`processed_growth`¶

SQL type: SimpleAggregateFunction(anyLast, Nullable(String)) — JSON Table: keywords.keywords_data_local Last validated: 2026-05-21

What it is¶

Growth metrics + classification for a keyword. JSON containing year-over-year and rolling-window growth rates, a trend-strength score, a seasonality flag, and a categorical growth_class tag (one of exploding / growing / stable / declining / crashing / gated).

How it's computed¶

Computed by build_growth_json() (function defined around processing.py:L459) over the combined_volume_data_sorted series — i.e. the blended trend that becomes processed_volume_trend.

The hybrid growth system has three tiers (memory: archived hybrid_growth.md covers the full logic, decision nodes will land in Phase 3):

Tier 1 (< 6 months of history) — month-over-month
Tier 2 (6–23 months) — rolling-window comparisons
Tier 3 (24+ months) — year-over-year with seasonality detection

Seasonality detection routes a keyword's growth calculation differently — seasonal keywords compare same-month YoY rather than rolling-window. Detected via lag-12 ACF (processing.py:KB-ANCHOR:seasonal-route-acf).

Gate¶

processing.py:KB-ANCHOR:growth-metrics-display-gate: growth metrics are computed only when processed_keyword_volume > 200 (the strict gate — no JS branch, unlike the trend display gate). Sub-200 keywords get an empty/null growth JSON so downstream consumers (extract_gt_year_keywords, pe_update, bot_scan, demo growth API) ignore them.

This intentionally matches prod's display threshold to avoid showing growth tags on keywords whose volume isn't displayed.

Schema (representative)¶

{
  "yoy_3m": 45.2,
  "yoy_6m": 32.1,
  "yoy_12m": 18.7,
  "growth_3m": 12.3,
  "growth_6m": 8.4,
  "trend_strength": 0.65,
  "is_seasonal": false,
  "growth_class": "growing"
}

The exact keys evolve — build_growth_json is the authoritative shape definition.

Edge cases¶

Sub-200 keywords → empty JSON or null. Downstream code must handle this — recent versions of the consumers do.
Bot-corrected keywords — growth is computed over the repaired trend, not the raw bot-inflated one. A keyword whose spikes were corrected will show calmer growth than its raw GSC would imply.
Trend-type routing — keywords whose trend-type-classifier outputs lumpy / isolated_event / dead skip the forecast but still get growth metrics over the observed portion.

Downstream¶

The product surfaces the growth_class tag and the YoY percentages as the "Growth" panel.
pe_update.py reads growth_3m for the growth-weighted ranking score: score = volume × (1 + clip(growth_3m, -50, 200) / 100) (processing.py:KB-ANCHOR:pe-growth-weighted-score).
Global aggregation produces processed_growth_global over the country-summed trend.

processed_growth¶