Dendro v0.9.1

Dendro v0.9.1

Direct Operator Overview — Modo Direto

Issue #101 adds an evidence-backed Direct Operator Overview with four decision
cards, predefined question chips, a contextual drawer, and a toggle between
operator and analytical modes. The Context-Tax spotlight helps operators inspect
context overhead and the evidence behind it. Production repairs align the
persisted overview, scoped replay, API, and dashboard paths and avoid nested
transactions during overview evaluation.

Windows reliability

• CLI stack exhaustion: CLI parsing and async runtime entry now run on a
  dedicated dendro-runtime thread with a 4 MiB stack, addressing Windows stack
  exhaustion reported as 0xc00000fd.
• Named pipe contention: A bounded async retry loop handles
  ERROR_PIPE_BUSY (231) while opening a connection, with at most 25 retries and
  a 10 ms delay between attempts. Other open errors return immediately; requests
  are not replayed after delivery.
• Synthetic fixture portability: Windows overview fixtures use
  create_user_private_root_with_choice with StorageRootChoice::UserProfile
  so synthetic databases and keys reside in admitted private roots. This changes
  fixture placement without weakening native privacy checks.
• Migration provenance: Canonical file identity deduplicates discovered and
  explicit provenance paths, normalizing aliases with and without the Windows
  \\?\ verbatim prefix. Backup names retain their source paths, and distinct
  files with colliding names still fail rather than overwrite a backup.

Historical migration proof

The historical migration proof fixture explicitly uses DuckDB's v1.5.6
storage compatibility setting and verifies storage format v68 in the file
header. This keeps the synthetic database readable by the historical binaries
used in the backward compatibility proof. The candidate's reported CLI version
is bound to its manifest version instead of a hard-coded historical release.
These are proof-harness changes, not a claim of compatibility with every DuckDB
storage format.

Publication disclosure

Publication is authorized under the person's explicit override: "force merge and deploy and conclude".
At PR #104 source merge into development (5ac4d4d82e1ae0ce14d10b2397642330f92d1a83), final required CI checks were pending and fresh independent QA and Phase 6 review acceptance were absent. This release does not claim completed acceptance by those gates or represent future CI conclusions.

Source basis

These notes cover changes after v0.9.0 (3f1b29cc649f1d7f14cf147c0bcad5c31218ca88)
through candidate source f1191fcd9be9252ef200cf57323a0b7b191bf4df.
The v0.9.1 preparation commit updates release metadata and documentation only.

Known limitations (Native Windows migration verification)

In source candidate f1191fcd9be9252ef200cf57323a0b7b191bf4df on native Windows CI Job 123275004360, four tests passed and two migration_v5 integration tests failed with Io(os error 5: Access is denied):

• exclusive_backup_pointer_cutover_and_rollback
• windows_provenance_aliases_backup_once_and_preserve_rollback

All Issue #101 targets (cli_doc_parity, connected_dashboard, direct_overview_regressions, quack_owner) passed on native Windows. Publication proceeded under person override; this notice records the known Windows migration test limitation truthfully without waiver.

• Tracked follow-up issue: #106

Dendro v0.9.0 — Connected Intelligence Studio & 5-Layer Capability Matrix

Dendro v0.9.0 Release Notes — Connected Intelligence Studio, 5-Layer Capability Matrix & Active Intelligence Consoles

Release Date: 2026-10-28
Tag: v0.9.0
Previous Release: v0.8.0

---

Executive Summary

Dendro v0.9.0 delivers an Apple-grade Connected Intelligence Studio, comprehensive 5-layer capability intelligence, 4 active intelligence consoles (I22–I25), offline fleet explorer fixtures, and single-owner trusted daemon services for multi-agent engineering fleets.

