Changelog#

All notable changes to the PALO Framework are documented in this file.

The format is based on Keep a Changelog, and this project follows Semantic Versioning where practical.

[Unreleased]#

Added#

• Added the first-class Ask PALO Governance Hub view and a separate production-capable Knowledge Copilot BFF with OIDC/PKCE sessions, in-memory CSRF handling, PostgreSQL session durability, Reader client credentials, exact six-tool MCP validation, stateless model synthesis, canonical citation validation and explicit insufficient-evidence results.
• Added a hardened standalone Ask PALO deployment profile with a pinned non-root Distroless image, read-only filesystem, loopback-only Compose port, file-mounted secrets, TLS/nginx rate limits, query-free OIDC access logging, deployment smoke tests and an exact-digest live qualification checklist.
• Added a permanent Observatory index connecting the AI Incident Observatory, Regulatory Watch, 2026 Tech Trends and Human Agency Risk Map through explicit Gold Case, Monitoring Brief and Candidate Under Review publication states.
• Added an AI incident investigation route to Platform Map, plus exact Knowledge Reader and Knowledge Curator maturity boundaries and the disabled-by-default external incident provider state.
• Added a dedicated stateless PALO Knowledge Reader production candidate: exact six-tool server, canonical-only content, fail-closed release digests, strict OIDC/HTTPS admission, request and rate limits, non-root minimal image, volume-free Compose profiles and deployment-specific live qualification gates.
• Added separate six-tool Knowledge Reader and ten-tool Knowledge Curator MCP profiles, provenance-bearing search/get, immutable reviewed local contributions, OIDC roles, canonical and terminal-/mcp HTTPS routes, profile instructions, paired configurations for 11 MCP hosts, a machine-readable conformance matrix, host qualification procedure, automated configuration validation and a portable PALO Knowledge Copilot skill.
• Added a public PALO 3.1 and PALO-AI 2.7 release verification record with the baseline source revision, CI run, validation totals, negative-test coverage, tagged artifact checksum and current maturity limits.
• Added a production release commit checklist with isolated Git staging groups, deterministic site-output handling, private-input exclusions and mandatory validation gates.

Changed#

• Updated root and Governance Hub dependency locks to patched compatible releases; both npm audit scopes report zero known vulnerabilities at implementation time.
• Published cache-safe, versioned Case 001 stylesheet, texture and PNG paths, plus non-transformed full-resolution PNG ZIP downloads, while retaining the previous asset names as compatibility copies.
• Migrated Community to the shared PALO 3.1 visual system, removed third-party style and font dependencies, reconciled shared navigation around the Observatory index and corrected public changelog routes to the generated HTML view.
• Remote OIDC MCP listeners now fail closed unless explicit client and tenant allowlists are configured, and every listener must use the exact host validated when its app was created. Loopback evaluation compatibility covers IPv4 127/8 plus canonical and expanded IPv6 loopback forms, while IdP audience identifiers distinct from the public resource URL remain supported.
• Redacted deployment identifiers, account addresses, credential rotation details and billing state from public Knowledge Reader documentation while preserving control design and qualification status.
• Corrected Knowledge Reader snippet offsets so punctuation normalization cannot move the displayed excerpt away from the matching source term.

[PALO-AI 2.7.0] - 2026-08-25#

Release status: Data-assurance developer preview. The bundled runtime remains denied for production use.

Added#

• Added the PALO-AI v2.7 data-assurance vertical slice: immutable external context evidence, deterministic purpose-bound Data Fitness Decisions, signed Data Disclosure Contracts and Receipts, AI System & Agent Registry records and continuous-assurance signals.
• Added Action Claim 1.4, binding an identity/tenant/Effect Contract to exact current Data Fitness and Disclosure Contract digests before policy evaluation and one-time capability issuance.
• Added zero-row, aggregated and row-level disclosure checks across source/read fields, row limits, sensitive categories, redaction, recipient, provider, model, region, endpoint, trace retention, export and output schema.
• Added an Actian Context Bridge normalization profile that retains normalized claims and a source-payload digest without retaining sample rows or the source payload.
• Added fail-closed invalidation of prior allowed fitness decisions and revocation of matching unconsumed capabilities after quality, classification, lineage, access, contract, incident, system or policy signals.
• Added 8 JSON Schema contracts, valid fixtures, 12 MCP tools and end-to-end tests for payload minimization, verified disclosure, mismatch incidents and stale-decision denial.

