BreakCompliance

Imported from apet97/BreakCompliance (AGENTS.md). Install upstream with npx skills add apet97/BreakCompliance. Copyright stays with the author.

Agents Guide — Break Compliance

Operational guide for AI agents. Read this and CLAUDE.md before changing code. The hard rules below are non-negotiable.

Mission

Read-only break-compliance reporter. Reviews whether Clockify users took the breaks their workspace policy requires. Never creates/edits time entries, never sends messages, never writes anything to Clockify. Fail-closed on auth, fail-loud on misparse.

Current hardening checkpoint (§41)

• CSV export parity is a current invariant: openOnly=true excludes ACKNOWLEDGED/OVERRIDDEN findings, and userIds=<id> composes after that so person-filtered exports match the visible sidebar rows.
• Current manifest evidence is schema 1.5 with structured settings and a one-character exemptUserIds TXT sentinel. Any schema 1.3 evidence is historical unless explicitly dated as current.
• /actuator/prometheus is currently public on Railway for operational evidence. Do not change auth/security behavior without an operator decision; update docs first if the posture changes.

Before changing code

1. CLAUDE.md — settings model, deploy info, hard rules.
2. CONTEXT.md — read-only mission, current domain model, ADR index.
3. docs/api-calls.md — outbound + inbound API shapes with live-probe evidence.
4. docs/clockify-marketplace/ — canonical marketplace docs mirror; cite paths in commit messages when adding new functionality.
5. docs/addon-java-sdk/ — Java SDK 1.5.3 source; the SDK already verifies — never reimplement.

Run + verify

# Full suite (JDK 21 required; system JDK 25 breaks Lombok).
JAVA_HOME=/opt/homebrew/opt/openjdk@21 PATH=/opt/homebrew/opt/openjdk@21/bin:$PATH \
  mvn -B -ntp test
# Expect 370 green. Postgres + Redis spin up via Testcontainers.
# Colima users add: DOCKER_HOST=unix:///Users/<you>/.colima/default/docker.sock

find src/main/resources/static -name '*.js' -print0 | xargs -0 -n1 node --check
NODE_OPTIONS=--no-warnings node --test src/test/js/*.mjs

# Targeted run.
mvn -B -ntp test -Dtest='LifecycleControllerTest,BreakRuleEngineTest'

# Deploy (push to main does NOT auto-deploy).
railway up --service BreakCompliance --ci

# Tail logs during smoke-test.
railway logs --service BreakCompliance

Finish + push hygiene

Before a direct main push:

1. Run git status --short --untracked-files=all.
2. Stage only intentional repo files. Local agent tooling (.claude/, docs/superpowers/) and OS junk (.DS_Store) are gitignored — keep them out. Leave stale plan drafts and other local helper artifacts out too unless the operator explicitly asks to remove or commit them.
3. If asked to remove untracked items, delete them explicitly and re-run status until it is clean.
4. Re-verify the relevant proof gates after tracked edits.
5. Prove fast-forward safety with git fetch origin main plus git merge-base --is-ancestor origin/main HEAD, then push.
6. Keep the final reply high-signal: commit SHA, pushed branch, verification results, and any live/deploy evidence intentionally skipped.

Probing live Clockify (dev workspace)

API key + workspace id in /tmp/clockify-livetest.env — never copy into the repo.

set -a; source /tmp/clockify-livetest.env; set +a

curl -s -H "X-Api-Key: $CLOCKIFY_API_KEY" \
  https://developer.clockify.me/api/v1/user | jq .

# Detailed report (matches the addon's outbound shape).
curl -s -X POST \
  -H "X-Api-Key: $CLOCKIFY_API_KEY" -H "Content-Type: application/json" \
  https://developer.clockify.me/report/v1/workspaces/$CLOCKIFY_WORKSPACE_ID/reports/detailed \
  -d '{"dateRangeStart":"2026-05-04T00:00:00","dateRangeEnd":"2026-05-17T23:59:59","detailedFilter":{"page":1,"pageSize":50}}' \
  | jq '.timeentries | length'

Probe-lab fixtures + findings at /Users/15x/Downloads/WORKING/clockify-api-probe-lab/ — refer to its findings/SUMMARY.md and ATTENDANCEANDTIMEREPORTS.md before debugging any API call shape.

