Methodology / Revision history

Every version, and what changed.

The full revision history of the CatalogIQ methodology. The version that produced a score travels with the score, so any number is always reproducible against the rules that were active when it was measured.

MAJOR
A structural change to the Index: a dimension added, removed, or redefined, or a change to the sub-index split.
MINOR
A change to the weighting within an existing structure.
PATCH
A bug fix or a non-substantive clarification.
v8.02026 · 04Current
Consolidates the v7 framework and Working Draft 2 into one document, describing the engine as it is actually implemented in June 2026. Material in five places.
What changed
+Standards-grounding is now multi-standard. Sub-index A dimensions resolve against a priority-ordered list, ETIM, UNSPSC, eCl@ss, Google, Amazon, with fallback events recorded. The single-anchor model is gone.
+Channel definitions are data, not code. Feed-protocol scoring is driven by channel definitions in a table; a channel without a definition does not appear in the output.
+Per-field provenance at the regulatory layer. Every attribute in a regulatory finding carries a source label and a confidence value.
+Catalog-level compliance is a published output. Per-product gap reports aggregate into a catalog-level summary: match rate, attribute coverage, biggest gaps, by-classification breakdown.
+The Regulatory Overlay exists. A rule-based compliance system ships alongside the Index as a sibling, producing per-product findings rather than a score.
Also closed in this release
Three measurement classes Three-state missing-data treatment Pillar view documented UNSPSC added to reference standards Per-field channel readiness
v72025 · Q3Major
The framework specification that introduced vertical tuning and tightened Sub-index B.
What changed
−Sub-index B reduced to four dimensions. Data Freshness was deferred until the feed pipeline is built, under Principle Zero: no placeholder dimension carries weight into a published score.
+The HVAC-tuned weight vector introduced. The first vertical-specific methodology, weighting foundational attribute precision above the global default.
v62025 · Q1
The original five-dimension Sub-index B specification, the first published structure of the Index.
Governance

How versions are governed.

Calibration cadence

Weights are reviewed at least annually against the scored corpus and re-released as a new minor version when a review produces a material change. Until the first calibration lands, every weight table carries a dated “review pending” note.

Standards & rule-set governance

Reference standards are tracked at their published versions; a score produced under an older version stays reproducible against that version. Regulatory rules carry their own versioning, frozen per observation.

The public change log

Every change to the methodology, the weights, the standards versions, or the rule set is recorded here, with what changed, when, and why. A methodology you cannot track over time is one you cannot trust.

Reproducibility

Given the same inputs and the same methodology version, the engine produces the same score. The version that produced a score travels with the score.

Shape the next version

Think a weight is wrong? Make the case.

Tell us which weight, standard, or dimension you would change, and the argument for it. Substantive challenges are reviewed in the calibration cycle and credited in this change log.

Submit a methodology challenge → Back to the methodology
Routes to the methodology-comment intake. Wire to the HubSpot form URL before launch.