v0.9.0-beta.1 - Internal Beta

v0.9.0-beta.1 - Internal Beta

This release finalizes a major Task Runs refactor and fixes key reliability/UX issues around after-action reports and tab refresh stability.

Highlights

• Refactored Task Runs UI into focused components:
  • filters, list/detail workspace, KPI summary, KPI compare, reliability review, reliability trends
• Extracted Task Runs orchestration logic into dedicated modules:
  • URL state, options derivation, filter transitions, action handlers, and data loading
• Fixed AAR behavior:
  • opening a task run no longer auto-generates/persists AARs
  • stored AAR retrieval now returns previously persisted reports (including stale ones) instead of hiding them
  • removed noisy 404s for missing stored reports
• Reduced Task Runs tab flicker/jitter:
  • removed duplicate refresh triggers
  • switched to deterministic refresh-key diffing
  • stabilized panel rendering during updates with stale-while-refresh behavior and layout guards

Why this matters

• Better maintainability: Task Runs is no longer a monolithic UI/logic block.
• Better correctness: AAR generation semantics now match user intent.
• Better UX: fewer visual glitches and more stable refresh behavior.

Validation

• Frontend build passes
• Task Runs UI smoke passes
• Task Runs URL-state smoke passes
• Backend smoke passes

v0.8.0-beta.1 - Internal Beta

v0.8.0-beta.1 — Internal Beta

This release operationalizes Weekly Review workflows and adds auditable review acknowledgements.

Highlights

• Weekly Review UX in Task Runs
  • Expanded Reliability Review panel with actionable sections:
    • KPI snapshot
    • anomaly flags
    • latest trend deltas
    • latest failure-mode mix
    • labeling backlog
    • top failing runs
  • Added export actions:
    • Copy markdown
    • Download markdown
  • Added optional Last 4 weeks compact view (week-over-week snapshots)
• Review acknowledgement audit trail
  • New persistent acknowledgements model:
    • review_acknowledgements table
  • New API:
    • POST /api/reliability/review/acknowledge
    • GET /api/reliability/review/acknowledgements
  • Reliability review payload now includes acknowledgement status/metadata for the selected period + runner.
  • UI supports:
    • reviewer + optional notes
    • Mark reviewed
    • reviewed badge with reviewer/time
• Review period consistency
  • Reliability review now returns explicit period bounds (from/to) for default windows, improving downstream UX and acknowledgements matching.

Quality and validation

• Added smoke assertions for acknowledgement lifecycle:
  • unacknowledged default
  • acknowledge success
  • acknowledgement listing
  • negative path: missing reviewer rejected
• Verified:
  • backend syntax checks
  • backend smoke suite
  • frontend build
  • existing UI regression smokes

Why this release matters

Suboculo now supports a lightweight, repeatable weekly reliability ritual:
review metrics → export/share summary → mark reviewed with auditable metadata,
without turning the system into runtime policy enforcement.

v0.7.0-beta.1 - Internal Beta

v0.7.0-beta.1 — Internal Beta

This release completes the KPI Compare fast-follow slice and hardens shareable analysis workflows in Task Runs.

Highlights

• KPI Compare usability upgrade
  • Preset windows (7/14/30) and custom A/B date ranges
  • Explicit Period A / Period B range labels with run counts
  • Delta interpretation with metric-aware direction (Improving / Degrading / No change)
• Guardrail-aware compare
  • Period-level sample sufficiency badges for:
    • canonical outcome sample
    • known-cost-success sample
  • Direction cells now show Insufficient sample when interpretation is not statistically trustworthy
• Shareable compare links
  • New Copy share link action in KPI Compare
  • URL now pins all compare-affecting state:
    • Task Runs filters
    • compare mode/preset/custom ranges
    • source context (derived_attempt)
• Deep-link reliability fixes
  • Shared links now open directly on Task Runs (tab=task-runs)
  • Fixed tab key mismatch (taskruns vs task-runs)
  • Added legacy compatibility for old shared links
  • Fixed startup URL hydration/sync race that previously redirected to Events

Test and quality updates

• Added frontend regression smoke for URL-tab/share-link behavior:
  • smoke:ui-taskruns-url
• Existing Task Runs regression smoke retained:
  • smoke:ui-taskruns
• Frontend build passes.

Documentation

• README updated with KPI Compare fast-follow capabilities
• Added release notes detail in:
  • docs/release-notes-kpi-compare-fast-follow.md

v0.6.0-beta.1 — Internal Beta

v0.6.0-beta.1 — Internal Beta

This release hardens reliability metrics correctness and makes After-Action Reports (AARs) durable.