Hard rules (don't break these)

Settings model (current)

Split surface: native structured-settings owns 17 admin-only fields: ten break-policy fields plus seven operational/admin controls (exemptUserIds, refreshDebounceSeconds, excludeUnsubmittedEntries, the three severity override fields, and nightShiftAttribution). The sidebar owns the preset chooser. The dropdown was removed from the manifest because Clockify's native UI renders each field independently and never re-fetches siblings on change — so backend-driven cross-field writes (the previous "preset-as-loader" pattern) weren't visible without a page reload. Persisted values still land on WorkspaceSettings columns (the custom_ prefix is historical — the customPolicyEnabled flag no longer gates evaluation; always-on). exemptUserIds is semantically blank by default, but the manifest emits a single space because Clockify's schema requires TXT defaults to have minLength: 1.

Preset selection: sidebar → POST /api/presets/apply {presetKey} → InstallationService.applyPreset(workspaceId, presetKey) overwrites all 8 threshold columns from RuleTemplatePresets.{key}.toEntity(…), sets appliedPresetKey, re-runs SettingsWarning.validate(...), and saves in one transaction. The lifecycle handler's defensive appliedPresetKey parser stays so any cached SETTINGS_UPDATED delivery still round-trips (Clockify won't push it post-manifest-removal, but the receiver is tolerant).

The engine uses synthesizeWorkspaceTemplate(input) to wrap WorkspaceSettings into a transient RuleTemplate. Per-user template resolution (RuleTemplate + TemplateAssignment tables) is dead code in evaluation; the tables remain for back-compat only.

When fallbackDetectionEnabled=true the engine adds a gap-as-break pass inside BreakRuleEngine.evaluateSegments: a wall-clock gap of [minBreakSegmentMinutes, 120] minutes between two consecutive WORK entries on the same day is credited as a synthesised qualifying break (counts toward breakMinutes, resets the continuous-work run, feeds longestQualifyingBreakMinutes, reported on findings as evidence.syntheticBreakMinutes). IGNORED (TIME_OFF/HOLIDAY) and explicit BREAK entries break the prev-work chain so no synthesis spans them. The 120-min ceiling is a hardcoded private constant (MAX_GAP_AS_BREAK_MINUTES) — gaps above that are treated as a new shift, not a break. Sidebar renders Break: 30m · 30m detected only when synthetic > 0. IGNORED entries are not credited as break minutes.

Approved time-off cache rows are evaluation-only inputs: FindingsService clips each overlapping WorkspaceTimeOff interval to the requested UTC date range and appends synthetic TIME_OFF entries without persisting them to breakcompliance_time_entries. Do not reintroduce whole-day user-date suppression for PTO; partial-day requests must leave same-day work visible.

Don'ts

• No deep-link "open settings page" from the iframe. Clockify's navigate postMessage only supports {"type":"tracker"} (see docs/clockify-marketplace/build/window-events.md). The active-template chip, the Switch… button (sidebar-side preset chooser), and the collapsible "where do I fine-tune" hint are the documented affordances.
• No new-window launch for the native settings page. Dev portal uses a catalog addon-id we don't have from JWT claims (claims.addonId is the per-workspace installation id).
• Preset selection lives in the sidebar. Don't re-add appliedPresetKey to the manifest — the field was removed because Clockify can't surface a backend-driven cross-field write without a page reload. The defensive lifecycle parser stays for legacy deliveries, but new code paths must go through POST /api/presets/apply.
• No new iframe controls for threshold fine-tuning. Individual fields stay native so admins land on Clockify's familiar settings chrome. The sidebar carve-out is the preset chooser only.
• No RuleTemplate lookups in new code paths. Engine ignores them.
• Don't drop Last-Page header parsing if you add paginated calls. We currently also approximate with entries.size() < PAGE_SIZE (documented in docs/api-calls.md).
• No outbound from INACTIVE installations. IngestionService is the single guard; new outbound paths must consult Installation.status before reading the token.
• No _WRITE scopes, ever. The mission is read-only. Adding any scope that lets the addon mutate workspace state breaks the marketplace listing commitment in docs/PRIVACY.md.
• Read-only fetches that close documented false-positive gaps are OK. The Detailed Report is the source of truth for break evaluation. Three supplementary read calls — GET /v1/workspaces/{ws}/holidays (P1.1), POST /v1/workspaces/{ws}/time-off/requests (P1.2), GET /v1/workspaces/{ws}/users (P2.3) — exist because workspaces that don't auto-create type=HOLIDAY / type=TIME_OFF time entries would otherwise produce false-positive findings, and stale userName columns make findings unreadable after a rename. Shape verified live on 2026-05-13 against the sacrificial workspace (docs/api-calls.md §1a / §1b / §1c). Don't add a fourth read endpoint without the same justification: documented false-positive class + live-probed shape + added to docs/api-calls.md.
• Don't tune Hikari by raising maximum-pool-size alone. The 3-phase ingestion split exists so the long Clockify HTTP call doesn't hold a DB connection. If the pool gets saturated, look for a missed split first.
• Don't add Redis calls in hot paths without a fail-closed contract. Today WebhookIdempotencyStore.markSeen and ClockifyRateLimiter.acquire let Redis exceptions bubble — that's intentional (Clockify retries the webhook, ingestion aborts). New Redis-backed safety checks should preserve that semantic.

