Long-record notebook findings: current-main triage

This triage compares the post-RC showcase notebook evidence with current main at 46c9ab81152ecf13f64310a211f9db3f65b3ec3f. An older notebook kernel reporting 0.1.0rc1 is not by itself evidence about current source behavior.

Finding	Classification	Current-main evidence and decision
Main install still prints 0.1.0rc1	NOTEBOOK / ENVIRONMENT ISSUE and DOCUMENTATION / DISCOVERABILITY ISSUE	Main intentionally retains the release-candidate semantic version. A running kernel can retain an older import, and --ignore-installed can churn unrelated dependencies. Add runtime/source identity and safe restart guidance; do not bump the release.
Hundreds of detected events look like regional episodes	REAL CURRENT DEFECT	detect_events creates one row per contiguous run at one grid cell, but EventResult does not state that scope. Add first-class scope and row-meaning metadata.
Local events need regional consolidation	GRAMMAR GAP / NEW FEATURE	No existing verb performs explicit space-time connected-component consolidation. Add a bounded, criteria-controlled semantic verb.
Annual event summaries require pandas	GRAMMAR GAP / NEW FEATURE	No event-aware metrics verb exists. Add a narrow period/metric operation returning ordinary xarray data.
Event-relative life-history trajectories	GRAMMAR GAP / NEW FEATURE	Existing synchrony compares event anchors but does not construct event-relative trajectories. Document the required design and defer numerical extraction until support/alignment and overlapping-event policies are explicit.
Annual/period signatures	GRAMMAR GAP / NEW FEATURE	Event metrics can already produce one explicit row per period. Defer a separate signature verb until multiple metric objects and unit-safe feature assembly have a stable contract.
block_signature(...).explain() loses degC	REAL CURRENT DEFECT	The reduced variable preserves units, but semantic inference reads only Dataset-level attrs. Infer a unit only when Dataset variables agree.
compare_blocks gives every metric source units	REAL CURRENT DEFECT	Current xarray propagation leaves pearson_r and n carrying source units. Assign metric-level units and retain per-source-variable units explicitly.
Large EventResult display overwhelms notebooks	REAL CURRENT DEFECT	The dataclass default repr prints the entire catalog. Replace it with a bounded scientific summary and small preview.
Long records may scale poorly	REAL CURRENT DEFECT, EXPECTED SCIENTIFIC BEHAVIOR, and DOCUMENTATION / DISCOVERABILITY ISSUE	State creation and overlap are lazy; event catalog construction necessarily materializes the state cube. Seasonal filtering also exposed that index-adjacent observations across a large coordinate gap could be joined as one event; event contiguity must respect coordinate cadence. Add a 20-year small-domain smoke test, document materialization, and use an active-window consolidation algorithm rather than unconditional all-pairs matching.
Multi-year quantile_state(.90) reference population is unclear	DOCUMENTATION / DISCOVERABILITY ISSUE	Current behavior correctly pools the selected time dimension. Record that reference population in attrs and explanations; do not add unimplemented climatology modes.
sync_with positive lag is ambiguous	ALREADY FIXED ON MAIN with a discoverability follow-up	Current code and an asymmetric regression define +5D as comparing left at t with right at t+5D; the right-hand condition occurs later. Strengthen output attrs, docs, and negative-lag coverage.
Rolling median-split output is not self-describing	REAL CURRENT DEFECT	Core variable labels and unitless attrs exist, but output-coordinate, edge, split, and reference descriptions are incomplete. Complete them without changing values.
Negative rolling tail values look invalid	EXPECTED SCIENTIFIC BEHAVIOR and DOCUMENTATION / DISCOVERABILITY ISSUE	The public verb is a signed upper-tail variance minus full-window variance contrast, not a probability. Negative values are valid. Preserve numerical behavior, declare squared units and an unbounded real range, and distinguish it from median-split Spearman synchrony.
Trend verb	GRAMMAR GAP / NEW FEATURE	Defer pending a stable independent-variable/unit and uncertainty contract.
Change-point verb	GRAMMAR GAP / NEW FEATURE	Defer; method choice, uncertainty, and multiple candidates need a design decision before a public verb.
Event classification	GRAMMAR GAP / NEW FEATURE	Defer opaque clustering. First stabilize scope, episode construction, metrics, and trajectory features.
Event-specific default plots	DOCUMENTATION / DISCOVERABILITY ISSUE and potential GRAMMAR GAP / NEW FEATURE	Existing plotting remains valid. Defer specialized defaults until the new event objects and metrics have settled; documentation will show concise ordinary plots.

The implementation must preserve exact spatial alignment, explicit temporal support policy, ordinary xarray interoperability, and existing numerical behavior unless a row above identifies a current defect.