_stats is the aggregated feedback across traces in the thread, keyed
by feedback key; shape matches feedback_stats on a single run.
first_inputs is a truncated preview of inputs from the earliest trace in the
thread for the query window.
first_trace_id is the root trace UUID for the chronologically first trace in
the query time window.
last_error is a short error summary from the most recent failing trace in the
thread. Absent when there is no error in the window.
last_outputs is a truncated preview of outputs from the latest trace in the
thread for the query window.
last_trace_id is the root trace UUID for the chronologically last trace in the
query time window.
latency_p50 is the approximate median end-to-end latency of traces in the
thread, in seconds.
latency_p99 is the approximate 99th percentile end-to-end latency of traces in
the thread, in seconds.
max_start_time is the latest trace start time in the thread (RFC3339
date-time).
min_start_time is the earliest trace start time in the thread (RFC3339
date-time).
num_errored_turns is the count of root traces in the thread (within the query
window) whose status was an error.
start_time is a reference start time for this row (RFC3339 date-time), such as
for sorting.
thread_id identifies this conversation thread within the project from the
request body project_id.
total_cost is the sum of estimated USD cost across those traces.
total_cost_details sums per-category estimated USD cost across traces in the
thread. Keys mirror total_token_details.
Example: {"cache_read": 0.012, "reasoning": 0.008}.
total_token_details sums per-category token counts across traces in the
thread.
Keys are model-specific category names (for example cache_read, cache_write,
reasoning, audio).
Example: {"cache_read": 400, "reasoning": 120}.
total_tokens is the sum of token usage across those traces.
trace_id is a representative root trace UUID when the summary includes one,
for example for deep links.
