Release notes
H&G All-Time Index v2.0.0 is active
Current version: v2.0.0, published 22 May 2026. The rankings are fixed until the next release.
Technical changes
H&G All-Time Index v2.0.0 is a major release that makes the rating fit fully deterministic: the order-independent bout merge (#47) lets the model compute Peak-form Elo and its uncertainty bands directly from the fit instead of a pinned draw. It keeps the common-scale model, validation suite, posterior bands and Data Confidence labels from v1.2.3, and the published rankings are score- and rank-neutral against v1.2.3 - the determinism did not move the order.
Release rules
Shareable URLs stay valid across reruns.
Major reranks become v2.0, not silent overwrites.
Small follow-up releases use patch numbers such as v1.1.3. Bigger changes become v1.2 or v2.0.
Older public releases are kept for comparison when they matter.
What changed in v2.0.0
v2.0.0 makes the rating fit fully deterministic. The order-independent, quality-based bout merge (#47) replaces a pinned, non-deterministic draw, so two independent cold rebuilds now produce a byte-identical board, Peak-form Elo curve and posterior uncertainty bands.
The board is score- and rank-neutral against v1.2.3: no fighter changes position and model scores are unchanged to rounding. Peak-form Elo and its bands are recomputed deterministically from the new fit - a mean shift of about +0.5 Elo on the rating scale - while the 0-100 ranking index is unchanged but for two 0.01 display-rounding ticks.
Two trade-offs are disclosed. The #47 fit costs roughly 0.7% of stage-1 calibration, accepted explicitly in exchange for determinism. Divisions are still post-patched from the published correction rather than computed from scratch, the same as v1.2.3; root-cause division computation is deferred to a later release.
What changed in v1.2.3
The top-1000 membership and feature weights are unchanged from the reviewed feature-refactor candidate, but the six dimensions are now put on a common public-list scale before weighting. That fixes the top-25 weight-sensitivity failure found in review.
The methodology page now publishes the validation suite: weight sensitivity, leave-one-dimension-out dependency, expert prior sensitivity, external-list face validity, scale fragility and rank stability.
Peak-form Elo now shows a two-sigma posterior band on the rating scale. Data Confidence now blends that band with career depth, opposition signal and rating-curve coverage.
Women's ratings remain out of this release because they need a dedicated review.
Release notes
v2.0.0 - deterministic-fit major release. The order-independent bout merge (#47) makes the rating fit fully reproducible: the board, Peak-form Elo curve and posterior bands are recomputed deterministically and are byte-identical across cold rebuilds. Rankings and model scores stay neutral against v1.2.3. Disclosed trade-offs: about 0.7% stage-1 calibration cost accepted for determinism, and divisions remain post-patched rather than root-computed.
v1.2.3 - common-scale validation and uncertainty release. It standardises the six dimensions before weighting, adds the public validation suite, Peak-form Elo posterior bands, and Data Confidence labels reconciled with the rating bands. It does not correct historical data.
v1.2.0 - public naming and transparency release. It renamed the rank-driving number to H&G All-Time Index, moved it to a fixed 0-100 display scale, and added Data Confidence without changing ranking order or top-1000 membership.
H&G All-Time Index is an Honour & Glory analysis layer built from public boxing record archives. Read the methodology and corrections process.