Evidaxis

As of 2026-11-07: TRL (Post-training & Alignment) is under weekly measurement. Axis coverage: 2 of 3 axes measurable. Gate: not computable — reference measurement

TRL

repo reference measurement
Post-training & Alignment · n = 16 · huggingface/trl
2 of 3 axes measurable · conf medium
convergence gate
Convergence gate: Gate: not computable — reference measurement Development velocity: measured × Citation momentum: not measurable ∅ Direct-dependents momentum (deps.dev): measured ×
gate
0 of 3 axes

Gate: not computable — reference measurement

Gate: not computable — reference measurement
position in cohort distribution one of 16 in cohort
percentile 53 within cohort p25 p50 p75 p0 p100
percentile
p53
momentum averages present-axis z-scores; two- and three-axis scores are not strictly comparable
velocity z
velocity z gaugevelocity z+0.31 within cohort, log-slope +0.031/wk.0+0.31
velocity z
+0.31 σ
slope
+0.031/wk
log-slope +0.031/wk
citation z Citation source: not measured.
direct-dependents z
direct-dependents z gaugedirect-dependents z+0.00 within cohort, log-slope +0.009/point.0+0.00
direct-dependents z
+0.00 σ
slope
+0.009/point
Direct-dependents axis: measured.
cohort
16
systems measured
weekly commits
57.8
Development velocity axis: measured.
total citations
n/a
Citation source: not measured.
momentum
51.9
percentile p53
commit heatmap, last 26 weeks57.8/wk
weekly commits, 26 weeks26 weeks of commits, peak 84 in a week, 26 active weeks, 1261 total.week 1: 21 commitsweek 2: 23 commitsweek 3: 26 commitsweek 4: 36 commitsweek 5: 51 commitsweek 6: 60 commitsweek 7: 26 commitsweek 8: 74 commitsweek 9: 44 commitsweek 10: 47 commitsweek 11: 77 commitsweek 12: 32 commitsweek 13: 21 commitsweek 14: 29 commitsweek 15: 48 commitsweek 16: 48 commitsweek 17: 37 commitsweek 18: 66 commitsweek 19: 51 commitsweek 20: 47 commitsweek 21: 55 commitsweek 22: 78 commitsweek 23: 74 commitsweek 24: 45 commitsweek 25: 61 commitsweek 26: 84 commits
weeks
26
peak/wk
84
active
26
total
1261
citations by yearCitation source: not measured.
Daily dependents count (unscored) 128 dependents 118 direct, 10 indirect · pypi/trl · deps.dev · the scored m5 axis uses weekly partitions

Derived signals

second-order reads, computed from the published series

Three signals ship when the commit series is statistically sufficient; the rest are held in a reserved state until the data supports them. A reserved signal shows its reason, never a fabricated number. How these are computed.

recency-weighted swing trailing 13-week vs prior 39-week log-slope ratio +2.27, regime accelerating. prior slope +0.023, recent slope +0.053 per week. +0.023prior 39w+0.053recent 13w+2.27×
swing
+2.27×
regime
accelerating
windows
13w vs 39w
snapshot 1 of 1 · · m5
commit-axis Sharpe log-linear slope over residual std, value +0.042 with 95% CI +0.020 to +0.063, 50 degrees of freedom. CI excludes zero.0+0.04295% CI +0.020, +0.063+0.020+0.063
sharpe
+0.042
95% ci
+0.020, +0.063
dof
50
snapshot 1 of 1 · · m5
commit changepoint52 weekly commit points. Changepoint at week 50 (2026-10-24), slope change +18.761 commits per week.2026-10-24
date
week
50 of 52
Δslope
+18.761/wk
snapshot 1 of 1 · · m5
code-leads-citation lag reserved

reserved: citation series too short / axis absent for cross-correlation

status
reserved
value
insufficient data
snapshot 1 of 1 · · m5
axis alpha vs cohort reserved

reserved: cohort N too small for regression (power ~0)

status
reserved
value
insufficient data
snapshot 1 of 1 · · m5
gate ETA reserved

reserved: needs snapshot history (1 snapshot exists)

status
reserved
value
insufficient history
snapshot 1 of 1 · · m5
methodology gate snapshot 1 of 1 · · m5
commit weeks
52
citation years
0
shipped
3
reserved
3

shipped

  • recency-swing
  • commit-sharpe
  • commit-changepoint

reserved

  • code-leads-citation-lag
  • axis-alpha
  • gate-eta
snapshot 1 of 1 · · m5

Reconstructed history

TRL reconstructed weekly commits50 points, range 12 to 77 commits/wk.
min
12 commits/wk
max
77 commits/wk
latest
51 commits/wk
points
50

Weekly commits over the past year, reconstructed from git history (GitHub commit_activity). This is reconstructed, not a point-in-time capture. Evidaxis captures the Type-2 signals (watchers, issues, dependents) live and never backfills them; those are the moat.