With this release, Dendro transitions from a local flight recorder into an integrated, causal operational intelligence platform. Fleets running across Claude Code, OpenAI Codex, Google Antigravity, Grok, Pi, and orchestrators like OpenRig and AgentFlow can be inspected in real time, scrubbed across arbitrary historical intervals, and evaluated with rigorous factual accountability. In addition, the repository has transitioned to the Functional Source License (FSL-1.1-Apache-2.0).

---

Highlights

1. Apple-Grade Connected Intelligence Studio (dendro dashboard --serve)

The studio introduces an immersive, fluid visual surface inspired by Apple design language and professional audio/video timelines:

• The Living Growth Rings: A centralized dynamic radial visualization displaying multi-agent fleet composition, token velocity, and health concentric rings that pulse with activity and react to time scrubs.
• Tactile Time Machine Scrubber: Fluid scrub timeline with one-touch presets (1h, 6h, 24h, 7d, 30d, All Time) alongside a dual-calendar Custom Date Range Picker that filters all telemetry, traces, and metrics consistently across the interface.
• Apple Health Story Highlights: 3 diagnostic metric cards highlighting top cost drivers, capability utilization bottlenecks, and runaway trajectory warnings with direct jump links to underlying causal layers.
• 4-Layer Exploding Stack X-Ray (⌘E / Ctrl+E): An interactive 3D perspective breakdown slicing fleet telemetry into its 4 foundational strata:
  1. Model Layer (Tokens, costs, cache hit rates)
  2. Harness Layer (Sessions, agents, turns, tool execution)
  3. Coordination Layer (OpenRig pods, seats, handoffs, message lineages)
  4. Environment & Security Layer (Process resource consumption, sandbox policies, authority boundaries)

2. 5-Layer Capability & Execution Intelligence Matrix

Dendro v0.9.0 categorizes every agentic operation across 5 discrete capability tiers:

• 10 Behavioral Tool Families: Deep classification across read, write, edit, command execution, web search, MCP calls, sub-agent invocation, scheduling, task management, and question asking.
• Harness Primitives: Native primitive classification for prompt turns, streaming chunks, tool execution, and context compactions.
• Skills Declared vs. Invoked (FUNNEL I06): Identifies loaded skills, evaluates activation rates, and reveals "zombie skills"—skills declared in prompts or configurations that consume context tokens without ever executing.
• Model Context Protocol (MCP) Servers: Tracks tool discovery, call latency, throughput, and error rates per MCP server endpoint.
• Lifecycle Hooks: Monitors pre-tool hooks, post-turn validators, and environmental traps.

3. 4 Active Intelligence Consoles (I22–I25) & Multi-Dimensional Rollups

Expands Dendro's analytical engine with four dedicated forensic consoles:

• I22 Trajectory Vortex Console: Pinpoints cyclic loops, runaway prompt expansions, and unproductive agent reasoning spirals with visual spiral intensity diagrams.
• I23 Schema Drift Console: Tracks protocol migrations, unexpected payload alterations, and deserialization warnings across evolving agent runtimes.
• I24 Authority Blast Radius Console: Maps tool capabilities to security scopes, pinpointing filesystem escapes, unconstrained shell access, and credential leakage risks.
• I25 Counterfactual Model Routing: Simulates fleet cost, latency, and token distributions if tasks had been dispatched to alternative models (e.g., Gemini Flash vs. Claude Opus vs. Grok Fast).
• Multi-Dimensional Group-By Rollups: Flexible slicing and aggregation across all 12 studio views by model, harness, provider, seat, pod, and status.

4. Offline Fleet Explorer Demo Dataset & Format Specification

• Portable Fleet Package (site/fixtures/fleet-explorer-demo.json): A zero-dependency, self-contained offline dataset enabling instant local exploration of fleet telemetry without requiring a running DuckDB daemon.
• Formal Format Specification (docs/OFFLINE_FLEET_EXPLORER_FORMAT.md): Complete schema documentation for exporting, archiving, and sharing deterministic fleet explorer packages.