When you change behavior

1. Update tests (src/test/java/me/apet97/breakcompliance/... or src/test/js/...).
2. Update CLAUDE.md if the settings model, hard rules, or build steps change.
3. Update docs/api-calls.md if any outbound or inbound API shape changes.
4. Commit message format: type(scope): short summary (matching fix(reports): …, feat(custom-policy): …, refactor(settings): …).
5. mvn test green BEFORE git push.

Numbered commit refs (archaeology)

• §1–§9 — initial takeover (contract fixes, de-minify sidebar, seed templates, settings persistence, custom policy, 401 graceful handling, CDN styling).
• §10 — ArbZG typo + preset reorder.
• §11 — webhook idempotency confirmed.
• §12 — iat replay protection.
• §13 — payload-drift logger.
• §14 — 429 Retry-After parsing + retry cap.
• §15 — verify.
• §16 — /v1/ path + ISO dates + response key (f7db0e6 reverted the camelCase mistake — live API returns timeentries lowercase).
• §17 — 9 granular custom policy fields.
• §18 — single-tab redesign, preset-as-loader, engine-from-WorkspaceSettings.
• §19/§20 — deferred (diagnostic logging, in-sidebar settings panel).
• §21 — userName captured; dropdown removed; Settings button later reverted.
• §22 — Settings button removed entirely, static caption added.
• §23 — security hardening, SDK conformity audit, test-suite verification.
• §24 — launch-readiness: SETTINGS_UPDATED canonical object wrapper accepted (SettingsUpdatedPayload), sidebar UI/UX (active-template chip + thresholds popover, "Last checked" relative timestamp, refresh button, empty-state polish, theme-flicker fix, dark-mode WCAG-AA, narrow-viewport responsive, full a11y), designed 64×64 icon, real support email.
• §25 — quality/perf/UX audit pass: EntryClassifier treats TIME_OFF/HOLIDAY as IGNORED (fixes false-positive findings on PTO/holiday days), shared RestClient bean (no more per-call RestClient.create()), V7 composite indexes on hot paths, pivot shows every day in range, visibilitychange-aware "Last checked" ticker, popover overflow fix, focus-on-Esc, prefers-reduced-motion, screen-reader labels, 4 new service tests (WorkspaceDataDeletion, RateLimiter, IdempotencyStore, RefreshSignal) + EntryClassifierTest. 226 tests green.
• §26 — UX pass after second review round: human-readable preset + timezone manifest labels (sidesteps the allowedValues: List<String> SDK limit), Title-Cased preset names, .required(true) on dropdowns to drop Clockify's auto-injected "None", .placeholder("0 = disabled") on the second-tier numeric fields. UTF-8 charset forced on /manifest (rescues § from mojibake). Cross-field validation (SettingsWarning) persisted on workspace_settings.validation_warnings (V9, additive) and surfaced in a sidebar banner via /api/session. Async ingest: bounded ingestExecutor (AsyncConfig), POST /api/ingest/detailed-report returns 202 with the run id; sidebar polls GET /api/ingest/runs/{id} with exp-backoff + Cancel link. Preset chooser relocated to the sidebar — new GET /api/presets + admin-gated POST /api/presets/apply, inline preview cards, Matches/Customized pill, confirm before overwrite. 255 tests green (+PresetControllerTest, IngestRunControllerTest, SettingsWarningTest, rewritten IngestionControllerTest with a SyncTaskExecutor override for in-test async).
• §27 — Marketplace readiness (P0 + P1 + active consumer + live validation; plan at ~/.claude/plans/verdict-do-not-zesty-gray.md). Commits 206e099..1257ffd on PR #1. Highlights:
  • P0: DetailedReportFetcher accepts timeentries AND timeEntries as a defensive fallback; WORKSPACE_READ scope dropped (unused); stale-DELETED guard in InstallationService.handleDeleted compares JWT iat to the stored installedAt (30s grace).
  • Active webhook consumer: RefreshSignalConsumer (@Scheduled fixedDelay=30s, debounce=20s) drains PENDING signals, groups by workspace, computes covering window from dateHint, dedupes against in-flight IngestionRun, dispatches via IngestionService.beginAsyncForRefresh(…, Consumer<runId>). V10 migration extends refresh_signals.status CHECK with CLAIMED / CONSUMED / FAILED / COALESCED + adds ingestion_run_id back-pointer.
