Changelog

Product changes, framework versions and corrections, with dates. A tool whose argument is that every number should carry a date and a source would be an odd thing to trust if it hid its own revision history. Corrections appear here as prominently as features.

Last updated 20 August 2026.


2026

20 August — correction: a published note attributed a figure to the wrong product form

This is a correction and it goes first because the rule below says corrections go first.

One alloy, two standards was published on 19 August with a two-row table giving 450 MPa to ASTM A240 and 500 MPa to EN 10088-2 for duplex 2205, both rows labelled plate, and attributing the eleven per cent between them to the difference between the two standards.

The 500 MPa figure is not the plate row. It is the cold rolled coil and sheet row. The EN 10088-2 plate minimum for that grade is 460 MPa, on the same table of the same datasheet, one line away. Held at a constant product form the two standards are about two per cent apart, not eleven, and the rest of the gap was the product form all along.

The note now carries three rows, a different title and a correction block of its own. It was found while opening the same datasheet to confirm a product form for an unrelated grade.

/compare/316l-vs-2205 was retired when the comparison pages became generated, and two published pages went on linking to it — the note above and what PREN can and cannot tell you. The quality gate did not catch it because the retired address was still listed in the gate’s own set of expected pages, which is a good illustration of why a hand-maintained allow-list decays. The gate now resolves links against the pages that exist.

20 August — data provenance: two licences resolved, two still unreadable

Rechecked all four database licence rows on the data page.

Resolved. OMDB’s terms page is readable and names no licence at all — the site describes itself as open access and its terms cover user conduct only. RepOD is a Dataverse instance whose licensing is per deposit, so no repository-wide licence exists to record.

Still unreadable. The Materials Project terms page returns 403 to automated retrieval; OQMD’s download page is disallowed by robots. Both remain marked for a manual check rather than filled in from memory.

Newly recorded. Structures contributed from GNoME inside Materials Project are licensed BY-NC, non-commercial, and accepting that licence is required to access them. A subset of an open database carrying a non-commercial condition is exactly the clause that travels silently into a commercial product through an aggregator.

20 August — grade coverage from 4 to 27, and an audit of the result

Twenty-three grades added across eight families, including the first ferritic, martensitic, heat-resistant and precipitation hardening pages. All generated from data files against open mill datasheets, with 58 pages now built from 27 of them.

Published with it: an audit of the whole data layer — 191 sourced values, 82 named gaps, none of the 191 above E3, and 188 of them resting on one publisher’s datasheets, all dated May 2015. Also new: a grade index, a comparison index, and a note on test orientation prompted by a figure that looked wrong and turned out to be measured the other way round.

20 August — the generator was producing invalid page metadata

Every generated comparison page carried a title and description that were not valid YAML, from a quoting bug that doubled quotation marks around any value containing a colon. It affected all of them, including the three published before this batch, and would have failed the site build. Fixed, and the quality gate now parses the metadata of every page rather than only its length.

19 August — TRACE v0.1 published

First public draft of the evidence framework: five steps, six evidence levels, eight rules, three conformance levels. Specification CC BY 4.0, reference code MIT.

Published alongside it: the tool audit checklist, a record schema, a validator, and our own score against it — currently does not conform at Level 1.

Known incomplete in v0.1, and stated in the specification rather than discovered later: no worked example containing genuine E2 or E1 values; uncertainty is recorded but does not propagate; formulated materials are poorly served by the provenance fields; Level 3 may be unreachable for any tool built on live third-party databases.


How this page works

Corrections are entries. If a number on this site was wrong and we fixed it, that is an entry here with the date, not a silent edit. The same rule the framework applies to values applies to the site.

Framework versions are entries. TRACE uses MAJOR.MINOR. A minor version clarifies and adds optional fields, and records valid under the previous minor stay valid. A major version changes the ladder, a rule in §4, or the conformance levels — which is why trace_version is mandatory in every record.

Data source versions are entries. When the upstream databases we read publish a new release and we pick it up, that is a change to what the product tells you, and it belongs here rather than in a release note nobody sees. Which sources, and their current versions.


What this page does not cover

  • Every deployment. Bug fixes and infrastructure changes are not listed.
  • The roadmap. This page records what happened, not what is planned. Publishing dates we then miss is worse than publishing nothing.
  • Upstream database changelogs. Materials Project and OQMD maintain their own, and ours records only when we picked a release up.