Changed#

• Action Claim 1.4 executor results are digest-only in persistent execution state; row payloads are not retained by the PALO runtime record.
• Execution Receipt 1.1 binds the signed Data Disclosure Receipt, and authoritative outcome verification treats disclosure mismatch as a held high-severity assurance incident.
• Data Fitness now denies conflicting current assertions conservatively, binds evaluated normalized claims into its evidence digest and preserves policy, decision and AI System record versions with database immutability guards.
• Data Fitness freshness now evaluates the oldest current evidence reference and caps each decision at both source validity and observedAt + maxEvidenceAgeSeconds, preventing a recent assertion from masking stale evidence.
• OIDC data-assurance operations now have explicit least-privilege scopes and tenant-scoped registry, execution and incident reads; the Rego network-intent check is explicit and disclosure obligations apply only to Action Claim 1.4.
• OIDC deployments can configure the tenant-claim name through PALO_OIDC_TENANT_CLAIM; disclosure observations must fall inside both the signed contract window and the governed execution window.
• Advanced the independently versioned PALO-AI developer-preview capability baseline to v2.7.0 while retaining zero production-ready capability claims and strict production-admission denial.
• Authority Context 1.4 now enforces trusted issuers, workload audience, authentication age, delegation continuity, claim-window coverage and non-widening scope instead of falling through the legacy validation path.
• Cached Action Claim 1.4 decisions are re-evaluated against current authority, Data Fitness and Disclosure Contract state, so continuous-assurance invalidation denies the same previously allowed claim.
• Post-effect disclosure-receipt failure now produces an unknown Execution Receipt, an inconclusive held incident and digest-only execution state without persisting executor row payloads.
• Tenant-aware execution and incident lookup now falls back to legacy Effect Contract or metadata bindings, scopes incident lists in one joined query and safely excludes unscoped legacy records from OIDC results.

Validation#

• Added negative Action Claim 1.4 authority tests for untrusted issuers and delegation windows that do not cover the full live claim.
• Added same-claim cache-invalidation, conservative freshness/expiry, OIDC cross-tenant and configurable-tenant-claim MCP, replayed disclosure observation, post-effect disclosure-receipt failure and legacy-incident isolation regressions.

[3.1.0] - 2026-08-23#

Release status: Governance control plane released. PALO-AI v2.6 remains a non-production developer preview.

Added#

• Added twelve canonical, applicability-aware governance control packs for fairness, system cards, affected-person notice and remedy, Article 50 transparency, data lifecycle, GPAI and systemic risk, serious incidents and decommissioning, accessibility, environment, AI literacy, ISO/IEC 42001 and PALO-AI production admission.
• Expanded the released registries from 9 to 31 controls and from 10 to 38 KPI/KRI definitions, with every new control mapped into one or more lifecycle gates.
• Added ten schema-validated evidence-contract families with valid and invalid fixtures for assurance results, system cards, affected-person cases, Article 50, data lineage, GPAI, serious incidents, decommissioning, AIMS and production profiles.
• Added explicit definitions of governance-complete, operationally complete, production-complete and certification-complete status, with stop conditions and residual boundaries for every domain.
• Added fail-closed PALO-AI production admission and OIDC tenant-to-Action Claim 1.3 binding. The bundled SQLite and in-process runtime is intentionally denied by its own production profile.
• Added PALO-AI v2.6 identity-bound Action Claim 1.3, durable approval/verification tasks, Effect Contract 1.1, RFC 8785/Ed25519 evidence verification, runtime guardrails and operational telemetry events.
• Added OIDC/JWKS MCP resource-server authentication with issuer/audience/expiry/algorithm validation, RFC 9728 metadata, least-privilege scopes/roles, scope challenges and authenticated reviewer attribution.
• Added an allowlisted OpenTelemetry span bridge and the dated August 2026 state-of-the-art radar covering primary sources, Eastern experimental projects and developer communities.
• Added the version-pinned OWASP Top 10 for LLM Applications 2026 source artifact, schema-validated PALO/PALO-AM/PALO-AI crosswalk and source-backed governance reference.
• Added a documentation-first OWASP 2026 analysis that separates direct, supporting and gap ratings from implementation effectiveness, compliance, certification and production authorization.
• Integrated the OWASP 2026 lens into Guide, Assessment Path, Evidence Bundle, Case File and the read-only MCP Guide Agent without creating a new activation module; applicable profiles keep all ten risks in scope and retain OWASP evidence as draft pending human review.
• Added a role-based repository reading guide that explains release tags, main, open pull requests, authoring files, generated dist, validation, contribution and authority boundaries.

