Release notes

In v2.6.0 and v2.7.0 this column read 0 for every one of the 922,097 rows, quietly reducing any board-gender analysis to a constant. A gender-wipe regression introduced in v2.6.0 had cleared the per-row gender values that the board-statistics stage reads as its input; because an all-zero result is arithmetically valid, nothing raised an error. The inputs are restored and the denominator now matches the codebook definition: the resolved gender count (male plus female) rather than total board size, so boards carrying many unknown inferences are not artificially depressed.

907,086 rows (98.4%) now carry a non-zero value; the mean board is 31.5% female.

The composite identity key behind the 745,125 unique-editor figure is now a column, so that count is reproducible without reconstructing the key yourself. It resolves in priority order and is prefixed with the tier that produced it: ORCID iD (346,713 records), ROR-resolved institution plus name (294,399), or affiliation plus name (280,985). The ORCID and ROR tiers are the reliable ones; the lowest tier is name-based, so distinct people sharing a name and affiliation string can still collide there.

Requested by a reviewer of the accompanying paper.

Earlier releases shipped the derived bibliometrics (h-index, publications, citations, academic age) without exposing the identifier they came from, so you could neither audit a match nor join back to OpenAlex. The Author identifier is now published.

Populated on 531,437 records (57.6%), which is 98.7% of the 538,550 records that matched an OpenAlex author.

Three columns replace blind trust with a visible match strength. oa_match_method records how the OpenAlex profile was found: orcid (344,957 records), name_ror (107,173), name_only (86,420) or unmatched (383,547). name_is_initials marks records whose given name is initials only, which blocks gender inference and weakens name matching. orcid_shared_flag marks iDs that map to several distinct name strings, usually a data-entry error at the source. Flagged records are retained, not removed.

name_is_initials true on 26,763 records (2.9%); orcid_shared_flag true on 239. For bibliometric work, consider restricting to the orcid and name_ror match tiers.

2,359 IEEE rows carried raw section headings in role_std, a standardized column that should only ever hold canonical values; this inflated the distinct-value count and corrupted any group-by on role. Those rows are re-mapped, and a senior_editor category has been added for 5,005 positions previously absorbed into other.

role_std now holds exactly ten canonical values.

All aggregates were recomputed on the corrected data. The largest movement is multi-board editorship, which drops from 25.3% to 15.0% of editors serving two or more boards. The earlier figure came from exact name matching, which merged distinct homonymous individuals into single editors and so overstated multi-board service by roughly ten percentage points. The new figure uses editor_key, which separates them wherever an ORCID iD or a resolved institution distinguishes the people involved.

Treat 15.0% as a corrected estimate: the residual name-tier collisions push in the same direction, just far less.