How the Activity Index works.

What it measures — and does not

It counts dated, first-party-sourced changes in ToolMarketCap’s change log: releases, pricing and plan changes, availability and platform changes, and ownership or name changes. Nothing else enters it.

It does not measure popularity, adoption, usage, market share, revenue, valuation, traffic, sentiment, or product quality. A higher value is not better: a tool that ships rarely and deliberately has a low value by design. It is not a ranking of tools.

The formula

Activity = round(100 × (0.7 × Volume + 0.3 × Breadth))

window     the last 90 whole UTC days, today included
weight     w(day) = 0.5 ^ (age in days / 30)
Volume     min(1, ln(1 + V) / ln(1 + 4))
           V = Σ w over distinct days with at least one change
Breadth    min(1, (Σ over kinds of the largest w in that kind) / 3)
           kinds: pricing & plans, releases, availability & platforms, ownership & names

• Recent active days (weight 70). A day with any number of changes counts once, so a burst of changelog entries on one day cannot inflate the value. Newer days weigh more: a change 30 days old counts half as much as one today. Volume grows logarithmically and reaches its maximum at 4 fully weighted active days, so a very dense changelog cannot run away with the scale.
• Kinds of change (weight 30). Activity that spans releases, pricing, availability, and ownership is broader than repetitions of one kind. Each kind contributes its most recent day’s weight; Breadth reaches its maximum at 3 kinds.
• No cross-tool normalization. Values are not percentiles or ranks. A tool’s value moves only when its own record moves, so a value stays comparable over time within a methodology version.
• No editorial adjustment. There are no per-tool boosts, penalties, or overrides. Corrections happen only by correcting the record.

Confidence

Confidence describes ToolMarketCap’s record, not the tool: how sure we are that the change log reflects the tool’s recent activity.

A value of 0 is different from no value: it means no verified change in 90 days on a record that covers the window.

Worked example — Replit, 6 Nov 2026

Chosen by rule, not by name: the tool with the highest observed activity today whose record has high confidence and at least two kinds of change. Computed live from the same records shown on its profile.

active days   8 (8 changes)
V             0.724 + 0.616 + 0.524 + 0.379 + 0.322 + 0.274 + 0.233 + 0.198 = 3.270
Volume        min(1, ln(1 + 3.270) / ln(5)) = 0.902
kinds         pricing & plans 0.724 + releases 0.379 + availability & platforms 0.616
Breadth       min(1, 1.718 / 3) = 0.573
Activity      round(100 × (0.7 × 0.902 + 0.3 × 0.573))
            = round(80.32) = 80

History and snapshots

The current value is calculated from the canonical record on every page view. Once per UTC day, a snapshot of each published value is stored with its methodology version, window, confidence, coverage, and every change it used — enough to reproduce it. Snapshots are immutable; a wrong one is invalidated, never edited.

History is never reconstructed. ToolMarketCap does not back-calculate past values. Changes before ToolMarketCap began tracking were compiled later from vendors’ changelogs, and that record is denser for recent months, so a reconstructed “30 days ago” value would partly measure our research rather than the market. Index history starts with the first recorded snapshot.

Known limitations

• Only changes ToolMarketCap has logged count. A real change we have not recorded is invisible to the index.
• Every change counts the same. An acquisition and a point release are one event each; the index does not measure size or importance.
• Vendors that publish granular changelogs can show more active days than vendors that batch announcements.
• Segments have different natural cadences. Segment medians are context, not a comparison of segments’ quality.
• The record before September 2026 is less dense than after; the 30-day decay limits the influence of older days.

Validation at publication

Before publication, v1 was checked against the full dataset: the order of tools barely moves under alternative constants (rank correlation of at least 0.97 for half-lives of 21–45 days, saturation of 3–6 days, and weights from 60/40 to 80/20). It was also checked for research-depth bias, the risk that the index mostly reflects how thoroughly ToolMarketCap has researched a tool: values track recent recorded changes much more strongly than they track ToolMarketCap’s source coverage. They are not independent of coverage, since a tool we have recorded more changes for can score higher, which is why the limitations above name it and confidence describes our record. Tests cover zero activity, decay and window edges, same-day bursts, dense single-kind changelogs, single large events, bounds, determinism, and every confidence rule.

Versions

Any change to an input, window, constant, weight, or confidence rule is a new version. Values from different versions are never mixed in one series.