Changed#

• Replaced categorical output from the public Risk Tiering tool with an explicitly non-authoritative screening hypothesis and corrected the voluntary-code reference from Article 69 to Article 95.
• Updated Regulatory Watch for the final Article 50 guidance, provider/deployer separation, GPAI sources and effectiveness evidence for AI literacy.
• Made source freshness date-driven so a source marked current fails validation when its next review date has passed.
• Advanced the platform and semantic release to v3.1.0 and the separately versioned PALO-AI capability baseline to v2.6.0.
• Added authorized public credit to Arshi Chadha, OWASP LLM09:2026 co-lead, for the focused technical review covering embedding inversion, retrieval-evasion testing and the LLM05/LLM09 ownership boundary; the credit is explicitly a personal contribution and does not imply OWASP review or endorsement.
• Refined the LLM09 crosswalk following an interaction with the dedicated LLM09 group: added explicit embedding-inversion, retrieval-evasion, similarity-collision and threshold-straddling evidence requirements; clarified the LLM05/LLM09 vector-poisoning boundary and accountable owners; retained the existing Supporting/Supporting/Gap rating and non-elimination boundary.
• Migrated the MCP reference server from the monolithic SDK 1.30 to split SDK 2.0 packages, with one dual-era factory for stateless 2026-07-28 and 2025 compatibility plus pinned-modern and header-consistency tests.
• Raised the root Node.js engine baseline from 20 to 22 and patched the Hono Node adapter to a production-audit-clean version.
• Updated the OWASP source registry entry with the 2026 v1.0 artifact boundary, provisional editorial status and a 30-day official-source reverification interval.
• Made Documentation Library text search global across depth levels, so source-backed OWASP guidance is discoverable directly from the default Start view while the active All-depth state remains visible and accessible.
• Reconciled the Platform Map public-web ledger with PALO Web and semantic release v3.1.0 while preserving independently versioned runtime and mobile components.
• Updated the Hostinger deployment runbook and added a verified package:hostinger command that produces a root-correct upload ZIP and SHA-256 sidecar.
• Updated SECURITY.md for the PALO Web v3.1.x baseline and released non-production PALO-AI v2.6 boundary, and corrected the fork clone path in CONTRIBUTING.md.

Validation#

• Added cross-registry validation for the twelve control packs, exact control/indicator/gate references, minimum evidence kinds, conditional gate integration, evidence schemas and invalid-fixture rejection.
• Added production-admission tests for schema validity, placeholder and expiry rejection, runtime incompatibility and cross-tenant Action Claim rejection.
• Added validation assertions that keep the LLM09 inversion, adversarial-query and LLM05 ownership-handoff requirements present in the machine-readable crosswalk and public dossier.
• Added crosswalk schema, source and control reference checks, route-count consistency, union-summary checks and pinned PDF SHA-256 validation.
• Added horizontal browser coverage for the OWASP route across PALO Guide, Evidence Pack, Documentation Library and Platform Map, including mobile layouts and release-facing consistency checks.

[3.0.1] - 2026-08-12#

Added#

• Added the preloaded local Evidence Pack experience, a voluntary SHA-256 validation receipt and three schema-valid gold cases.
• Added case:contribute, a one-command generator and validator for Community Casebook contributions and pull request preparation.
• Added the five-day activation runbook, 20-seat founding review protocol and public changed-because-of-feedback log.

Changed#

• Normalized tracked source and generated publication text to plain ASCII punctuation, separators and status labels.
• Added an automated text-style gate for source files, generated dist files and current Git metadata.
• Started a 30-day new-module freeze, enabled one canonical GitHub Discussions channel and added the five-day activation and changed-because-of-feedback records.
• Made PALO Evidence Pack the only primary website and README activation route while retaining existing modules as downstream tools.

[3.0.0] - 2026-08-12#

Added#

• Added the canonical PALO semantic spine with stable HTTPS identifiers, definition versions, evidence/authority classes, source references and review dates across 46 public graph entities and their relationships.
• Added lifecycle, Gate Instance and append-only Gate Decision Record contracts with actor authority, rationale, immutable input snapshots and digest chaining.
• Added atomic Evidence Artifact, Evidence Claim, Evidence Evaluation and Evidence Bundle Manifest contracts while retaining P1 Case File and Evidence Bundle v1 compatibility.
• Added a JSON-LD context, RDF ontology, SHACL shapes, valid/invalid Turtle fixtures and executable semantic invariants.
• Added a versioned atomic mapping registry across controls, indicators, gates and sources, plus change-impact and digest-bound semantic release manifests.
• Added a public Semantic Inspector, evidence/authority filters in Platform Map and Documentation Library, and semantic record drawers in the Governance Hub prototype.

Changed#

• The Explorer graph asset is now generated deterministically from the semantic spine instead of being an independently maintained catalog.
• Delivery state is now separate from evidence authority: Canonical definition, Source-backed context, Illustrative local preview and Human review required.
• Executive and Technical in the Governance Hub are explicitly workspace lenses, not access-control roles; all prototype records and exports declare their non-authoritative local-preview boundary.
• P2 registry documentation now reports the canonical 9 controls and 10 indicators.
• Root web/package/release metadata and shared asset cache keys advance to v3.0.0.
• Release metadata now distinguishes the PALO v3.0.0 platform release from independently versioned developer-preview components.
• Documentation Library is the canonical source-backed documentation destination; Documentation Hub remains a backward-compatible transition page.
• README, citation metadata, lifecycle narrative, naming convention and roadmap are reconciled with the v3.0.0 Semantic Foundation release.
• Local assessment outputs and specifically identified private or third-party research inputs are excluded from accidental staging.
• The P0 gate now bootstraps locked Governance Hub dependencies and the verified OPA binary for clean-clone reproducibility.