Highlights

• Corrected KPI cost semantics:

  • Suboculo now separates unknown cost from known zero cost.
  • Added estimated_cost_known in task-run derivation.
  • cost_per_success now uses only successful runs with known cost.
  • Cost aggregates and guardrails are aligned to known-cost sample size.

• Durable After-Action Reports:

  • AARs are now persisted in backend storage (task_run_reports), not just transient UI state.
  • AAR endpoint uses read-through caching:
    • returns stored report when fresh
    • regenerates and stores when missing/stale
  • Outcome updates invalidate stale reports for affected runs.

• Task Runs UX reliability:

  • Task Run detail now auto-loads persisted AAR on open.
  • Debounced Task Runs reload behavior remains in place for stable filtering.

API/Behavior Notes

• Reliability KPI outputs now include known-cost counters used for cost metrics.
• /api/reliability/kpi-definitions reflects the updated cost formula and null conditions.
• /api/task-runs/:id/after-action-report now returns cache metadata indicating generated vs stored response path.

Validation

• Backend smoke suite updated and passing.
• Frontend build and Task Runs regression smoke checks passing.

Why this release matters

This release fixes a subtle but important KPI integrity issue: unknown-cost runs could previously bias cost metrics downward. It also turns AARs into persistent artifacts that survive browser reloads and remain consistent across sessions.

v0.5.0-beta.1 - Internal Beta

v0.5.0-beta.1 - Internal Beta

This release turns Suboculo’s reliability data into an operational review workflow for teams.

Highlights

• Added configurable KPI thresholds (local-first):
  • threshold targets loaded from .suboculo/thresholds.json (or SUBOCULO_THRESHOLDS_PATH)
  • anomaly flags now include target breaches (below_target_, above_target_)
• Added Reliability Review API:
  • GET /api/reliability/review
  • returns period summary with KPI snapshot, anomalies, trend deltas, failure-mode summary, labeling backlog, top failing runs
  • includes pre-rendered markdown for sharing/review
• Added MCP reliability review tool:
  • suboculo_get_reliability_review
  • enables CLI-native weekly/period review workflows
• Added Task Runs “Reliability Review” card in frontend:
  • executive summary above KPI detail cards
  • anomaly badges
  • clickable labeling backlog (opens needs-labeling queue)
  • top failing runs with click-to-open detail
  • markdown copy action

Install + docs improvements

• Installers now copy .suboculo/thresholds.example.json into target projects.
• README/backend docs updated for thresholds configuration, review endpoint, and MCP workflow usage.

Validation

• Backend smoke suite passes.
• Frontend production build passes.

This beta is focused on making reliability monitoring actionable in day-to-day team operations.

v0.4.0-beta.1 - Internal Beta

v0.4.0-beta.1 - Internal Beta

This release adds post-run reliability analysis as a first-class workflow in both UI and MCP/CLI.

Highlights

• Added After-Action Reports (AAR) for task runs:
  • backend endpoint: GET /api/task-runs/:id/after-action-report
  • deterministic report sections: what happened, variance vs expected, risks, remediation
  • explicit insufficient_evidence state when canonical outcome is missing
  • markdown export payload for sharing
• Added Task Run AAR UI:
  • generate report on demand
  • copy markdown directly from Task Run detail view
  • report state resets after outcome changes to prevent stale summaries
• Expanded MCP reliability coverage with new tools:
  • suboculo_get_reliability_kpis
  • suboculo_get_reliability_trends
  • suboculo_get_failure_mode_trends
  • suboculo_get_task_run_after_action_report
  • suboculo_record_task_run_outcome
• Completed CLI workflow docs for reliability operations:
  • added prompt examples for Claude Code and OpenCode usage
  • added repeatable MCP Reliability Workflows (triage, comparison, postmortem)

Why this matters

Suboculo now supports a full loop:

1. derive task runs,
2. label outcomes,
3. track KPI/trend signals,
4. generate actionable after-action reports,
5. run the same analysis from CLI via MCP.

Validation

• node --check backend/server.js
• node --check backend/mcp-analytics-server.mjs
• npm --prefix backend run smoke
• npm --prefix svelte-app run build

All checks pass for this release scope.

v0.3.0-beta.1 - Internal Beta

v0.3.0-beta.1 - Internal Beta

This release advances Suboculo from event capture into practical reliability analysis for agent work.