5. Trusted Single-Owner Daemon (dendro serve) & Pre-Flight Advisor (dendro advise)

• Single-Owner Daemon (dendro serve): Background telemetry collector and local HTTP endpoint bound strictly to the local user session, guaranteeing private storage isolation and race-free DuckDB access.
• Pre-Flight Advisor (dendro advise): System diagnostics CLI command validating local platform prerequisites, directory permissions (0700), process sampling capability, and harness log readability.

6. License Update (FSL-1.1-Apache-2.0)

• Updated workspace licensing to the Functional Source License 1.1 with Apache 2.0 conversion after two years (LICENSE and NOTICE).
• Added official status badges to README.md including the OpenRig integration badge (Built-with: Openrig).

---

Upgrade & Parity Notes

• Workspace Version: All workspace members (dendro-core, dendro-connectors, dendro-resolver, dendro-watcher, dendro-platform-policy, dendro-windows-private-io, dendro-cli) are synchronized at 0.9.0.
• Database Schema: Schema 5 remains the canonical storage format. Ingestion automatically initializes and upgrades local DuckDB instances without data loss.
• Verification: Passes all workspace tests (cargo test --workspace), doc parity checks (cli_doc_parity), Clippy (cargo clippy --workspace -- -D warnings), code formatting (cargo fmt --check), and web test suites (node --test site/site.test.cjs).

Dendro v0.6.0 — Operator Usability, Zero-Ceremony Dashboards & Smart Defaults

Dendro v0.6.0 Release Notes — Operator Usability, Zero-Ceremony Dashboards & Smart Defaults

Release Date: 2026-10-19
Tag: v0.6.0
Previous Release: v0.5.0

---

Executive Summary

Dendro v0.6.0 focuses on operator usability and zero-ceremony workflows, shaped by a 5-member multi-model Usability Council (Solo Operator Persona, Fleet Supervisor Persona, UX Specialist, Information Architect, and Product Owner across AGY, Codex, and Claude).

Previously, inspecting fleet health or asking a semantic question required knowing internal plumbing: running dendro run, discovering exact RFC3339 --from and --to timestamps in the database, manually running dendro materialize before dendro ask, and matching 6 exact canned strings. Running dendro ui --window 7d against a missing database failed with an opaque Error: analytics database could not be opened, and running it against recorded historical corpora anchored to Utc::now() instead of the recorded data window, rendering 100% UNKNOWN.

In v0.6.0, operators can go from zero to insights in a single command while preserving Dendro's strict analytical truthfulness, read-only source guarantees, #![deny(unsafe_code)], and static/parameterized DuckDB queries.

---

Highlights & New Capabilities

1. Zero-Config Entry (dendro & dendro status)

• Running bare dendro (with no arguments) now runs dendro status instead of exiting with a clap usage error; dendro --help and dendro --version remain unchanged.
• dendro status [--db <path>] [--rig <rig>] [--format text|json] inspects the analytics database read-only and reports:
  • Database status (ready, empty, or not_initialized)
  • Redacted rigs and seats, plus usage_rows, activity_rows, and queue_rows counts
  • Exact recorded timestamp bounds (min_timestamp, max_timestamp), scoped to --rig when supplied
  • Semantic snapshot status (semantic_fresh: true/false and semantic_snapshot scope from dendro_semantic_snapshot / dendro_semantic_guard)
  • Discovered local harness roots and copy-pasteable next commands
• Missing database errors now include actionable guidance:
  analytics database could not be opened (not found; run 'dendro run' or 'dendro dashboard --auto-ingest' first)

2. One-Command Dashboard (dendro dashboard) & Conditional --auto-ingest

• dendro dashboard: Launches the interactive terminal explorer (--once for a non-interactive snapshot) over the database's active time window (--window all by default) and displays the resolved window and window_resolution mode in the header.
• dendro dashboard --html [path]: Exports the self-contained, secret-redacted, seat-aliased offline HTML dashboard (dendro-dashboard.html by default) in one step.
• --auto-ingest: Supported on both dendro dashboard and dendro ui. When --auto-ingest is passed and the target DuckDB file does not exist or has no recorded timestamps (status != "ready"), Dendro runs a local read-only ingestion pass over default harness roots before opening the dashboard.

