[PredictabilityAtScale]

Unknown or irregular periods currently disable statistical analysis, while Guard needs useful evidence without inventing consumption. This adds schema-1.1 per-source missing/coverage policies, bounded calendar grids, explicitly authorized empty additive zeros, comparable event-index observations, availability/revision replay, gap-aware frozen phase references and rule/reset handling. Schema 1.0 and its generated contracts remain unchanged.

Point support and pattern continuity are separate. Unknown/incomplete/excluded slots do not train or calibrate. First-value range rules remain available without a learned baseline; changes respect gaps/resets/time separation. Adaptive/explicit-trend spans need explicit post-gap reset/restart so a suspected episode is not silently trained as normal. Relationships preserve undefined denominators, measured exclusions and original source lineage.

Schema-1.1 multi-resolution keeps native and completed UTC-day views, ratio-of-sums, separately declared daily rules and case/policy replay. Unsupported daily lag/normalization/joint definitions are explicit. Complete measured daily sums remain available to rules even when excluded from modeling.

The existing README progression gains chapters 14–22, eleven checked-in fixtures and nine rendered figures, using the existing generator/demo/renderer and style. The normal Pages pipeline publishes the numbered progression and sparse API guide. Fixed the older generator's omitted single-reference setting for shift/steps without changing their existing fixtures or detections.

Validation: 212 tests pass locally, including 46 focused sparse cases; legacy and new demos pass; all progression figures regenerated and new figures visually reviewed; 1.0 schemas unchanged; actionlint and capability-map preview pass. Clean wheel/sdist API/CLI/MCP installs pass locally. Hosted CI verifies Python 3.11–3.13 installs plus the offline Linux/Python 3.12 Lambda bundle and bounded 1,344-hour/event benchmarks with runtime/result-size reports.

Version 0.2.0 is unused. After green CI, this authorized task will merge, enable immutable releases, tag the merged source, stage the tested artifacts and publish through the existing verified owner helper. Publication additionally verifies the actual public wheel URL/hash and downloaded offline bundle. No PyPI, AWS infrastructure or UsageTap changes are included.

Known limits are documented: unique event timestamps; caller-owned comparability/coverage/history origins; fixed UTC days and frozen multi-resolution children; no simultaneous multi-seasonality, episode quarantine, durable append/state or automatic historical result diffing. Full detailed results can be several MB and need application-specific persistence/budget choices.