Result aggregation

Root aggregations summarizes calculation rows, events, window segments, or request/response results produced by one analysis. For example, first calculate mean power for each loaded interval, then calculate the P95 of the two means, 10 W and 20 W: 19.5 W. Each activity contributes one equally weighted row; activity duration is not used again as a weight. This feature does not load arbitrary historical CSV files or combine independent batch jobs.

Run the complete example

power-statistics.yaml includes ASC input, signals, load rules, per-activity means, and aggregation. Run chronaxis run power-statistics.yaml --output result-statistics from its directory. Locate the aggregation table through this run's manifest outputs: load_mean_p95 is 19.5 W, with two candidates and two included values. Complete actual results

Replace the root aggregations section in that file with this excerpt; its calculations already defines load_mean:

yaml

aggregations:                 # Summarize existing result rows
  - name: load_mean_p95       # Name of this new aggregate result
    source:                   # Select exactly one named result source
      calculation: load_mean  # Mean power for each loaded activity
    operation: quantile       # Quantile of those activity means
    q: 0.95                   # Probability for P95

Choose a source and operation

Sole field in source	Input rows	Allowed operations	Numeric field
calculation: name	Rows of a scalar calculation; select is not allowed	count, sum, mean, variance, quantile, median	Numeric operations infer value; field can be omitted
sequence: name	Discovered requests for this pairing rule	count, sum, mean, variance, quantile, median	Numeric operations use successful delay_ns values; field can be omitted
rule: name	Events of the named rule	count	Omit field
window: name	Final segments of the named window	count	Omit field

count measures objects, not temporal coverage. Numeric operations use eligible rows. Confirmed timeouts and insufficient observation are counted separately; missing delays are never zero-filled.

Fields and combinations

Field	Type, requirement, and default	Meaning and constraints
name	Required unique name	Must not collide with other rule, window, calculation, or related node names
source	Required object	Exactly one field from the table above, naming a source within this analysis
operation	Required string	Choose from the source's operations; sum, mean, variance, quantile, or median summarize numeric values
field	Optional string	Only value for scalar calculations or delay_ns for numeric pairing aggregation; forbidden for count
group_by	String list; default []	[] makes one group; duplicate or unsupported keys are rejected
unit	Optional unit string	Source unit or a compatible target; variance squares the source unit; count accepts only 1
q	Finite number; required only for quantile	Inclusive range 0–1; P95 uses 0.95. Forbidden for other operations, including median
ddof	Integer; variance only, default 0	0 or 1; denominator N-ddof. Insufficient values produce no number; forbidden for other operations
include_partial	Boolean; default false	True is allowed only for numeric calculation aggregation and includes usable partial values; not for count or other sources
partial_policy	report or strict; inherits execution.partial_policy	report may return a partial known value; strict withholds the number when required data are incomplete

Quantiles sort included values and interpolate at position (N-1)×q. Median fixes q=0.5. Exact quantiles retain the selected values and need sorting space within the analysis memory budget.

Grouping keys

All sources allow dataset_id, entity_id, session_id, and their matching calculation_id, sequence_id, rule_id, or window_id. Per-event/per-window calculations also allow parent_id: the event or segment that produced that calculation row. A parent_id group exists only if its event or segment was actually produced. Session calculations cannot group by parent_id. Independent batch jobs form separate groups.

Empty values and completeness

With sufficient observation and no events, count for that event source is zero. A mean with no numeric values is empty, not zero.
An event may exist while its mean's filter selects no samples. That empty mean cannot contribute zero. A complete zero integral or duration is a valid value and can participate.
If the condition signal was interrupted, discovering three events does not prove that no events were missed during the gap. discovery_complete states whether observation is sufficient to identify all candidates; it is separate from each numeric result's completeness.
Partial values are excluded by default. Enabling include_partial does not make missing, invalid, or insufficient-coverage values usable.

The output satisfies candidate_count = included_count + definition_excluded_count + quality_excluded_count. For five requests with four successful delays of 100, 200, 300, and 400 ms and one confirmed timeout, P95 is 385 ms: five candidates, four included, one excluded by definition. If observation ends before the last deadline, that request has insufficient observation instead of a confirmed timeout. Output columns and statuses