fix(nodes): calculate analytics rates over available history in full and summary endpoints #485

Assessment baseline: fork master b6852ba0876b02426b6200c1c79c7e0ebe3b9389, checked on 16 October 2026. Recheck master and competing work before implementation.

Problem

Node analytics can divide availability and packets/day by the requested 30/365-day range even when the in-memory store contains much less history. That presents missing history as inactivity. The fork's full analytics and lightweight summary endpoint share an accumulator, so both must change together; fixing only the upstream-shaped full endpoint would break their parity.

Priority: high — correctness of user-visible statistics. This is an adaptation task, not approval to copy the source patch unchanged.

Fork-specific implementation

• Derive a conservative coverage boundary from configured retention, actual startup/background-load progress, load caps and memory-pressure eviction. Do not infer it from the first transmission's FirstSeen: an old transmission heard again can still be loaded recently.
• Filter the numerator and denominator to the same supported interval. Define empty-store, zero-duration and partial-hour behavior explicitly; do not produce NaN/Inf or claim complete coverage while loading.
• Share the boundary and rate calculation between full analytics and summary; preserve all existing response fields and their parity. Return explicit partial-coverage metadata and show a clear UI notice.
• Normalize comparisons to UTC and preserve inclusive boundary behavior. Any tolerance around the retention edge must be documented and tested.
• This does not add older SQL history or silently change sender/recipient/relay membership semantics.

Acceptance criteria

1. Deterministic tests cover full coverage, a longer requested range, empty/loading/capped stores, pressure eviction, re-heard old transmissions, exact boundaries and non-UTC host time zones.
2. Full and summary endpoints return matching computed statistics at the same clock instant; bins before the effective boundary are excluded.
3. Browser tests verify the partial-data notice and its absence for genuinely complete ranges; API documentation explains requested versus effective coverage.
4. A regression test fails on the current calculation and passes with the fix. Independent review checks denominator semantics and startup/eviction behavior.
5. No per-request full SQL scan; measure relevant latency and allocations. No staging/production change or automatic deployment is part of this issue.

Source reference (literal text to avoid cross-repository notifications)

Repository: Kpa-clawbot/CoreScope
Pull request: 2211
Head: 78f82a6122af4e512d04702093158a6dfce3f08e
Merge: 927a8eec8a430b8813e8afa1668d536ca62ed3a0
Changed files:
.github/workflows/deploy.yml
cmd/server/chunked_load.go
cmd/server/node_analytics_coverage_test.go
cmd/server/node_hop_analytics.go
cmd/server/store.go
cmd/server/types.go
public/node-analytics.js
public/style.css
scripts/non-unit-tests.json
test-all.sh
tests/e2e/test-issue-748-node-analytics-coverage-e2e.js
tests/unit/test-issue-748-node-analytics-coverage.js