Release notes

PRAXIS now routes concrete public policy missions into capsule creation, wires pricing CTAs to billing, and records the authenticated golden-path proof contract for CI.

• Public and signed-in home surfaces now share the same mission-first entry, with sanctions, treaty, and stakeholder/crisis examples handed to Capsule Builder through capsulePrompt.
• Practitioner and Team pricing CTAs call the Polar checkout route for authenticated users and preserve signup-to-checkout intent for new users, while Institution remains a contact path.
• Settings exposes Manage billing through the portal route and public pricing labels sandbox/unconfigured billing posture honestly.
• Legal and Learn pages now state the current beta, US-data, Polar sub-processor, and premium-web posture instead of pending or aspirational claims.
• The CI smoke workflow can run the authenticated golden path when TEST_FIREBASE_EMAIL and TEST_FIREBASE_PASSWORD secrets are present.

PRAXIS now makes the public Guide reachable, labels current-versus-planned Learn content more clearly, and strengthens the runtime contract that attaches policy capsules to generation and durable agent work.

• Added public Guide, Help redirect, and Changelog redirect routes, then exposed Guide in public navigation, sitemap, and mobile marketing links.
• Expanded Learn with deep-link anchors for sourceability, claim-source alignment, role capsules, output quality, and thinking devices so starter capsule references land on concrete guidance.
• Ask and Guide turns now receive an explicit capsule-generation contract that treats the active capsule as governing context, separates reference capsules, labels evidence gaps, and recommends capsule-bound AgentRuns for durable work.
• AgentRun creation and Cloud Tasks dispatch now use idempotency and deterministic task names so duplicate enqueue attempts produce one durable run receipt instead of parallel work.

PRAXIS now keeps docs, capsule-depth audit expectations, and the credential-gated production canary aligned to the Home + Capsules cockpit contract.

• Reconciled repo and product docs around public Home plus signed-in Home + Capsules, with legacy routes treated as redirects or deep links into the shared cockpit.
• Repaired capsule-depth auditing so starter totals and category counts are derived from the live catalog instead of stale fixed counts.
• Extended the doctor canary contract to record public reachability, authenticated session, active capsule, Ask thread, AgentRun dispatch/completion, Memo Studio output, same-thread receipt, and cleanup posture.

Ask PRAXIS and runtime flows can now promote source-pack gaps into the active ContextCapsule repository as proposed review notes before any canonical source mutation.

• Added a capsule-service promotion path that dedupes source gaps and stores them as proposed capsule notes with source, run, and review metadata.
• Added an authenticated capsule action for source-gap promotion while preserving sample-mode mutation guards.
• Capsule trust posture now records proposed source gaps as review blockers and keeps the user in the existing capsule repository workflow.

PRAXIS now closes the May 19 static pre-ship blockers, preserves resumable agent run streams, and archives retired graph staging endpoints before production deployment.

• Agent run stream clients now build deterministic resume URLs from the last event id while keeping the server-side Last-Event-ID contract explicit.
• The authenticated app loading surface uses the shared Skeleton primitive and carries explicit empty, error, and retry state contracts for pre-ship coverage.
• The deep pre-ship checklist now has package scripts for workspace tabs, substrate writers, orchestrator writes, and ontology-system audits.
• Retired graph staging endpoints now return archived 410 responses that point callers to the graph proposals API instead of exposing stale staged mutations.
• Added named E2E smoke specs for briefing assembly, desk command center, and temporal/timeline display flows.

Every Ask PRAXIS turn now runs through triage, ContextCapsule, tool-router, and supervisor receipts, with deep/current research automatically queued into the durable agent runtime when configured.

• Added a deterministic conversation orchestrator that classifies intent, complexity, context health, tool routes, approval gates, and the selected answer/run path before streaming.
• Ask answer envelopes and saved thread turns now carry orchestration receipts alongside source previews, claim grounding, and Memo Studio document-pack candidates.
• Deep, current, or multi-source Ask turns can auto-create and dispatch capsule-bound Research Memo agent runs while simple turns remain fast direct answers with receipts.
• The Ask panel now shows compact triage, capsule, tool-router, and AgentRun chips without adding a new page or top-level navigation item.
• The account settings hub is canonical at /app/settings, and starter ContextCapsules now include source-pack, readiness, suggested-action, and preferred-agent workflow metadata.

Ask PRAXIS now carries richer source cards, source-aware claim grounding, actionable follow-ups, and citation-preserving document packs into Memo Studio.

• Ask source cards now normalize uploaded chunks, Knowledge Library corpus records, grounded fetches, source receipts, and document digests.
• Claim grounding now matches each answer claim against source-preview text instead of assigning the first available sources heuristically.
• Source previews show locators, source receipt links, and text-extraction fallback labels before a pack is sent to Memo Studio.
• Draft document packs preserve answer text, quote previews, source receipt ids, capsule/context provenance, gaps, and generated timestamps.

PRAXIS now exposes a public-safe runtime semaphore, keeps operational repair detail behind admin auth, and ships a live audit for production health, Cloud Run revisions, and retired worker posture.

