Changelog

1.7.6 — 2026-11-03

• Suppress drift publication for clean runs; retain issues for genuinely unchecked sources and review proposals for actual changes. Existing issues are not closed automatically.
• Separate governance certification from explicit authorization of an unverified read-only analysis. Retain strict organization policies, known revocations, semantic blockers and normal query/result checks.
• Keep authority setup and repair with maintainers; carry accepted governance limitations into result verification rather than repeatedly blocking users.

Upgrade notes:

• Existing context repositories need a reviewed support-file upgrade for the drift fix and governance guidance. Updating the plugin alone does not replace those files. Authority runtime and certified export checks remain fail-closed.
• No ACF schema migration or aiden-core change is included.

1.7.5 — 2026-11-02

• Prefer scoped dbt evidence and compact tool discovery while preserving model tests, cross-column constraints, provenance, and required continuations.
• Make the provisional analytical IR explicit and strengthen evidence stopping rules without weakening plan approval or governance checks.
• Add four retrieval behavior eval cases and a synthetic dbt retrieval probe. The probe declares its sources, uses a local test macro, rejects incomparable token measurements, and checks combined-path acceptance against the baseline.

Upgrade notes:

• Update the installed Nodal Analytics package; no ACF migration is required.
• Compatible with existing connectors. Scoped aiden-core retrieval can ship independently and is used when its capabilities are advertised.
• Real MCP token savings have not yet been measured; supplied-evidence evals and the probe's offline tests do not establish production retrieval savings.

1.7.4 — 2026-10-28

• Fix merge approval never recording in scaffolded repos. The governed content digest now uses the same file scope as the merge snapshot, so the shipped evals/seeds/_seed-template.yaml (and other non-contract files) no longer make the checkout and merged revision disagree.
• Add README instructions for updating an existing installation.

Upgrade notes:

• Context repositories created from 1.7.2 or 1.7.3 could not record merge approvals. After a support upgrade, merge the next reviewed change to record the first approval.
• Content digests change once for repositories with non-contract files under governed directories, so existing freshness records need one fresh source check (any push to the approval branch or a manual context-drift run).

1.7.3 — 2026-10-27

• Remove leftover authored approval-status instructions and Markdown markers from interview guidance, extraction documentation and customer templates. Keep owner rosters status-free; distinguish interview evidence from protected merge approval.
• Stop drift patches from inventing legacy status markers on unmarked query patterns. Preserve explicit legacy migration support and runtime eligibility.

1.7.2 — 2026-10-27

This release includes all changes since 1.6.4: the ACF 0.3 migration, merge-based approval, adversarial-review fixes and customer workflow relocation. Versions 1.7.0 and 1.7.1 were superseded before publication.

• Move customer approval and drift workflows into the context template so they no longer execute in the package source repository. Fresh scaffolds and support upgrades still install both workflows under .github/workflows/ in customer repositories, with their existing approval and freshness behavior unchanged.

• Correct adversarial-review regressions in CI revision identity and skip exits, clearance inheritance, post-merge freshness, deterministic/clean exports, scoped governed consumption, dbt lineage, method literals, merge approval history, notebook rollback, routing, simulation and numeric snapshot row keys.

• Replace authored definition approval with protected human-reviewed merge and revision-bound live authority checks. Keep branch testing explicitly unapproved.

• Persist stale blocks independently of repair PRs; propagate through dependencies and require reviewed, checked, revision-specific clearance with retained history.

• Migrate shipped legacy content with owner-authorized admission, preserve review history, separate result verification, and verify exports against approved sources.

• Adopt ACF 0.3 per-metric YAML, shared fragments, aliases and families, review history, explicit blockers, ratio policies and cross-domain dependencies. Retire ACF 0.2 and legacy metric layouts immediately; provide a one-time, comment-preserving YAML converter with backups and conflict protection.

• Add governed methods, static notebook import, and independent notebook/method execution comparisons with separate human confirmation and revision checks.

• Match questions structurally and report definition, slicing and verification coverage separately. Add negative seeds, reviewed variants and resumable observed-demand interviews; keep answer keys out of answer generation.

• Add encrypted approved SQL, customer-controlled regression execution, dependency-aware drift review proposals, gated seed re-grading and optional semantic warnings. Missing capabilities are reported, never counted as passes.