  • P1 hardening: IngestionRunReaper (@Scheduled) marks runs stuck in RUNNING past 10 min as FAILED + releases their CLAIMED signals; IngestionService.prepareRun throws IngestionRunInProgressException (→ 409 with existingRunId) when a RUNNING run for the same workspace+range exists; Prometheus metrics via micrometer-registry-prometheus (/actuator/prometheus emits breakcompliance_webhook_received{event} / _refresh_signals_processed{outcome} / _ingest_run_duration / _ingest_entries_processed / _ingest_run_failed{reason}); HSTS only set when request.isSecure() (Railway-aware via X-Forwarded-Proto).
  • CI: Dependabot weekly + CodeQL Java analysis.
  • Build: Testcontainers bumped to 1.20.4 + surefire system property api.version=1.44 + TESTCONTAINERS_DOCKER_SOCKET_OVERRIDE so the suite runs identically under Docker Desktop and Colima.
  • Live validation: production install/uninstall captured in docs/LIVE_VALIDATION.md — three webhook-driven ingest cycles (entries=33→34→35), /actuator/prometheus emitting real values, 12/12 workspace tables at zero rows post-uninstall.
  • Marketplace packet: new docs/LISTING.md, docs/SUPPORT.md, CHANGELOG.md; refreshed docs/PRIVACY.md, docs/SECURITY.md, docs/DATA_RETENTION.md.
  • Test count: 279 green (+RefreshSignalConsumerTest, IngestionRunReaperTest, +1 stale-DELETED case in LifecycleControllerTest, +1 dedupe case in IngestionControllerTest, +5 iat extraction cases in ClaimsNormalizerTest, parser fallback cases in DetailedReportFetcherTest, dateHint case in RefreshSignalServiceTest; ClockifyRateLimiterTest.overBudget… de-flaked with a bucket-boundary alignment).
• §28 — P2 product polish (0.2.0; plan at ~/.claude/plans/all-you-re-picking-up-zesty-moth.md). Six commits on main:
  • Non-admin sidebar gating: Check Compliance / Refresh / Switch-preset disabled with "Workspace admin required" tooltip when state.session.workspaceRole isn't ADMIN/OWNER; no 403 round-trips. Diverged customized-pill drops its Reset affordance for non-admins.
  • Staleness indicators: new GET /api/ingest/runs/latest (204 when no COMPLETED run) seeds state.lastRunAt on sidebar load; amber "Pending refresh · webhook Xm ago" pill rendered when any PENDING/CLAIMED refresh_signals row has receivedAt newer than the latest completed run's completedAt. Repository gains findFirstByWorkspaceIdAndStatusOrderByCompletedAtDesc.
  • CSV export: new GET /api/findings/export?format=csv — RFC 4180 attachment, columns date,userId,userName,severity, code,message,workMinutes,breakMinutes,syntheticBreakMinutes, templateId,createdAt. Sidebar download via blob + a tag with download attribute.
  • Finding review UX: new admin-gated POST /api/findings/{id}/review upserts breakcompliance_finding_reviews (table + enum already shipped in V1; controller was inert). GET /api/findings now embeds review: {status, note, updatedAt} | null per row. Sidebar Checklist cycles a finding through OPEN → ACK → OVERRIDE via per-row button with optional window.prompt audit note (replaced by the §30 accessible dialog). FindingsService.exists(workspaceId, findingId) is the workspace-scoped guard.
  • Locale-aware date labels: NormalizedClaims.userTimeZone (canonical claim + legacy userTimezone / tz aliases) flows to the sidebar; pivot weekday + M/D labels rendered via Intl.DateTimeFormat (anchored at 12:00 UTC to avoid DST midnight shifts).
  • Docs: docs/WHAT_COUNTS_AS_A_BREAK.md — admin-facing explainer of WORK/BREAK/IGNORED classification, the gap-as-break heuristic, the minBreakSegmentMinutes floor, and the hardcoded MAX_GAP_AS_BREAK_MINUTES = 120 ceiling.
  • Test count: 296 green (+3 IngestRunControllerTest, +5 FindingsControllerTest review cases, +4 FindingsControllerTest CSV cases, +3 ClaimsNormalizerTest userTimeZone, +2 SessionControllerTest userTimeZone).