• Public /api/health/runtime now redacts accidental URL, secret-name, stack, and service-account leaks while preserving only status, labels, latency, checkedAt, and the system debug path.
• Admin /api/admin/system-health keeps endpoint detail and repair guidance behind the existing admin gate.
• Added audit:runtime-health-live to verify praxis.tacitus.me health endpoints, frontend/backend Cloud Run ready revisions, and the absence of the retired gdelt-poller worker pool.
• Kept pipeline/ingest-gdelt as an intentional no-op compatibility guard instead of recreating or deploying the retired GDELT worker.

Focused /agents/[runId] workspaces now behave like live agent sessions: they refresh active runs, expose retry/cancel/approval controls, and send completed outputs straight back into the PRAXIS app.

• Focused agent workspaces poll active queued, running, and waiting-approval runs so run status and receipts update without leaving the canonical URL.
• Added owner-only retry, cancel, and approval decision routes, then wired them into the focused run panel inside the existing capsule Agents lane.
• Retry creates a child run for failed, cancelled, or expired sessions and re-dispatches queued/running runs through the existing Cloud Tasks boundary.
• Research Memo and Output Formatter workers now return Memo Studio output links instead of generic app links.

Capsule-bound agent runs now open through canonical /agents/[runId] workspaces on the same PRAXIS domain, so a queued run becomes a durable conversation-style workspace instead of only a recent-run row.

• Added the private /agents/[runId] route that loads the shared PRAXIS cockpit with the selected run, capsule, events, tool calls, approvals, and output link focused.
• Run creation now navigates directly to the canonical agent workspace after queueing and dispatch, while /app/agents/[runId] redirects for compatibility.
• Login and middleware preserve private agent workspace redirects and keep onboarding enforcement aligned with /app.
• Expanded the agent-process contract audit so future cleanup must preserve the deep-link route, middleware matcher, and focused-run workbench wiring.

The one-page cockpit now treats Research Memo and Output Formatter as capsule-bound agents with durable run metadata, visible run status, and source-aware output receipts.

• Added a capsule Agents lane with registry cards, readiness, risk, capsule slots, allowed tools, and recent run status.
• The cockpit can now create capsule-bound agent runs through the agent run ledger and dispatch route instead of treating agents only as chat shortcuts.
• Added a live Output Formatter agent that creates export-ready Memo Studio drafts with citation and gap-preservation receipts.
• Tightened agent run DTOs so the UI receives objective, update time, readiness, risk, output destination, and policy-agent labels without exposing internal cost fields.

PRAXIS now has a user-scoped Knowledge Library backend that organizes uploaded documents into repositories, capsule links, and AI-generated digests for better Ask PRAXIS retrieval.

• Added owner-scoped knowledge repositories and knowledge-document records over the existing document ingestion pipeline.
• Document uploads can now carry repository, category, tag, and capsule association metadata while preserving the canonical upload/chunk/embed path.
• Added Vertex/Gemini-backed structured document digestion for summaries, claims, dates, causal links, human-friction signals, and follow-up research questions.
• Ask PRAXIS corpus retrieval now includes Knowledge Library digests alongside uploaded document chunks.

The one-page cockpit now exposes capsule building, repository review, message actions, research actions, and text-span binding without adding new app pages.

• Added cockpit sheets/dialogs for the capsule builder, repository, capsule inspector, and capsule-aware research actions.
• Assistant messages now expose copy, retry, save-to-capsule, memo, and source-check actions.
• Selected text in a PRAXIS answer can now be attached to Sourceability, Temporality, Causality, Human Friction, Rules, or Open Questions.
• Added capsule span persistence under the owner-scoped capsule API and Firestore rules.

The one-app model now keeps Learn, Pricing, Release notes, and the external TACITUS link working as the only public support surfaces.

• Restored /learn as a concise public explanation of the ContextCapsule workflow.
• Changed /tacitus into an external redirect to tacitus.me.
• Kept /app and legacy app URLs consolidated into the single signed-in cockpit.

PRAXIS now has explicit capsule templates for time, causality, decision windows, and human friction in complex policy work.

• Added Temporal and causal map, Human friction map, and Decision clock capsule templates.
• Policy-team categories now seed these structures where timing, blockers, incentives, and causal uncertainty matter most.
• Added a combination recipe that turns temporal and friction capsules into a decision-window brief.
• Expanded the capsule capability map with explicit Causality and Human Friction areas.

The capsule manager now understands policy-team categories, creates category-seeded capsules, and suggests the right PRAXIS tools from capsule metadata.

• Added a canonical policy-team taxonomy for economic, diplomatic, security, humanitarian, rights/legal, governance, climate/development, information/cyber, crisis/ops, and research/strategy work.
• New capsules can be created from policy-team presets that save category tags, output goals, and recommended tool cores through the existing capsule contract.
• Capsule details now show policy-team assignment and capsule-aware suggested tools with one-click attach.
• Core library search now indexes capsule category tags and attached tool/product ids so user capsules are easier to find and reuse.

