# Churn — data, dimensions, constraints, and inheritance Churn organizes five complementary ways to model a magnitude. Each numbered set has a distinct primary dimension, but all sets share provenance and quality rules, the three-domain grouping (math, physics, social), and a constraint layer. See [DATA-CATALOG.md](DATA-CATALOG.md) for the cross-dimensional data catalog, shared record contract, import policy, and integration roadmap. ## Set index | Set | Dimension | Core model | Question | |---|---|---|---| | [1 — Longs](1/README.md) | temporal series | `Long` and pairwise relevance | How do measurements co-vary across periods? | | [2 — Structures](2/README.md) | standing scalar | `Structure` | What is the value, unit, and evidence kind? | | [3 — Facets](3/README.md) | object + selected face | `Facet` | Which property of which object is described? | | [4 — Areas](4/README.md) | two-dimensional extent | `Area` | What extent is occupied, and relative to what? | | [5 — Lines](5/README.md) | one-dimensional count/length | `Line` | How far does a count or length reach, and what Spark guides its reading? | ## Organization Each set keeps its model source and local explanation beside the data it interprets. Sets /2–/5 group inputs under `data/` and domain explanations under `models/`; /1 keeps its annual series in `longs.csv` and its separate assumptions layer under `assumptions/`. - `N/README.md`: purpose, sources, run instructions, and file index. - `N/*.sleela`: SLeeLa model for the set's dimension. - `N/Constraints.java` and `N/constraints.sleela`: JVM and SLeeLa constraint representations. Java report classes share a fully-qualified class name, so each set is compiled into a separate `build/N/` directory. - `N/data/`, `N/models/`: input facts and domain-level explanations. - `N/BASIS.md`: basis document for sets that define one. - `5/SPARKS.md`: interpretive guidance for Lines; Sparks are not evidence. ## Shared data contract and relationships These are conceptual/data-contract relationships, not claims that SLeeLa currently implements class inheritance between the five model structs. 1. **Shared evidence contract.** Factual records should identify their name, value(s), unit or measurement basis, evidence kind (`EXACT`, `MEASURED`, or `OBSERVED` where applicable), source/provenance, and date or period when values can change. 2. **Domain is cross-cutting.** Math, physics, and social classify the subject; they do not replace Long/Structure/Facet/Area/Line. 3. **Derived relationships must be explicit.** A Facet should identify its source object(s); an Area derived from geometry should identify its formula and inputs; a Line derived from a count or length should retain its unit and source. Do not conflate formulas, measurements, and observations. 4. **Temporal joins need matching periods.** Longs should document period, frequency, units, and fiscal/calendar-year differences before series are compared. Correlation is descriptive, not causal. 5. **No speculative imports.** Add data only with a verifiable source, retrieval/update date, units, evidence kind, and an explanation of how it enriches the model. Prefer updating a canonical dataset over duplicates. ## Build and validation From the repository root: ```sh make churn make churn-data-check make churn-test ``` Or from this directory: `make all`, `make test`, `make clean`. The data check parses JSON/XML/CSV/TXT, checks CSV headers and row widths, detects duplicate IDs in supported record arrays/CSV ID columns, validates evidence-kind values, and enforces the opt-in `churn-1` JSON contract. It does not establish scientific correctness, unit compatibility, referential integrity, or that sources are still current. Java constraint reports are generated outputs and remain ignored by Git. See [DATA-CATALOG.md](DATA-CATALOG.md) for the recommended validation roadmap. ## Evidence discipline - `EXACT` means definitional/formula-based, not merely a rounded decimal. - `MEASURED` values should retain source uncertainty where available. - `OBSERVED` values need a stated period or retrieval context. - Record units explicitly; convert before comparing unlike units. - Keep assumptions, derived values, and published observations distinguishable. - A Spark is explanatory guidance, never a factual source. --- # SLeeLa

SLeeLa — The Now. The Ingredient.

SLeeLa Marketing Logo