• Preserve quick simulated-analyst runs and resume, with private draft-capture summaries and no fabricated approval or execution evidence. Update package guidance and scaffolds, and add a complete fictional ACF 0.3 worked example.

• Report incoming authoring-template differences during support upgrades while preserving customer instructions, configuration, evidence and lineage baseline.

• Add behavioral evals for family choices, method fidelity, open-decision blockers and missing-data declines, retaining the should-not-fire control and ablation.

• Live host, warehouse, dashboard and customer notebook acceptance remains an opt-in maintainer check; offline tests do not establish those results.

Upgrade notes:

• Existing context repositories need an explicit ACF 0.3 migration and support upgrade; updating the plugin alone does not migrate authored context. Back up existing content and review the conversion before merging. Admission of legacy drafts requires each repository owner's authorization.
• Production consumers require the protected approval branch, governance state, workflow identity and trusted live reader described in GOVERNANCE.md. Conversion alone grants no live approval.
• ACF_BRANCH_TEST=1 permits local testing of proposed context only. Remove it for production use and verify approval and freshness through the live authority.

1.6.4 — 2026-10-19

• Preserve the substantive business request through clarification and approval.
• Carry explicit question/plan references and client lineage across delegation when supplied; never collect transcripts.

1.6.3 — 2026-10-16

• Add privacy policy and support links to the Claude plugin manifest for the plugin directory listing.

1.6.2 — 2026-10-16

• Declare the Nodal logo as the Claude plugin directory listing icon.

1.6.1 — 2026-10-16

• Prepare OpenAI directory metadata with a compliant subtitle, support link, desktop-only execution requirement, and guided setup after installation.
• Refresh submission instructions and release notes, including clean-package checks and the separate user-configured warehouse connection requirement.
• Point the Launch subscription link at the Nodal-branded checkout domain.

1.6.0 — 2026-09-16

• Add ACF 0.2 nested AND/OR filters, per-entity eligibility conditions, and same-file metric references with inherited lineage.
• Validate measure grammar, reference cycles and dimension subsets; optionally check lineage columns against local dbt manifests.
• Update authoring guidance, draft worked examples, and generated scaffolds.

1.5.4 — 2026-09-09

• Add the portable Agent Plugins manifest and production-ready Nodal logo for OpenAI's universal plugin directory while retaining Codex and Claude compatibility manifests.
• Add public privacy, terms, and support pages for the open-source plugin.
• Add OpenAI listing copy, release notes, starter prompts, and reproducible reviewer cases for a skills-only submission.

1.5.3 — 2026-09-09

• Require a fresh progress line immediately before every interview question, including selection-tool calls, follow-ups, and verification prompts.
• Reinforce the question loop across full interviews and test drives, retain visible progress when counts need rebuilding, and keep remaining estimates scoped to the domain round across stage transitions.

1.5.2 — 2026-09-08

• Show a combined text progress indicator during full interviews and test drives, with the current stage, answered questions or completed core topics, and an adaptive estimate of remaining questions.
• Explain added follow-ups, keep deferred topics outside the quick-test budget, and show live-verification progress separately from interview questions.

1.5.1 — 2026-09-04

• Add a shipped Sigma dashboard-verification playbook with explicit organization, workbook, and page resolution for multi-workbook deployments.
• Capture Sigma workbook/page identity, version, controls, resolved date state, and partial-filter limitations alongside each page-scoped result.
• Define safe tiered extraction for Sigma network payloads, visible elements, and local on-demand exports without credentials, scheduled delivery, or writeback.

1.5.0 — 2026-09-03

• Add opt-in isolated host-state testing with a fresh Codex home or Claude bare mode while preserving the existing clean-project default.
• Launch interactive Codex clean-room tests with the required workspace-write sandbox so the sibling context repository is accepted as a writable root.
• Install local or Git-ref Codex marketplace snapshots inside the isolated home and support authenticate-then-launch prepared rooms without copying secrets.
• Capture non-interactive transcripts and add a browser-install lifecycle scenario that checks configured state, restart and approval guidance, and suppression of repeated installation offers.
• Add a public agent onboarding guide and an isolated pre-install scenario that verifies one-distribution installation, consent boundaries, and the new-task handoff to setup-nodal.
• Document the separate macOS-account or VM smoke test required for faithful Codex Desktop, browser, approval, and restart coverage.

