Guidance Rule Statistics

Generated: 2026-05-24 21:12:39 UTC

This page measures how the translation-guidance rules are behaving as reusable editorial knowledge. The denominators are explicit: scan coverage is based on headwords with guidance scan evidence, and rule coverage is based on active, non-retired rules with matched evidence.

Introduction Date Basis

Rules imported from Gabe's spreadsheet can predate their database row. Estimated dates use the earliest Gabe-linked kappa human translation whose headword is matched by that rule; rules without that evidence are kept separate from actual timestamps.

Estimated Rules by Introduction Date

Butterfly Collecting Estimate

Here each guidance rule is treated like a species and each scanned headword is a collecting site. Rules that appear only once or twice drive the estimate of how many additional rules are likely still unseen.

Chao2 currently estimates 10.56 unseen active firing rules beyond the 95 active rules already observed in guidance-match evidence.

Kappa Rule Accumulation

This uses the dated kappa human translations as the collecting sequence. Each step is one Gabe-linked kappa headword in timestamp order. The upper curve shows cumulative distinct active rules first observed in that sample; the lower panel shows new-rule bursts and the number of new rules seen in the trailing 10 sampled entries. The current sample observes 89 distinct active rules across 86 dated kappa entries.

Estimate by Rule Kind

Zipf-Like Rank Frequency

Rules are ranked by the number of distinct headwords where they fire. A Zipf-like distribution would appear close to a straight line on the log-log plot, with slope near -1.

Top-N Headword Coverage

This asks how much of the scanned corpus is covered if we keep only the most frequently firing rules.

Top Firing Rules