Watches evaluate a metric per time bucket as their owner (SPIKE, DROP,
OUT_OF_RANGE, NEW_CATEGORY, NULL_RATE_CREEP with robust baselines) on a
hidden scheduled process, raise and resolve data.analysis.anomaly alerts
through QAlerts, and offer Evaluate now / Enable / Disable record actions
and a preview endpoint (extension 1.3.0). Pivot and crosstab labels are
human (Yes/No, Average Annual Salary). Contingent fields: a watch's saved
analysis depends on its table, table names use the core tables PVS, and
analyzable/excluded fields pick from the chosen table's fields.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
POST /tables/{table}/pivot (extension 1.1.0): up to 3 row and 2 column
dimensions with date buckets or nice numeric bins, fields from to-one
exposed joins, count/sum/avg/min/max/distinct/percentile measures,
totals, Other-merging by top-N, the privacy guard and a drill per cell.
FieldAnalysisAccessResolver maps fields to FULL / EXACT_MATCH_ONLY /
NONE (core secured-data tiers when present, FieldSecurityLock otherwise);
only FULL fields are analyzable, drills never name others, and exact-run
cache keys include the user's field-access fingerprint.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
The phase (b) analysis engine: typed results (metrics with drill
filters, charts as data), column and table profiles, distinct values and
duplicates, AnalysisSecurity (analyze permission, table READ, personalized
table and field security, analyzable/excluded fields), the effective
per-table config, the minimum group size guard, data versions, async exact
runs cached in data_analysis_run, and DataAnalysisService as the facade.
The AnalysisBackend port has an RC.5 stand-in that pushes down what core
supports today and streams the rest with row limits.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
james.maesandClaude Opus 5.5