Methodology · vocabulary
Glossary
These terms describe Evidaxis measurements. The latest published snapshot uses m5; historical records retain their own methodology. Each definition names the signal, window, and threshold used to interpret the open data.
Terms on every page
- System
- One open AI project that Evidaxis measures: a code repository, a model or a tool, kept under one record with a stable id. Evidaxis measures systems, never people.
- Niche
- A group of systems that do the same job, such as Coding Agents. Every comparison on the site (niche median, niche medal, Rising) is made inside one niche.
- Field
- A group of related niches, such as AI Agents. A field page lists its niches side by side; it never merges their systems into one list.
- Not yet assigned
- A system that is measured every week but has no niche yet. Until a niche is assigned it has no niche median, no medal and no niche Rising comparison; its own figures are still shown.
- Snapshot
- The weekly, dated record of every measurement, frozen on its date and released under CC0. Every figure on the site belongs to one snapshot, named by its week and date.
- Commits per week
- The average number of commits per week to the system’s GitHub repository over the trailing 12 weeks, from GitHub’s weekly commit activity.
- Citation
- A work in OpenAlex that cites the system’s paper. The count is all citing works indexed to date, all years including the current one. A system with no linked paper has no citation count.
- Verified direct dependent
- A package that depends directly on the system’s package, counted by deps.dev. Evidaxis shows the figure only for packages whose identity is verified as the system’s own; when the trend is not scored yet, the figure is a count only.
- Star
- The GitHub stargazer count. Evidaxis records it and shows it, but never scores it: size is not momentum.
- Not measured / count only
- Not measured: there is no reading for this measure, for example no paper is linked or the package is outside the panel. Count only: the count was measured, but there is not yet enough history or coverage to score its trend; the number is still shown.
- Niche median
- The middle value of one measure among the systems in a niche that have that measure. Systems without the measure are left out, not counted as zero.
- Niche medal
- A medal for a system whose citations or verified dependents reach 2×, 10× or 100× its niche median: bronze, silver or gold. The badge shows the multiple and the measure; it is a threshold, not a place.
- Rising
- At least two of the three measures (commits per week, citations, verified dependents) are rising together in the system’s niche on this snapshot. It describes the recorded series; it is not a forecast.
How a figure is computed
- Momentum
- The slope of a system’s public signals over time: how fast it is moving. Evidaxis scores that slope and nothing else. A repository with two million stars and a flat curve scores zero.
- Momentum Score
- A 0–100 readout that maps a system’s within-cohort axis z-scores onto a single number within a cohort, collapsed to one number. A convenience view only. Recognition is decided by the convergence gate, never by this number.
- Development velocity
- Axis 1: the least-squares slope of log(1 + weekly commits) over the trailing 26 weeks. A fast, leading signal of where active work is accelerating.
- Citation momentum
- Axis 2: the slope of log(1 + citations per year) to a system’s canonical paper (OpenAlex). A slow, confirming signal of scholarly acceleration. Three rules keep it honest: drop the partial current year, drop the birth year once four full years exist, and require at least three completed years.
- Direct-dependents momentum
- Axis 3 in m5: the log-slope of unique direct dependents over a system's own packages. A package counts when it is declared in the system's own tree, published under that name and not attributed by deps.dev to another repository. Packages without repository metadata are unverified. History is derived exactly from stored per-package weekly rows or restarted. At least 14 usable points of the new quantity and 5 latest dependents are required; cohort agreement and fragility can withhold a vote.
- Within-cohort robust-z
- Normalization: a raw axis slope is converted to a z-score using the median and median absolute deviation of its cohort (resistant to a single outlier), so "rising" means rising relative to peers, not in absolute terms.
- Residualization
- Axes 1 and 2 adjust their robust z against size proxies (log stars and log total citations). Axis 3 in m5 residualizes raw slope against log(1 + latest dependents) before robust z. This removes the fitted size relationship within each cohort.
- Convergence Gate
- The recognition rule in m2, m3 and m5: a non-incumbent system is Rising only when its cohort has at least five members and at least two independent axes are present and rising (positive slope and within-cohort z at least one, plus axis-specific floors and vetoes). M2 has two axes; m3 and m5 accept any two of three.
- Rising
- Status: two or more independent axes converge upward within the cohort. The full badge, and the rare one. It fires only when two separate signals agree, so a single pushed number cannot earn it.
- Watch
- Status: exactly one axis is rising; a second independent axis has not yet converged. A candidate signal, not a badge.
- Tracked
- Status: measured on two or more axes, none currently rising relative to the cohort.
- Single-axis
- Status used when fewer than two axes are available and no other status rule applies. The convergence gate needs at least two measured axes under the record's methodology version.
- Calibration
- Status: a mature incumbent included to anchor a cohort’s statistics. Measured but never badge-eligible: the badges go to systems on the rise.
- Cohort
- A peer set (a sub-niche within an industry) against which a system’s momentum is normalized. Membership controls the z-scoring, so a human draws the cohort line by hand.
- Snapshot
- A frozen, dated, hash-verified record of every measurement at a point in time, released under CC0. You cannot back-fill a baseline after the flare. The snapshot series is the part that compounds: a longitudinal map that no one can reconstruct from scratch later.
- Score receipt
- Three fields ride with every published number: methodology version, snapshot id, and manifest hash. With them, any score is independently checkable against its hash-pinned inputs in the public CC0 data.