3. Universal Smart --window Anchored to Recorded Data (Rig-Scoped)

• --window 1h|24h|7d|30d|all is supported across dashboard, ui (defaulting to 7d), report, trace, materialize, export, ask, and simulate topology. (report and trace preserve their v0.5.0 --from/--to defaults when --window is omitted.)
• When --from/--to are omitted on a populated database, resolve_window scopes timestamp bounds to --rig (when provided), floors min_timestamp to whole seconds, sets the half-open [from, to) upper bound to max_timestamp + 1s so events at max_timestamp (including sub-second timestamps and single-timestamp databases) are never excluded, and discloses "window_resolution" ("recorded_bounds", "anchored_to_recorded_data", "explicit", or "trailing_utc").

4. Zero-Ceremony dendro ask (Auto-Materialization & Natural Aliases)

• Positional Question: dendro ask "highest idle" works directly alongside --question.
• Auto-Materialization: When the semantic snapshot (dendro_semantic_snapshot / dendro_semantic_guard) is missing or stale, the CLI dendro ask automatically refreshes the 4 canonical views (v_seat_idle_ranking, v_handoff_rework_analysis, v_mission_token_burn, v_executive_governance) over the active window (writing the derived semantic tables in DuckDB) unless --no-refresh is set, recording "auto_materialized": true and "window_resolution" in the JSON response. Calling dendro_core::semantic::answer in Rust or passing --no-refresh preserves strict read-only refusal on stale snapshots.
• Natural Intent Discovery (--list-intents): dendro ask --list-intents (or bare dendro ask with no question) prints all 6 canonical intents and example phrases without requiring a database.
• Expanded Natural Phrases: Recognizes natural questions for idle time ("idle", "highest idle", "where is time being wasted?"), yield/token burn ("worst yield", "token burn", "where do tokens go?"), rework ("rework", "handoff friction", "why is there rework?"), verification ("verify", "provenance"), trace ("trace idle", "trace tokens", "trace handoffs", "trace rework"), and honest limits ("limits", "known limits"), while still refusing out-of-scope mission/date qualifiers truthfully.

5. In-App TUI Keyboard Help (? / h)

• Pressing ? or h in dendro ui / dendro dashboard opens an interactive help view documenting panel cycling (Tab, Left/Right), seat selection (Up/Down, Enter), forensic trace inspection (1/2/3), and exit (q, Ctrl-C).

Dendro v0.5.0

Dendro v0.5.0 Release Notes (Goals 1–5 Complete)

Release-candidate documentation and independent verification receipts for Dendro v0.5.0, completing Goals 2–5 atop the released v0.1.1 (R1) core engine. Tagging and external publication remain governed by operator release authorization.

---

Release Candidate Identity & Provenance

Field	Value
Version	0.5.0 (workspace.package.version across dendro-core, dendro-connectors, dendro-resolver, dendro-watcher, dendro-cli)
Base Commit (main)	25e2c65 (incorporating PR #56, PR #57, PR #58 81f5a90, PR #68 b15397e, PR #69 25e2c65)
Release Binary (target/release/dendro)	SHA256 555c2ed059b36f0bb250b0a90e6688e63d099c5995201fafb471845ee628bd23 (dendro 0.5.0)
Toolchain	rustc 1.98.0 (88d9e12ae 2026-08-18) / cargo 1.98.0 (797e8a9bc 2026-08-05)
Base Analytical Contract	docs/ACCEPTANCE_R1.md (cc365c88…), Clarifications Register K1–K11 (44a3032e…), docs/ACCEPTANCE_G2_STAGED.md

---

What Ships in v0.5.0 (Since v0.1.1)