1.4.0 — 2026-09-02

• Clarify the current technical operator and domain-expert audiences, including enterprise adoption through verified domain-by-domain rollout.
• Separate representative evaluation from production onboarding and document optional automated dashboard verification before browser consent.
• Persist browser intent as ask-when-needed, manual, or automated; suppress repeated installation offers and distinguish a required restart from a missing binding.
• Clarify how Nodal's active interview, evidence, and human confirmation differ from passive documentation generation.
• State explicitly that the open-source package sends no data or credentials to Nodal, and distinguish it from the separate paid, opt-in hosted path.

1.3.0 — 2026-08-27

• Add read-only query-history visibility probes and document permissions, retention windows, degraded modes, and privacy caveats for Snowflake, BigQuery, and Redshift.
• Add product and context-validation diagrams covering the interview, governed answer, dashboard verification, and enterprise learning loops.
• Strengthen the five-question test drive and dashboard-validation handoff, including filter-aware captures and human-read fallback behavior.

1.2.0 — 2026-08-27

• Add challenge-result for user-requested skeptical second reviews of completed analytics answers.
• Add context-isolated and structured same-context challenge modes that test both the reported number and the user's expectation without treating either as ground truth.
• Add UPHELD, REPLAN, and INCONCLUSIVE outcomes with assumption and hypothesis ledgers, bounded read-only diagnostics, and escalation gates.
• Route material corrections back through analytics-plan so a revised plan receives fresh approval before any new execution.

1.1.0 — 2026-08-27

• Add analytics-plan as the single plan-before-query workflow across ACF, KTX, dbt, local documentation, approved documentation MCPs, and warehouse evidence.
• Add an auditable, explicitly uncalibrated uncertainty v0 with expert-escalation gates before and after execution.
• Add verify-result for plan-fidelity, grain/join, and plausibility review.
• Add analyst-handoff as a transition-focused orchestrator over context-interview.
• Replace generated data-question with the universal analytics-plan; generated context repos remain directly consumable through AGENTS.md and CLAUDE.md.
• Extend .nodal.local.json with classified local and MCP context sources while continuing to reject secrets and raw connection errors.

1.0.0

• Package context-interview, dashboard-verify, and explicit-only setup-nodal for Claude, Codex, and skills.sh installation.
• Add the manifest-driven, hash-checked context scaffold distribution mirror.
• Add guarded project-local configuration and consent-based browser binding setup.
• Preserve clone-and-run development and generated-repository workflows.
• Add a tracked, non-destructive clean-room integration harness while keeping operator briefs, connection configuration, and permissions local.

Release checklist

• Run python3 scripts/sync_skill_payload.py --check and all tool CI locally.
• Install from a temporary checkout with Claude; confirm all seven namespaced skills and bundled hidden files/scripts are present.
• Install from a temporary checkout with Codex; list the available plugin and confirm all seven skills load.
• Install into a temporary project with npx skills@latest add; confirm scripts, executable modes, .claude, .gitignore, and .gitkeep survive.
• Invoke setup in each native host, confirm .nodal.local.json, and consent to a test .mcp.json merge without overwriting an existing binding.
• With authenticated read-only connections, run all three capability probes.
• Complete a context-interview test drive against a temporary context repo.
• Exercise analytics-plan against ACF, KTX, and an approved documentation MCP; verify an evidence conflict produces a clarification and escalation.
• Run verify-result against a missing mandatory filter and a fanout fixture.
• Run challenge-result against an upheld answer, a material time/filter defect that produces REPLAN, and a missing-evidence case that produces INCONCLUSIVE; confirm no revised SQL runs before approval.
• Complete an analyst-handoff domain and confirm it emits eval seeds.
• Run the tracked clean-room harness against the source checkout, installed Claude and Codex plugins, and a temporary skills.sh installation; complete its resume check and generated-repository assertions.
• Install the Codex plugin in a fresh isolated host home and run the browser-install-lifecycle scenario against both the local release candidate and its Git release ref.
• Run agent-guided-onboarding from an empty isolated Codex home, then paste the public agent-guide prompt into the clean Desktop smoke-test account.
• From a separate macOS user or restored VM snapshot, smoke-test Codex Desktop installation, project-server approval, browser launch, and the post-install restart path.
• Invoke dashboard-verify against an authenticated dashboard and reconcile at least one value and filter window.
• Confirm repository and isolated installed-skill scaffold smoke tests pass.
• Bump plugin semver for every published package change and add release notes.