Daily dependents count (unscored)

TRL dependents over time19 points, range 128 to 140 dependents.
min
128 dependents
max
140 dependents
latest
128 dependents
points
19

Downstream packages depending on pypi/trl (deps.dev), captured point-in-time each snapshot. This daily series is unscored. The scored axis in m5 uses weekly partitions.

Measurement states

TRL is under weekly measurement. History: 19 weekly observations; minimum 4. Development velocity axis: measured. Citation source: not measured. Direct-dependents axis: measured. Gate: not computable — reference measurement

Sources

huggingface/trl development-velocity axis

Others in Post-training & Alignment

AReaL Post-training & Alignment
1 of 3 axes measurable
59
p73
vel +0.68 cit no axis deps no axis
axis dials
+0.68 n/a n/a
ART Post-training & Alignment
1 of 3 axes measurable
55
p60
vel +0.37 cit no axis deps no axis
axis dials
+0.37 n/a n/a
Axolotl Post-training & Alignment
1 of 3 axes measurable
57
p67
vel +0.53 cit no axis deps no axis
axis dials
+0.53 n/a n/a
LLaMA-Factory Post-training & Alignment
1 of 3 axes measurable
16
p0
vel -2.75 cit no axis deps no axis
axis dials
-2.75 n/a n/a
ms-swift Post-training & Alignment
2 of 3 axes measurable
40
p27
vel -0.88 cit no axis deps -0.67
axis dials
-0.88 n/a -0.67
NeMo-RL Post-training & Alignment
1 of 3 axes measurable
61
p80
vel +0.84 cit no axis deps no axis
axis dials
+0.84 n/a n/a

Full Post-training & Alignment cohort →

Direct-dependents momentum (deps.dev)

Direct-dependents axis: measured.

Coverage state
scored
Cohort z
+0.00
OLS slope
+0.009
Theil-Sen slope
+0.008
Latest direct dependents
158
Confirmed-clean points
29
Reconstructed history points
15
As-of partition
2026-10-19
Fragility veto
inactive
Rising vote
no

This axis measures the trajectory of a system's own packages: declared in its own tree, published under that name, and not attributed by deps.dev to another repository. A package with no repository link is unverified and stays. The self-name exclusion list is fixed at the v2h.1 panel. Systems with no admitted package are not measured on this axis. Points that cannot be derived exactly are withheld, and the series restarts after a withheld point.

The earliest weekly points of each series were read from deps.dev's BigQuery history on 2026-07-21; later points are weekly captures. m5 does not replay those partitions: each point is the stored system total, one package's unique_direct, or a complete sketch union, and a point that is not exact is withheld. Each record reports how many of its points are reconstructed history.

Source: deps.dev (Google), CC-BY 4.0. Modified: aggregated to system-level counts and momentum scores.

Axis definition · Snapshot axis correlations

Cite this measurement

Plain citation

Evidaxis. TRL measurement record. Snapshot 2026-11-07. https://evidaxis.org/e/e_B1Q69HZTTMA/ Concept DOI: 10.5281/zenodo.21076011.

BibTeX

@dataset{evidaxis_e_B1Q69HZTTMA_20261107,
  title = {TRL measurement record},
  publisher = {Evidaxis},
  year = {2026},
  url = {https://evidaxis.org/e/e_B1Q69HZTTMA/},
  doi = {10.5281/zenodo.21076011}
}

Download record (JSON)

entity_id e_B1Q69HZTTMA methodology m5 snapshot 868e3c39d9c7 period 2026-w45 record JSON ↓ source snapshot

Score derived from public data on Evidaxis methodology m5, never edited by hand. Cite as: “Evidaxis, TRL, entity e_B1Q69HZTTMA, snapshot 2026-w45.”

Durable reference (format-independent): urn:evidaxis:claim:e_B1Q69HZTTMA:m5:2026-11-07