Skip to content

Refresh surge consistency multiplier

A sub-multiplier on the GSC surge signal that boosts urgency for single-spike surges — they're more likely to be data-quality artifacts and need faster verification than a steady three-month climb.

What it is

After detecting a surge, the code counts how many of the 3 recent months individually exceed 1.5 × avg_prior. The count drives a multiplier applied to surge_strength:

Months above 1.5 × prior Multiplier Meaning
1 of 3 (single spike) 1.3 × Suspicious — could be a bot spike or one-off event; verify fast
2 of 3 (mixed) 1.1 × Partially consistent — mild boost
3 of 3 (monotonic) 1.0 × Steady trend — baseline urgency, no boost

The multiplier also doubles as the single-spike detector for GT tier assignment: consistency ≥ 1.3 forces tier 1 regardless of the priority score.

How it's computed

At processing.py:KB-ANCHOR:refresh-surge-consistency-multiplier. The threshold value avg_prior * 1.5 is the same 1.5× baseline used as the per-month qualifier; the surge itself uses the broader 3-month-average comparison.

Why this choice

A single 8× spike in one of three months looks identical, in average terms, to a 2.7× rise sustained for three months — but operationally they need different responses:

  • Steady climb — the new equilibrium is real; GT will confirm it on the next monthly refresh; capacity allows it to wait.
  • Single spike — could be a bot incursion, an indexing bug, or a transient event (news cycle, viral moment). The signal-to-noise on the rest of the data degrades until we know which one, so the keyword jumps the queue.

The 1.3 / 1.1 / 1.0 ladder is chosen to land single-spike keywords on the tier-1 side of the gt_priority ≥ 4.0 cutoff at moderate surge strength, while leaving monotonic surges in tier 2 unless other signals pile on.

Edge cases

  • Surge from a low baseline — if avg_prior is near zero, every recent month trivially exceeds 1.5 × avg_prior. The surge signal's volume ≥ 50 floor mostly prevents this, but a keyword that just crossed 50 can briefly enter monotonic-3-of-3 territory with all months above the threshold.
  • Equal-magnitude dips and surges — the count is asymmetric: v > threshold only fires for positive spikes. A symmetric dip ("2 of 3 below 0.5 × prior_avg") doesn't get the same single-spike boost, since dips are less commonly bot-driven.

See also