The case
CASE-180 measured how often a declared thing is revised. The owner pointed past it, the same day: “some of the most important rates are when and how schema refinement happens. authoring documents happens faster than vocabularies expand, which happens faster than the fields that comprise a scheme are added/modified, which happens faster than new SCHEMEs are specified, which happens faster than new projects are spun up”. The last level is not visible inside one record, “although you can weakly guess at it from the rate at which codes to remote schemes are defined”.
Method. scratch/formal/schema_rates.py compares each commit’s luria.yaml with its parent’s and counts declarations at each level:
- a vocabulary term;
- a vocabulary;
- a member of a scheme, which is a field or a relation;
- a scheme;
- a remote, a project the record can cite.
It counts as documents the files authored commits add to a scheme’s directory. Each event is assigned to its unit of work: the pull request where the commit can be tied to one, otherwise the commit. For each unit it notes whether documents were filed and whether an ADR was added, and for a new term, whether a document changed in the same unit carries the term. The scaffold commit, which declares the starting grammar at once, is left out of the rates below.
Rates in nucleation, 2026-09-25 to 2026-10-27 (32 days):
| level | added | per day | days with one | first half / second half |
|---|---|---|---|---|
| document | 5030 | 158 | 32 | 2271 / 2759 |
| vocabulary term | 106 | 3.3 | 18 | 82 / 24 |
| scheme member (field or relation) | 79 | 2.5 | 8 | 67 / 12 |
| vocabulary | 8 | 0.25 | 3 | 6 / 2 |
| scheme | 5 | 0.16 | 1 | 5 / 0 |
| remote project | 1 | 0.03 | 1 | 1 / 0 |
Besides the additions, members were redeclared 26 times on 9 days, and terms were reworded 10 times and removed twice.
How the refinements came in, against a baseline of every pull request that touched the record (296):
| units that added a… | filing documents | adding an ADR | also adding a term |
|---|---|---|---|
| any pull request (baseline) | 86% | 16% | |
| term (35) | 100% | 91% | |
| member (9) | 100% | 89% | 67% |
| vocabulary (4) | 100% | 75% | 100% |
| scheme (2) | 100% | 100% | 100% |
Of the 106 terms added after the scaffold, 94% appear in the frontmatter of a document changed in the same unit. Commit subjects agree: “moral-psychology joins the topics (ADR-018)”, “ADR-042: a required kind field on CASE”.
Projects. Two weak signals:
- Remotes. This record’s remotes name three projects: the anthology and luria from the scaffold, and Mathematics of Meaning from 2026-10-11.
- First commits. The records in this session were begun on these dates: strata-g on 2026-06-24, luria (extracted from strata-g) on 2026-08-03, and nucleation on 2026-09-07. The anthology’s and Mathematics of Meaning’s start dates were not read.
That gives about five projects in four months, against five schemes added here in one day.
The other two records. Their grammars are small (ADR and DP only). Since its luria.yaml was written, luria added two terms and four members, and strata-g added nothing. They filed 33 and 10 ADRs and principles over the same weeks. Both moved their grammars onto luria.yaml whole, so its history does not show how their grammars were built.
Limits.
- One month.
- One record with a grammar of any size.
- Units are pull requests, and a large pull request is more likely to carry both a term and an ADR. The baseline is not matched for size, so “by decision” rests on the 91% against 16% and on the commit subjects, not on a matched comparison.
- Five schemes on one day and one remote are single events, so their rates are not estimates.
What it shows
- The owner’s ordering holds by frequency. By days with an event: documents 32, terms 18, members 8, schemes 1, remotes 1 here, and projects about one a month across the owner’s records. Whole vocabularies (3 days) fall between members and schemes. A term grows an existing vocabulary, while a new vocabulary arrives with the field that will hold it (44% of member units added one).
- The levels are nested, not parallel. Every unit that added a scheme also added members, a vocabulary and terms. Two thirds of the units that added a member also added a term. A rarer refinement arrives with the commoner ones, in the same unit of work as the documents that needed it. It is triggered by those documents: 94% of new terms were used at once. It is also ratified there: nine in ten of those units added an ADR, against one in six of all units. So the slow levels are not slower processes running on their own clocks. They are rarer outcomes of one loop, in which filing a document can meet a misfit, and the misfit is fixed by a decision at the lowest level that holds it.
- The gap in rates widens. From the first half of the month to the second, filing rose by a fifth, term additions fell to under a third, member additions to under a fifth, and schemes from five to none. The ratio of documents filed to terms added went from about 28 to about 115. The grammar is settling while the record grows.
What the rates are evidence of
Added at the owner’s review of the pull request, which asked that event frequencies be kept apart from dynamical time-scale separation. Three readings of these numbers differ in support:
| reading | status |
|---|---|
| The layers’ events occur at different rates. | Demonstrated for this record’s first fifty days, with the limits above. |
| The layers are separate operating regimes, worked on by different activities and under different checks. | Plausible and partly supported. Revisions of the grammar come by decision and lint enforces the declared types, but they occur inside the same work as filing (above). |
| The layers are a dynamically generated hierarchy of time scales, slow modes constraining fast ones as in CLAIM-478. | An interpretation, not shown. |
On the third reading:
- An inverse event rate is not a relaxation time. A layer that rarely changes need not be a slow mode, and nothing here measures a spectral gap or a relaxation.
- The history is short and not stationary. The rates fell over the month as the grammar settled.
- The counts mix kinds of event. Authoring, bot repair, status transitions and schema changes are counted alike.
- The rare changes arrive in shared work. This case found the rarer changes arriving in the same pull requests as ordinary document work. That fits threshold-triggered revisions within one workflow better than distinct processes running on separate clocks.
A study that could separate the readings would:
- give uncertainty for each rate;
- compare matched cohorts of entries filed at the same time and of the same kind;
- record what triggered each schema revision.
What would change it
A record begun with a small grammar and grown by use, measured over a longer period, would show whether the falling rates are settling or only the end of a founding period. So would a second large record, such as the anthology. A size-matched baseline would test “by decision”.
Note of 2026-10-27 (later): predicted before it was measured
Two of the exchange’s later replies described this case’s result in general terms, without the numbers. A16 §5, at 14:46 and before the owner asked for the measurement, gave inquiry “its own timescale hierarchy”: observation, hypothesis revision, problem reformulation and ontology revision, slower in that order. It warned that “This isn’t a universal ranking.” A17 §6 gave the mechanism this case found and added a cost: “If revising a schema is vastly more expensive than adding another record under the existing schema, inquiry may remain trapped in a locally convenient representation. … The cost of changing a distinction influences how long the distinction persists.” Here the refinements came in the same pull request as the documents that needed them, so revising a distinction cost one ADR in a unit of work already open. The measured rates fell over the month (CLAIM-475). Whether they fell because the grammar fits or because revision costs more than it did is not separated by this case.
Note of 2026-10-31: what would make this a test of radial extension
Tsoukas and Chia (THEORY-1279) hold that a category changes when it is applied to nonprototypical cases. This case is consistent with that: 94% of the terms added were used in the pull request that added them, so the grammar was extended at the case that needed the extension. As measured, though, the case cannot test it. THEORY-1279 is promoted by a study that classifies cases as prototypical or not before seeing which categories change. The record could run that study on itself. Take the documents filed under each vocabulary term, code each one blind as prototypical or not against the term’s blurb as it stood when the document was filed, and see whether the terms whose blurbs were later revised, or whose vocabularies were split, had more nonprototypical filings. No status, title or warrant changes.