Highlights

• Added first-class reliability workflow around task runs and outcomes.
• Expanded KPI coverage with:
  • aggregate reliability metrics
  • runner-level KPI comparison
  • time-bucketed trends
  • trend insights with sample-confidence guardrails
  • failure-mode trends over time
• Completed provenance enrichment in task-run derivation:
  • git revision
  • normalized model attribution
  • runner/runtime version
• Added bulk outcome operations for faster labeling:
  • needs-labeling queue
  • multi-select apply in Task Runs UI
  • batch outcome API

Reliability and UX improvements

• Improved event normalization and ordering behavior.
• Strengthened OpenCode session lifecycle capture and metadata handling.
• Reduced noisy/redundant OpenCode events in stored telemetry.
• Added/expanded smoke coverage for reliability endpoints and derived views.
• Updated backend API docs for new reliability and outcome endpoints.

Release intent

This is an internal beta focused on validating ARE workflows with real team usage: labeling quality, KPI interpretability, and cross-runner operational signal.

v0.2.0-beta.2 - Internal Beta

Suboculo v0.2.0-beta.2

Documentation and stabilization follow-up to v0.2.0-beta.1.

What’s new

• Synced project docs with current implementation:
  • attempt-based task run semantics
  • Events-tab attempt filtering
  • Task Runs + outcome workflow
  • backend API coverage for task runs/outcomes/benchmarks
• Added/clarified operational guidance:
  • derive task runs after upgrades
  • optional legacy run cleanup command (cleanup:legacy-runs)
• Minor reliability hardening for attempt boundary derivation:
  • malformed session.start without sessionID does not create attempt splits

Why this release

v0.2.0-beta.1 introduced major Phase 2 behavior changes. This release aligns docs and operational instructions so teams can adopt the beta without ambiguity.

Validation status

• Backend smoke suite: passing
• Frontend production build: passing
• No intentional breaking changes vs v0.2.0-beta.1; this is a documentation/stabilization patch release.

v0.2.0-beta.1 - Internal Beta

This release advances Suboculo from event logging toward task-level reliability analysis for internal beta usage.

Highlights

• Attempt-based task runs
  • Task runs are now derived as attempts, not one perpetual run per root session.
  • Attempt boundaries are inferred from:
    • session.start
    • post-session.end continuation
    • inactivity gap (45 minutes)
  • Active runs without explicit session.end remain running (no false completed).
• Outcome semantics enforcement
  • Structured outcome taxonomy is now enforced on write:
    • evaluation_type
    • outcome_label
    • failure_mode
    • failure_subtype
  • Backend rejects invalid or inconsistent outcome payloads.
  • New taxonomy metadata endpoint supports controlled UI forms.
• Outcome analytics
  • Added summary API for task run outcomes with aggregate breakdowns by:
    • outcome label
    • failure mode
    • failure subtype
    • evaluation type
  • Added filterable Task Runs views for canonical outcome and intervention-related analysis.
• Attempt-aware event exploration
  • Events API now supports filtering by attempt key.
  • Events UI includes an Attempt filter and displays attempt context per event.

Reliability/Data quality improvements

• Claude Code session.start capture now prioritizes hook payload session_id over environment-only fallback, reducing null session IDs.
• Derivation logic ignores malformed session.start without concrete session identity for attempt boundary splitting.

Operational additions

• Added optional cleanup utility for legacy run model rows:
  • npm --prefix backend run cleanup:legacy-runs
  • removes task_runs with source='derived_session' after migration to attempt model.

Validation status

• Backend smoke suite: passing
• Frontend production build: passing
• Syntax checks: passing

Compatibility note

• Existing historical data is preserved.
• Attempt model is now the canonical source for new task-run analysis (source='derived_attempt').
• Legacy rows may remain until optional cleanup is run.

v0.1.0-beta.1 - Internal Beta

Suboculo internal beta release.

Suboculo is a local per-project monitoring and analysis tool for AI coding agents. It captures agent activity into a project-local SQLite database, serves a web UI for inspection, and exposes MCP analytics tools for CLI workflows.

Included in this beta:

• Claude Code integration via hooks and direct SQLite capture
• OpenCode integration via plugin and direct SQLite capture
• Real-time web UI with filtering, tagging, notes, saved analyses, and selection-to-CLI bridge
• MCP analytics server for querying sessions, events, and selections
• Per-project installation with custom port support

Supported workflows:

• inspect tool usage and agent/subagent activity
• analyze selected traces in Claude and save results back to the UI
• compare Claude Code and OpenCode activity in the same project database
• run Suboculo locally without requiring a central service

Install:

• ./install-suboculo.sh /path/to/project
• ./install-suboculo-opencode.sh /path/to/project

Documentation:

• README.md
• INSTALL.md

Notes:

• this is an internal beta / team preview release
• intended for local per-project use
• rough edges are still possible; report issues as you find them