Build · 3 Jul 2026
Derivation log schema settled, versioned and open
The record format fixed before anyone asked for it, under a permissive licence, with a changelog.
Specifics
The numbers, and where they come from
| Quantity | Value | Basis |
|---|---|---|
| Schema version | v1.0.0 | Stated |
| Fields | 14 | Counted |
| Format | JSON Schema draft 2020-12 | Stated |
| Licence | CC BY 4.0 | Stated |
The record
Field by field
- 01
Turn and session identifiers
Stable across a conversation, so a regulator can reconstruct an exchange rather than a single answer.
- 02
Input flags
Every typed flag raised on the way in, with the stage that raised it. A flagged turn is not a failed turn, but it is a turn where the repair path ran.
- 03
Selected intent and selecting branch
Not just what was chosen, but which branch chose it. The second half is what makes the record re-derivable.
- 04
Artefact version hash
The build identity. Determinism holds within a build, so a record without this field cannot be re-derived and is not a record.
- 05
Round trip result
Whether comprehension recovered the intent that production started from, per response.
- 06
bridges_licensed
Which qualia relation licensed each bridging inference, with referent and relation named. No statistical architecture can populate this field.
Why publish it now
It needs no code and no customer
A schema is one of the two credentials a pre-product company can ship honestly. It costs nothing to publish and it is standards-path currency: a record format other vendors could adopt is a contribution, whereas a record format only we use is a file format.
It is published at a stable versioned URL with a changelog, under a permissive licence, with an explicit invitation to implement it. If a competitor adopts it we have gained more than we have lost.
What a record does not prove
The log shows that production and comprehension agreed. It does not show that a human recovered the intent. Verification here is sound with respect to the grammar, not with respect to English.