Validation#

• Added generation-drift checks, JSON Schema validation, RDF parsing, SHACL-subset conformance checks, invalid-fixture assertions, mapping/digest validation and public build coverage to the P0 gate.

[2.5.0] - 2026-07-18#

Release status: Full-Cycle Developer Preview. Isolated evaluation only.

Public surfaces and documentation#

• Added the three-entry PALO/PALO-AM/PALO-AI governance map, cognitive front door, canonical Why PALO-AI and Quickstarts pages, browser-local outcome comparison, capability rail and nine-gate Production Readiness route.
• Added the searchable Documentation Library, deterministic Markdown publication pipeline, Governance Hub prototype, product specification, role-based guide, workflows, diagrams and nested Vite publication.
• Added local, hybrid, managed-cloud and private-cloud guidance, cyber-assurance and scale planning, staged community entry and explicit production-readiness boundaries.
• Added n8n governance architecture patterns, presentation assets, launch material, integration-boundary documentation, installable alpha package, local n8n runtime test report and deferred provenance publishing workflow.
• Consolidated primary navigation around Start, Agentic Governance, Governance Hub, Tools, Documentation and Readiness, with progressive disclosure of contracts and evidence.
• Hardened Governance Hub navigation, downloads, boundary tests and accessibility while preserving the explicit developer-preview and non-certification claims.

Added#

• Added Action Claim 1.2 and immutable Effect Contracts for authoritative preconditions, expected effects, forbidden effects and inconclusive handling
• Added signed one-time Execution Capabilities bound to the exact claim, decision, tenant, resource, executor and verifier
• Added operator-provisioned trusted executor and verifier registries, signed Execution Receipts and signed Outcome Attestations
• Added verified, mismatch and inconclusive outcome semantics so an allowed action is no longer confused with a correct result
• Added held Assurance Incidents with explicit acknowledge/resolve transitions; compensating actions remain separately governed claims
• Expanded the MCP toolkit from eleven to nineteen tools and added full-cycle REST execution, outcome and incident endpoints
• Added n8n-nodes-palo-ai 0.2 with the PALO Governed Action node and four visible outputs: Verified, Review Required, Denied and Execution Failed
• Added a synthetic multi-tenant catalog demo covering verified effect, stale-state prevention and authorized-but-wrong mismatch detection
• Added automated tests for capability replay resistance, idempotent retries, signed receipts, outcome verification, incident holds and ledger integrity

Changed#

• Disabled caller-supplied REST execution evidence by default; governed execution now generates receipt and outcome evidence inside the trusted runtime
• Updated the reference Rego policy to accept Action Claim 1.2 only when a bound Effect Contract is present
• Retained Action Claim 1.1 and the original n8n decision-only node for migration compatibility

Known developer-preview limitations#

• SQLite, shared bearer tokens, environment HMAC keys and in-process connector handlers are not production infrastructure
• External tools cannot join the local database transaction; exactly-once behavior depends on connector idempotency and is not universally claimed
• Executor and verifier workloads are not attested, mobile reviewer identity remains a prototype, and multi-replica durable recovery is not implemented
• Production adoption still requires workload identity, scoped RBAC, KMS/HSM, PostgreSQL and durable queues, HA, observability and independent security assessment

[2.4.1] - 2026-07-17#

Release status: Developer Preview. Not approved for production authorization or consequential tool execution.

This release publishes contracts and a reference implementation for evaluation. Runtime enforcement, cryptographic evidence hardening, authenticated human approval, production connectors, and collaborative agent-team execution remain under development. The included bearer-token transports, SQLite/HMAC components, Vibe Gate metadata, Web/Android approval clients, and n8n/Dify integrations are prototypes and must not be represented as production security controls.

Added#

• Added an experimental bearer-authenticated MCP Streamable HTTP transport using the official SDK, with protocol parity across eleven stdio/remote tools
• Added a reference SQLite WAL registry and append-only evidence-ledger prototype for agents, authorities, policies, replay state, approvals, decisions and signed events
• Added normalized Action Claims with path, host, network intent, validated arguments, nonce, idempotency key and monotonic sequence number
• Added prototype Web and Android Approval Inbox clients, PALO-AM profile/decision exchange, and a demonstrative Vibe Coding claim-metadata gate
• Added an honest public machine-readable capability matrix and an in-process register, deny, approval, execute, sign, persist and verify acceptance test
• Added the non-production PALO governance MCP reference server, expanded from its initial seven-tool baseline to eleven tools for registry, policy, approval, evidence and ledger operations
• Added canonical, interoperable JSON contracts for agent profiles, action claims, policy decisions, approvals, and HMAC-signed evidence envelopes
• Added an experimental localhost governance gateway for non-production Dify, n8n, and workflow examples; policy and signing secrets remain server-side
• Added a persistent approval state machine with expiry, single terminal resolution, exact claim-digest binding, accountable resolver identity, and rationale
• Added an append-only SQLite evidence-ledger prototype with UUID event IDs, secret-field redaction, HMAC-SHA256 signatures, previous-event hash chaining and immutable update/delete triggers
• Added pinned, checksum-verified OPA installation, Rego v1 compilation and policy tests, MCP protocol smoke tests, runtime security tests, connector tests, and CI enforcement