• §29 — Marketplace submission hardening: Maven project version aligned to 0.2.0; sidebar/session settings deep links removed in favor of the documented breadcrumb only; TIME_OFF/HOLIDAY ignored-entry semantics clarified and pinned with tests; V15 retires any pre-existing duplicate RUNNING rows and releases their CLAIMED signals before installing a partial unique index preventing duplicate RUNNING ingests under concurrent starts; detailed-report pagination / live-shape fixture tests added; operations and submission checklist docs added. 304 tests green on 2026-06-12.
• §30 — Codebase improvement pass: no-inline /sidebar CSP contract with /theme-init.js; production URL guard rejects local/non-Clockify Clockify base URLs unless dev/test opts in; group-only holidays no longer suppress the whole workspace; non-report Clockify fetchers paginate; detailed-report rows now parse into DetailedReportEntry while retaining raw JSON; IngestionService delegates time-entry upsert and suppression-cache refresh; engine input no longer carries dead template/assignment/group fields; review lookups are finding-id scoped; review notes use an accessible dialog instead of window.prompt; sidebar includes an admin audit panel; sidebar JS and CSS are split into first-party modules; CI checks all static JS modules; patch/minor deps staged (PostgreSQL 42.7.11, Lombok 1.18.46, Testcontainers 1.21.4). 339 tests expected at that point.
• §31 — Deferred dependency migration completed: Spring Boot 4.1.0 + Flyway 12.8.1. Boot 4's split modules are now explicit in pom.xml (spring-boot-flyway for migration auto-config, spring-boot-jackson2 for Jackson 2 ObjectMapper compatibility with the Clockify SDK/adapters, and webmvc/data-jpa/jdbc test modules for slice auto-config). Tests now use Boot 4 test packages and Spring Framework @MockitoBean / @MockitoSpyBean.
• §32 — Adversarial cleanup: IngestionService no longer marks a run COMPLETED before the holiday/time-off suppression refresh attempt returns, so sidebar evaluation and refresh-signal callbacks do not observe stale suppression data. Sidebar boot now loads persisted findings for the latest completed run before showing "All clear", 409 ingest_in_progress attaches to the existing run id, duplicate admin-role JS was removed, and async Spring tests mock supplemental Clockify fetchers. 341 tests green on 2026-06-12.
• §33 — Quality hardening: V16 adds a non-null scope_key identity for workspace holidays so applies_to_user_id = null can persist workspace-wide holidays; detailed-report parsing fails loud on missing entry arrays and page cap exhaustion; supplemental holiday/time-off/user-directory refreshes are independent attempts; sidebar token iat replay checks now accept Date, NumericDate numbers, and numeric strings; Redis repositories are disabled because Redis is template-backed only. 352 tests green on 2026-06-13.
• §34 — Audit remediation: approved time-off cache rows now become synthetic, non-persisted TIME_OFF entries so partial-day PTO does not hide same-day work outside the approved interval; time-off webhooks read timeOffPeriod.period.start with a legacy direct-start fallback; blank/null Detailed Report bodies fail loud; DSAR exports include actor audit logs; lifecycle deletion tests and emergency SQL cover workspace holiday and time-off cache tables. 362 tests green on 2026-06-13.
• §35 — Plan-queue refresh: detailed-report ingests reconcile the cached time_entries range before persisting the fresh page set so deleted Clockify rows disappear locally; stale comments were aligned with synthetic partial-day TIME_OFF; finding messages now route through English-only Spring MessageSource; sidebar high-visibility copy loads from same-origin /i18n/en.json; marketplace proof docs record 2026-06-14 local verification and explicit live-evidence skips. 366 tests green on 2026-06-14.
• §36 — Manifest schema repair: /manifest pins served JSON to schemaVersion: "1.5" because SDK 1.5.3 does not expose a 1.5 builder and Clockify's dev portal rejects structured settings under older schema validation. Optional TXT defaults now satisfy schema minLength; exemptUserIds stays semantically blank through the existing blank-string parser. 367 tests green on 2026-06-14.