1. Goal 2 — Conversational & Semantic Analytics, Sealed Replay & Terminal Explorer (#7, #12–#14, #16–#18, #30–#34, #52–#55)

• Rework Vortex & Pair Friction Matrix (#12): Pair-level handoff/return breakdown, M3 rework chain attribution, handoff-loop detection (>2 passes on the same pair), and configurable friction alert thresholds (--friction-return-rate, --friction-chain-returns, --friction-rework-tokens, --friction-loop-duration).
• Live-Producer Snapshotting & Sealed Replay (#13, #31, #32): Atomic byte-cutoff manifests (dendro run --write-manifest <path> and dendro run --sealed <path>), unified source-state disclosures (available, gone, replaced, absent), and single-OpenRig-root-per-DB enforcement.
• Canonical Semantic Views & Local Interrogation (#16, #17, #18):
  • dendro materialize populates four typed DuckDB semantic views (v_seat_idle_ranking, v_handoff_rework_analysis, v_mission_token_burn, v_executive_governance) using typed appenders under a 1 GB refresh cap.
  • dendro ask answers six deterministic English interrogation patterns (Compare, Rank, Explain, Verify, Trace, HonestLimits) with exact token/handoff conservation against base tables.
  • dendro init-nao creates a local-only Nao workspace (0700/0600) without overwriting existing directories.
  • Legacy sidecar compatibility (#59): sidecar databases missing optional tables (file_source_baselines, openrig_row_records, mutable_source_disclosures) refresh and invalidate cleanly using parameterized information_schema.tables probes.
• Interactive Terminal Explorer (#14): dendro ui (24h/7d trailing or fixed --from/--to window, plus --once) and dendro report --viz, with keyboard navigation, single-ESC reset on navigation, and C1 control-character escaping (\uXXXX).
• Extended Fleet Signals (#52–#55): Ceremony vs. engineering tool categories, $k$-seat concurrency histograms, mid-phase compaction hazard detection (>50% context drop while in-progress), recovery item separation (qitem-recovery-*), ring boundary classification (in_ring/out_of_ring), and (model, effort) breakdown.

2. Goal 3 — Portable Harness Connectors, SDK & Cross-Project Portfolio (#8, #19–#21, #27)

• Verified-Export Stream Connectors (#19, #20): Incremental watermark-backed connectors for Cursor, Windsurf, Devin, and Goose portable verified-export session logs, enforcing secret redaction and path containment.
• Custom Harness Developer SDK (#21): Documented in docs/extensions/README.md and docs/extensions/formats.md with tutorial and example (crates/dendro-connectors/examples/custom_harness.rs).
• Cross-Project Fleet Portfolio (#27): dendro portfolio aggregates independently ingested analytics databases (--sources or --source-dir) in json, ndjson, or table formats with B1–B4 hold evidence and strict non-symlink hold_dir containment (#59).

3. Goal 4 — Interactive 3-Layer Visualizer, Offline HTML Export & Waste Calculator (#9, #22, #23, #28)

• Interactive Landing Visualizer & Dynamic Cost-of-Inaction Calculator (#22, #23): Shipped under site/ with deterministic Sankey/timeline rendering and empirical financial waste sliders.
• Scrubbed Offline HTML Export (#28, #59): dendro export --format html --output <path> writes a self-contained, aliased 0600 HTML viewer and unlinks partial files on write failure without touching pre-existing files.

4. Goal 5 — Advisory Governance, Anomaly Spotlights & Topology Simulator (#10, #15, #24–#26, #29, #66)

• Closed-Loop Backlog Feeder (#15): dendro issue-draft generates local Markdown operational-debt drafts with quantified cost-of-inaction (and opt-in --publish --repo OWNER/REPO).
• Automated CI Quality Gate (#25): dendro assert evaluates M3 rework rate and idle-with-claimed-work thresholds with deterministic CI exit codes (0 pass, 2 breach, 3 insufficient evidence).
• Idle-with-Claimed-Work Advisory (#26) & Anomaly Spotlights A1–A6 (#29): dendro report spotlights surfaces six proactive operational hypotheses (ceremony tax, serialization bottleneck, container hold, recovery churn, context-drop hazard, builder saturation) as report-only evidence.
• Historical Topology What-If Simulator (#24): dendro simulate topology models seat removal, replacement seat routing, direct handoffs, tier multipliers (--tier-policy), boundary resets (--boundary-reset-threshold), and model-family diversity (--enforce-model-family-diversity) over canonical decision reports while conserving historical engineering demand.