Changed#

• Replaced legacy Rego syntax and permissive delegation shortcuts with fail-closed checks for registered tools, operations, read/write scopes, network access, hosts, delegation depth, subagent count, roles, and human validation
• Replaced local Dify and n8n allowlists, mock signatures, and second-based evidence IDs with canonical claims evaluated by the trusted PALO gateway
• Changed agent profiles so they reference only a key ID and algorithm; signing secrets can no longer be embedded or derived from public configuration
• Preserved every PALO-AM and Assessment Path run as a unique historical record and calculate source freshness from recorded source state instead of marking every export current
• Made Case File merge linear for identified arrays, reject cross-case merges, require strict RFC 3339 timestamps, and surface save or handoff persistence failures before navigation

Security#

• Added nonce replay detection, immutable claim-ID checks, constant-time signature and gateway-token comparisons, authenticated gateway access, request-size limits, redaction defaults, atomic registry writes, and fail-closed OPA outage behavior
• Prevented raw tool arguments, access tokens, passwords, cookies, API keys, and private keys from entering agentic decision evidence

Known developer-preview limitations#

• A single bearer token does not provide principal identity, role separation, reviewer authentication, administrative authorization, rotation, or production transport security
• The reference executor is not an exactly-once execution engine; cached decisions, action expiry, state/evidence atomicity, and decision provenance require further hardening before real tool use
• Approval clients demonstrate state transitions but do not yet provide sufficient human-readable action context or an enterprise reviewer identity lifecycle
• Vibe Gate metadata is a self-attested demonstration and is not a trusted signed attestation or an unavoidable pre-tool-call proxy
• Collaborative Agent Teams remain specified only; Team Registry, Shared Task Claims, peer coordination, leases, conflict handling, and team-level evidence are not implemented
• n8n and Dify artifacts are non-production examples and do not provide a certified end-to-end approval-resume or execution connector

[2.4.0] - 2026-07-12#

Added#

• Added PALO Spotlight, a local ER-aware platform search across modules, guide decisions, artifacts, actors, controls, sources, stakeholder intents, and weighted relationships

• Added deterministic intent scoring, six-phase route tracing, relation traversal, operational starter questions, keyboard navigation, and reusable PALO_SPOTLIGHT.open(query) integration for a future conversational assistant

• Added a pinned Node toolchain for HTML, links, fragments, canonicals, sitemap, RSS, release metadata, and browser smoke validation

• Added an explicit public-file allowlist and deterministic root-to-dist build with SHA-256 exactness verification

• Added release-manifest.json, portable _headers, hosting guidance, and P0 lateral analysis

• Added PALO case-file and evidence-bundle v1 schemas with valid/invalid fixtures, migration guidance, source freshness fields, and incident/material-change gate triggers

• Added a local-only browser API for case create, validate, import, export, merge, board pack, and cross-module handoff while preserving unknown fields

• Added case import/resume and handoff to Stakeholder Onboarding and Assessment Path, including board-review Markdown output

• Added the embedded PALO-AM Simulator MVP with deterministic tiering, control/evidence/KPI recommendations, JSON/Markdown export, and case linkage

• Added critical P1 browser flows and P1 lateral analysis

• Added P2 structured control, KPI/KRI, decision-gate and source starter libraries with cross-reference validation

• Added six educational Case File worked examples, five operational templates, a PolicyWatcher signal contract, vendor-neutral connector patterns, and non-production policy-as-code examples

• Added P2 dependency/license inventory, publication hygiene, module contracts, deterministic publication coverage, and lateral analysis

• Added the public Platform Map with honest Implemented, Foundation, and Research states; P0-P3 lateral impact panels; stakeholder-intent navigation; and a synchronized table fallback

• Added navigation entities and weighted navigation relations to the Operationalization Explorer with a dedicated graph mode, destination properties, fallback coverage, and P3 browser tests

• Added Platform Map links to the homepage, shared navigation, Documentation Hub, sitemap, RSS, release manifest, README, and deterministic publication allowlist

• Added P3 lateral analysis and the required desktop, mobile, navigation-graph, and PALO-AM Simulator screenshots

• Added the v2.4.0 PolicyWatcher maintenance integration: a local Assessment Path signal receiver, pending-human-review Case File mapping, live Platform Map/Explorer destination, and exact import/boundary tests