• §37 — MessageSource runtime repair: added the root messages.properties fallback alongside messages_en.properties so Spring Boot auto-configures a concrete MessageSource in production. /api/findings/evaluate no longer 500s when the engine emits finding.* text, and the app-context regression pins all current finding message keys. 368 tests green on 2026-06-14.
• §38 — Sidebar diagnostics repair: first-load/latest-run hydration now fills findingsCreated from the persisted findings list after it loads, and the diagnostics renderer omits unknown values instead of printing null. Added the repo's first focused Node sidebar behavior test at src/test/js/sidebar-diagnostics.test.mjs.
• §39 — Triage-first sidebar redesign, built from the bundled design system (.claude/.skills/Break Compliance Design System/). New default Triage view in the results toolbar (Pivot + Checklist kept as alternate detail views): honest summary KPIs (Open fail/warn split · People affected · Reviewed n/total — deliberately no compliance %, since the backend persists findings only for problem days and there is no compliant-day denominator), a prioritized Needs attention feed of design-system FindingCards with inline Acknowledge/Override + evidence drill-down, and a risk-sorted People with findings roster whose rows filter the feed (kept in sync with the existing user-filter dropdown). Vanilla JS, no bundler, CSP script-src 'self' intact; no backend/engine/Flyway/scope change. DS tokens added additively to sidebar/css/base.css; bc-* component CSS in new sidebar/css/triage.css; pure derivations in sidebar/triage-metrics.js covered by new src/test/js/triage-metrics.test.mjs; finding.rule.* short labels are presentation-only (persisted codes/messages/severity/CSV unchanged). Documented JS gate widened to src/test/js/*.mjs. 368 Java + 12 JS tests green on 2026-06-14.
• §40 — Thermo-nuclear review follow-up: structural simplification of the sidebar after §39 left sidebar.js at 1667 lines. Each findings view is now its own module — sidebar/views/{triage,pivot,checklist}.js — plus the preset surface sidebar/views/preset-ui.js (orchestrator injects api/showBanner via configurePresetUi) and the admin gate sidebar/roles.js; sidebar.js is back to a ~900-line orchestrator. De-duplicated helpers: displayUserName (deleted the parallel bestName), pickWorstSeverityFinding + evidenceNotes (single copy in findings-rendering.js), visibleFindings (in triage-metrics.js), and one kpiCard (merged the kpiCardOf variant). Deleted the dead severityToStatus alias and the cycleReview indirection (cycle logic inlined in the checklist). Fixed the renderUserFilter hidden state mutation (filter is now reconciled once in renderResults), built the finding-card date formatters once per render instead of per card, and renamed the misleading .bc-roster-pct → .bc-roster-count. Added an import-smoke test src/test/js/sidebar-modules.test.mjs that loads the module graph (catches renamed-export breaks node --check misses). 368 Java + 15 JS tests green on 2026-06-15.
• §41 — Final-state audit remediation: CSV export filter parity restored so "All open" and person-filtered sidebar views export exactly the visible findings. GET /api/findings/export now accepts openOnly with the same review semantics as GET /api/findings (no review row or explicit OPEN = exportable; ACKNOWLEDGED/OVERRIDDEN omitted) and composes that with the existing userIds allowlist. The sidebar preserves state.lastRunRange.openOnly and passes state.userFilter to the CSV URL builder. Live evidence was refreshed to schema 1.5 / one-character exemptUserIds TXT default, and public Prometheus exposure is documented as the current Railway posture without changing auth/security behavior. Expected proof gate after this section: 370 Java tests + 17 JS tests.