---

Release-Readiness Verification Receipts

1. Deterministic Quality & Security Gates

• cargo fmt --check: pass
• cargo check --workspace --locked: pass
• cargo clippy --workspace --all-targets --locked -- -D warnings: pass (0 warnings)
• cargo test --workspace --locked: 379 passed, 0 failed, 1 ignored across 22 test suites (including the 5,500-fact + 16 MB extension materialization benchmark under 512 MB RSS).
• node --test site/site.test.cjs: 13 passed, 0 failed, 1 optional fixture skipped.
• #![deny(unsafe_code)]: enforced across all 5 workspace crates (dendro-core, dendro-connectors, dendro-resolver, dendro-watcher, dendro-cli); 0 unsafe blocks in workspace source.
• SQL parameterization: 0 dynamic format!() SQL queries in production library/binary code; all runtime SQL uses static statements or duckdb::params! / rusqlite::params!.

2. Real-Log Corpus Benchmark (AC1 — ac1_real_logs_finish_within_five_minutes)

• Measured elapsed time: 96.24 s (96,243 ms), leaving 203.76 s of margin below the 300 s AC1 ceiling.
• Corpus ingested (--release binary):
  • Claude Code: 138 session files
  • Codex: 127 session files
  • Pi: 126 session files
  • OpenRig SQLite snapshot: 71,686 events

3. Release Binary 12-Subcommand CLI Smoke Sweep

Executed with /home/sam/code/dendro/target/release/dendro (0.5.0, SHA256 555c2ed059b36f0bb250b0a90e6688e63d099c5995201fafb471845ee628bd23):

• dendro --version → dendro 0.5.0
• dendro report decisions / dendro report spotlights / dendro report decisions --viz → pass
• dendro trace --metric-id tokens → pass
• dendro materialize / dendro ask / dendro init-nao → pass
• dendro export --format html → pass (0600 file mode verified)
• dendro portfolio → pass
• dendro simulate topology → pass (status: projected / status: no_data, conserved: true)
• dendro issue-draft / dendro assert (exit code 3 on insufficient-evidence fixture) / dendro ui --once → pass

---

Explicit Known Limits & Truthful Scope Boundaries

1. Observational Intelligence Only: Dendro opens sources read-only and never mutates live OpenRig state, agent seats, or runtime routing.
2. Topology Simulation (dendro simulate topology): Computes deterministic accounting projections over recorded historical tokens and handoffs under explicit operator assumptions. It is not a causal predictor of future LLM productivity or hazard elimination.
3. Conversational / Nao Layer (dendro materialize, dendro ask, dendro init-nao): Local DuckDB semantic views, deterministic English intent answering, conservation checks, and local Nao workspace generation are verified in automated tests. Live external cloud-LLM Nao operator sessions are not fabricated in local CI.
4. Editor & Autonomous Connectors (Cursor, Windsurf, Devin, Goose): Ingest verified portable JSONL/JSON session exports under authorized local roots; private proprietary cloud databases and remote webhooks are explicitly out of scope.
5. Cross-Project Portfolio (dendro portfolio): Aggregates independent per-project DuckDB databases and local hold manifests; it does not claim cross-database atomic point-in-time snapshots or live unrecorded hold residence durations.
6. Codex Tool Calls (inherited from v0.1.1 Known Limit 11): Codex structured tool-call loop counts remain marked not read where Codex logs omit structured file-write/tool-result pairs.