Changed#

• Updated AJV to 8.20.0 after the final dependency audit, removing the reported moderate ReDoS advisory from the validation toolchain
• Unified all shared palo-v21.css and palo-v21.js cache keys on release v2.4.0
• Changed GitHub Pages CI to fail on validation errors and deploy only the generated dist artifact
• Upgraded Assessment Path exports to evidence-bundle v1 and connected onboarding, assessment, simulator, and board review through one portable local case
• Corrected the mobile Explorer navigation landing so it remains hash-free, opens on the visible graph, and focuses the Navigation mode control
• Raised Explorer mode, phase, relationship, inspector, semantic-search, and result targets to a 44px minimum
• Offset focused PALO-AM results below both fixed navigation bars and added keyboard-focusable, text-labelled mobile scroll affordances to the matrix and KPI/KRI registry
• Standardized PolicyWatcher links on https://www.policywatcher.online/, retained the local schema as the contract, and made additive signal fields forward-compatible without changing the 1.0.0 format

[2.3.2] - 2026-07-11#

Fixed#

• Unified the homepage Toolbox, AuditBench, and Poisoning Study entry bands around the shared PALO workspace pattern
• Restored cold-load and direct-file styling for the homepage lifecycle and proof metrics using local CSS
• Rebuilt the Companion App page on the shared shell with valid section navigation and operational lifecycle routes
• Linked the homepage five-phase proof point to the lifecycle and synchronized changed public files with dist/
• Restored the PALO-AM Controls navigation target and completed a 27-page internal and external link audit

Added#

• Stakeholder onboarding
  • Added a three-step, plain-language onboarding for decision, working context, and material signals
  • Added deterministic routing that returns exactly one primary action, one evidence artifact, and no more than two contextual PALO modules
  • Added personalized six-phase routes for executive, governance, product, engineering, public-sector, procurement, audit, assurance, and research perspectives
  • Added local JSON and Markdown route exports without accounts, analytics, or transmission of onboarding answers
  • Added resume, restart, change-answer, and explore-all paths using session-local storage with a graceful storage fallback
  • Connected the result to the weighted ER workflow and 3D knowledge graph with phase, module, camera, inspector, and URL synchronization

Changed#

• Made Stakeholder Onboarding the primary Start here, hero, command-center, and theory-to-practice entry from the homepage
• Updated the public release marker to v2.3.2 on the homepage, Explorer, and Community shell
• Added canonical, Open Graph, Twitter, robots, and WebApplication metadata for the Operationalization Explorer

Accessibility#

• Used native fieldsets, legends, radio buttons, and checkboxes with text progress, 44px controls, visible focus, preserved answers, intentional result focus, reduced motion, and a semantic WebGL fallback
• Verified responsive onboarding and Explorer behavior across desktop, tablet, and mobile layouts

[2.3.1] - 2026-07-11#

Added#

• Theory to Practice value proposition
  • Added a dedicated homepage operating loop that explains how PALO moves from a use-case question to classification, impact assessment, controls, KPI/KRI, evidence, and review
  • Connected the operating loop to the available PALO modules: Model Canvas, Human Agency Risk Map, Tech Trends Observatory, Risk Tiering, Framework Comparison, Regulatory Watch, FRIA, PALO-AM, Vibe Coding Governance, AuditBench, Poisoning Boomerang, KPI/KRI Generator, Assessment Path, Documentation Hub, and P.A.L.O. Toolbox
  • Made the practical outputs explicit: decision record, impact and control record, KPI/KRI register, and versioned JSON or Markdown evidence bundle

Changed#

• Pointed the primary Start here navigation to the theory-to-practice value proposition
• Updated the public release marker to v2.3.1 on the homepage and Community shell

[2.3.0] - 2026-07-11#

Changed#

• Workspace UI refresh
  • Replaced the crowded homepage toolbar with a compact command bar organized around Start here, Assess, Observe, Docs, Community, and Toolbox
  • Replaced the marketing-first hero with a workspace entry view and three operational routes
  • Added responsive mobile navigation with clear touch targets and no horizontal overflow
  • Simplified the homepage footer to policy, resource, contact, and RSS links
  • Removed the non-essential WCAG target badge and privacy-status card from the primary interface; accessibility and privacy remain available as policy pages
  • Applied the command bar to the Community page and aligned the public shell with the PolicyWatcher workspace language

Fixed#

• Prevented top navigation wrapping and overlapping at desktop widths
• Removed the legacy homepage mobile-menu script that could target missing navigation nodes after the shell refresh

[2.2.1] - 2026-07-11#

Added#

• PolicyWatcher ecosystem link
  • Added PolicyWatcher as a separate civic-tech monitoring companion for public privacy-policy and terms-of-service changes
  • Added links to the PolicyWatcher Observatory, timeline, confidence methodology, and Trust & Quality materials
  • Added an explicit integration boundary: PolicyWatcher signals can inform monitoring, but do not replace original sources, official requirements, human review, or legal advice

Changed#

• Added PolicyWatcher references to the homepage, Assessment Path, Regulatory Watch, Documentation Hub, Recognition page, README, and release feed

[2.2.0] - 2026-07-11#

Added#

• PALO Assessment Path
  • Added three guided entry points: classify the case, assess impacts, and build the evidence
  • Added routing from Risk Tiering to contextual FRIA, controls, KPI/KRI, PALO-AM, and AI Dev Governance when signals apply
  • Added local JSON and Markdown evidence bundle export with PALO version, route, evidence readiness, official sources, and disclaimer
• Regulatory Watch 2026
  • Added dated status cards for Article 4 AI literacy, GPAI milestones, Article 50 transparency, and current high-risk implementation guidance
  • Added official Commission and EUR-Lex source links and an explicit review-date disclaimer
• Documentation Hub
  • Added indexable web-native lifecycle documentation, interactive module directory, primary document links, source set, and contribution links
• Proof and Community media kit
  • Added public links for the Android and iOS/iPadOS companion apps, GitHub, framework image, PDF, and community page
• Unified v2.2 shell assets
  • Added responsive shared CSS and JavaScript for navigation, local downloads, current year labels, and Documentation Hub filtering

Changed#

• Added Start Here, Regulatory Watch, Assessment Path, Documentation Hub, and Proof & Community sections to the homepage
• Updated README, sitemap, RSS feed, and release history for v2.2.0
• Updated the roadmap to place PALO-AM Simulator after the guided evidence workflow

[2.1.0] - 2026-07-11#

Added#

• Recognition and source notes page
  • Added PALO_Recognition.html with public references, primary sources, link review date, and verification notes

Changed#

• Corrected the Risk Tiering transparency reference from Article 52 to Article 50
• Reframed FRIA copy as Article 27 readiness and pre-screening for in-scope deployers and use cases
• Added current Commission timeline notes for high-risk AI implementation and linked the official guidance
• Updated privacy, security, and accessibility statements with July 2026 review status
• Replaced unverified WCAG/GDPR compliance badges with review status and privacy information labels
• Completed SEO description, canonical, and Open Graph metadata for previously incomplete pages
• Added Recognition to the homepage footer, README module list, sitemap, RSS feed, and release metadata

[2.0.1] - 2026-06-22#

Changed#

• Updated README release table to use exact release dates and version numbers
• Reworked the README roadmap into a versioned roadmap from v2.0.1 through v3.0
• Aligned the PALO-AM page roadmap with the repository roadmap numbering
• Aligned homepage changelog entries with CHANGELOG.md
• Reordered RSS entries for chronological consistency and synchronized dist/feed.xml

[2.0.0] - 2026-06-22#

Added#

• PALO-AM: Agentic Governance Modality v2.0
  • New PALO extension for AI agents and agentic systems
  • Five operational governance objects: Agent Identity, Agent Authority, Agentic Risk Matrix, Agentic Control Layer, and Agentic Evidence Layer
  • Action-Space vs Autonomy Matrix for routing agentic systems into governance tiers
  • Five-phase PALO lifecycle overlay for delegated-action systems
  • KPI/KRI registry for agentic AI governance
  • Worked enterprise scenarios for procurement, workflow automation, and autonomous support agents
  • Alignment references for IMDA MGF v1.5, EU AI Act, ISO/IEC 42001, ISO/IEC 42005, and NIST AI RMF
• Public recognition and reference links
  • Added references to Reuters-distributed coverage, World AI Council mention, Rivista AI coverage, and Rivista Corporate Governance special issue index

Changed#

• Updated README to reflect the current 2026 framework state, mobile ecosystem, new modules, documentation, and roadmap
• Updated sitemap and RSS feed to include newer modules and mobile release information
• Corrected Google Play links to the active package ID: com.fabriziodegni.paloframework

[1.8.0] - 2026-05-15#

Added#

• Enterprise Security & Governance for AI-Assisted Software Development Environments
  • New PALO extension covering AI-assisted software delivery, rapid prototyping, vibe coding, and coding assistants
  • Three-layer governance model: Functional Intent, Controlled Environment, and Evidence & Assurance
  • Six decision gates for AI-assisted software development workflows
  • KPIs/KRIs for traceability, security, review coverage, and evidence quality
  • Dedicated module page: PALO_VibeCoding.html

Changed#

• Added AI Dev Governance links to homepage navigation and toolkit shortcuts
• Integrated AI-assisted development governance into the homepage changelog

[1.7.1] - 2026-04-08#

Added#

• P.A.L.O. Companion App - iOS/iPadOS Release
  • App Store listing added for the P.A.L.O. Framework Toolbox
  • Universal iPhone and iPad support
  • Evidence Vault, biometric protection, PDF report generation, and direct access to PALO web modules
  • App Store link: https://apps.apple.com/it/app/p-a-l-o-framework-toolbox/id6761771299

Changed#

• Updated companion app messaging from Android-only to cross-platform availability

[1.7.0] - 2026-03-25#

Added#

• The Poisoning Boomerang
  • New research module analyzing data poisoning as a cross-cutting AI governance risk
  • Coverage of six poisoning and anti-scraping tools: Miasma, Nepenthes, Nightshade, Glaze, Cloudflare AI Labyrinth, and AttackAI
  • Five detection strategies and lifecycle controls for data integrity governance
  • EU AI Act Article 10 and Article 15 governance notes
  • PALO lifecycle integration for data provenance, robustness, monitoring, and remediation
  • Dedicated module page: PALO_PoisoningStudy.html

Changed#

• Added Poisoning Study links to homepage navigation and related research sections
• Expanded the homepage changelog with March 2026 research updates

[1.6.0] - 2026-03-12#

Added#

• P.A.L.O. Companion App - Android Launch on Google Play
  • Official public release of the P.A.L.O. Framework Toolbox for Android devices
  • Available on the Google Play Store
  • Offline, privacy-first AI governance pre-screening toolkit
  • Full PALO Framework integration: FRIA, Model Canvas AI, Risk Tiering, KPI Generator, and more
  • Optimized for Android smartphones and tablets
  • Built with Flutter for a native mobile experience

Changed#

• Updated homepage to reflect official Android app availability
• Enhanced Companion App landing page with Google Play Store links

[1.5.0] - 2026-03-03#

Added#

• PALO Companion App v2.0 landing page
  • Dedicated page for the offline, privacy-first mobile app
  • 8 professional assessment tools presented for mobile use
  • Evidence Vault and mobile-first workflow messaging
• AuditBench Explorer
  • Interactive deep-dive into the AuditBench alignment auditing benchmark (Sheshadri et al., 2026)
  • 14 hidden AI behavior cards with PALO analysis
  • Category filtering: Manipulation, Deception, Bias, Safety, and Autonomy
  • Investigator Simulator with five auditing tools: Prefilling, User Persona, Scaffolded Probing, SAE Probe, and Reveal
  • PALO Alignment Self-Assessment with checklist and radar chart visualization
  • Export functionality for reports and assessment results

Changed#

• Updated homepage with Companion App showcase section
• Added Companion App links to hero buttons and toolkit navigation
• Updated homepage with AuditBench Explorer showcase section
• Added AuditBench link to hero buttons and mobile navigation
• Enhanced PALO toolkit with alignment auditing capabilities

[1.4.0] - 2026-01-13#

Added#

• Community & Open Collaboration Page
  • Dedicated page for open collaboration, contributors, and partners
  • Open collaboration roles for contributors and reviewers
  • Partners and partnerships section for academic, industry, and standards-body collaborations

Changed#

• Updated homepage navigation with Community link
• Enhanced footer with Community page link

[1.3.0] - 2026-01-04#

Added#

• 2026 Tech Trends Observatory
  • Analysis of predictions from McKinsey, BCG, Accenture, PwC, EY, KPMG, and Gartner
  • Theme Coverage Matrix
  • Blind Spots analysis revealing governance gaps
  • PALO Governance Gap Score
  • Cross-firm correlation insights
  • Direct links to public source reports
• Full GitHub repository structure with README, CONTRIBUTING, and documentation

Changed#

• Updated homepage with 2026 Tech Trends CTA button
• Enhanced January 2026 changelog entry

[1.2.0] - 2026-01-02#

Added#

• Human Agency Risk Map
  • Observatory tracking 18 activities humans are delegating to AI
  • Interactive cards with mitigation strategies
  • Psychological impact analysis
  • Category filtering: Cognitive, Behavioral, Social, and Professional
  • PALO Framework integration CTAs
• Human Agency Risk Map Italian version
  • Full Italian translation
  • Language toggle between EN and IT versions

Changed#

• Updated homepage hero section with Human Agency Risk Map button
• Added "2026 Spotlight" badge to homepage changelog

[1.1.0] - 2025-12-17#

Added#

• Risk Tiering Calculator
  • 3-step wizard for EU AI Act risk classification
  • Minimal, Limited, High, and Unacceptable risk tiers
  • Required documentation guidance per tier
  • Visual risk indicators
• KPI Generator
  • AI governance metrics aligned with PALO principles
  • Customizable KPI templates
  • Export functionality

Changed#

• Updated homepage with Risk Calculator and KPI Generator buttons
• Enhanced mobile navigation with Toolkit section
• Improved footer navigation

[1.0.0] - 2025-12-01#

Added#

• Initial release of PALO Framework website
• FRIA Assessment Tool
  • Fundamental Rights Impact Assessment wizard
• Model Canvas AI
  • Visual planning tool for AI projects
• Comparison Tool
  • Framework comparison utility
• Complete PALO Framework v1.0 PDF documentation
• Accessibility statement
• Privacy policy and security policy
• SEO optimization: sitemap, robots.txt, and meta tags
• RSS feed for updates

Security#

• Implemented security.txt in .well-known
• Added HTTPS-only policy

[0.9.0] - 2025-11-15#

Added#

• Beta version of PALO Framework tools
• Initial design system and color palette
• Responsive layout foundation