The one-page cockpit now has stronger visual parity, clearer policy-team actions, Markdown answer rendering, and inline capsule categorization.

• Public and signed-in cockpit surfaces keep the same composer, capsule bar, policy-team actions, and capsule manager shape.
• Ask PRAXIS responses render as structured Markdown with headings, lists, and tables instead of plain text.
• Capsule cards use inline category entry instead of browser prompts, keeping add, select, improve, archive, upload, and refresh inside the app page.
• The Ask prompt contract now asks PRAXIS to format answers for policy teams with findings, source posture, gaps, and next actions when relevant.

PRAXIS now treats the capsule library as the only substantive app page beside public pricing and release notes.

• The signed-in cockpit now loads real user capsules, supports capsule search and category filters, and shows stored ids, versions, sources, documents, tools, and outputs.
• Capsule cards can select, ask, improve, categorize, refresh, upload into, or archive capsules without leaving /app.
• Public navigation is reduced to Home, Capsules, Pricing, Release notes, Learn, and the external TACITUS link.
• The app sidebar and command palette now point users back to the capsule manager instead of separate Memo, Canvas, Library, Ask, or Advanced pages.

PRAXIS now enters one authenticated app surface at /app while Learn, Pricing, Release notes, and TACITUS remain the only public navigation links.

• Collapsed the authenticated dashboard, assistant, context, canvas, writing, tools, situations, profile, and settings pages into the single /app cockpit.
• Kept compatibility redirects for old app URLs so shared links land on the one-page PRAXIS workspace.
• Added public Pricing and Release notes pages beside Learn, with TACITUS routed to tacitus.me.
• Document uploads now carry the active ContextCapsule id from the cockpit so the library intake binds files to the selected capsule.

The public surface is narrower: legacy demo pages and old tour UI are removed while compatibility redirects stay in proxy.

• Removed the retired demo route surface so users enter PRAXIS through authenticated, user-owned workspaces.
• Deleted the unused guided-tour component stack and its stale unit test.
• Replaced the dashboard command-center smoke with checks for the shared Tasklet-style PRAXIS capsule cockpit.
• The demo fixture verifier now validates seed data only, not a public demo route implementation.

The model selector now tracks the current Google Cloud Gemini 3 guidance for deep policy reasoning while preserving Gemini 2.5 fallbacks.

• Deep memo and long-context routes now default to Gemini 3.1 Pro Preview in the central model matrix.
• The cockpit exposes Gemini 3.5 Flash, Gemini 3.1 Pro Preview, and Gemini 2.5 fallback choices.
• The runtime catalog test now guards the expanded frontier model selector.

The shared cockpit bar can now discuss directly from the selected ContextCapsule stack instead of depending on the floating assistant shell.

• Chat with PRAXIS now calls the Ask PRAXIS endpoint directly from the cockpit when signed in.
• Each run sends the concatenated capsule stack, selected tools, output template, user lens, and selected Gemini model as context.
• The cockpit renders an inline discussion panel so the dashboard remains visually aligned with the public Home while gaining signed-in functionality.
• Public Home keeps the same surface and routes attempted runs through login handoff.

The Ask cockpit now treats Context Capsules as a real editable stack and exposes the current Gemini 3 plus stable Gemini 2.5 Vertex choices.

• Capsules in the rail above the Ask bar can now be changed, removed, searched, or replaced from the capsule library.
• The same popup includes a tool and core library so policy tools can be toggled without leaving the discussion surface.
• The composer supports Enter to send, Shift+Enter for a new line, and Shift+Up to restore the last queued message for editing.
• The model selector now includes Gemini 3.5 Flash, Gemini 3.1 Pro Preview, and Gemini 2.5 Pro/Flash/Flash-Lite Vertex options.

The signed-in dashboard now matches the public cockpit more closely, with one visible capsule-concatenation control and a quieter context management page.

• Removed the dashboard floating Ask launcher so the signed-in Home keeps the same bar-first cockpit as public Home.
• Moved capsule recipe actions into the single Context Capsule concatenation rail above the Ask bar.
• Removed duplicate capsule composition UI from Context management; that page now focuses on context health, portable .core packages, sharing posture, and governance.
• Updated cockpit audits to guard the single concatenation affordance.

The public and signed-in cockpit now stay visually aligned, while Learn carries the deeper product explanation.

• Public Home and signed-in Home share the same capsule cockpit surface.
• Internal provider/runtime/token labels were removed from the main cockpit.
• The cockpit uses a quieter policy-grid background, rounder surfaces, and clearer Context Capsule language.
• /learn now has a sidebar with sections for capsule cores, ten policy paths, tools, outputs, connectors, and planned scaffolding.

PRAXIS now has a typed output-product catalog behind the same simple cockpit.

• Added schema-versioned OutputCapsule templates for daily reads, country packs, media, risk, humanitarian, diplomatic, memo, and talking-point outputs.
• Added deterministic planning, quality blockers, hidden-tool mappings, and country-pack composition.
• The cockpit shows a compact Produce selector with missing cores and suggested producer steps.
• Persistence from live agent/tool runs remains the next runtime wave.

The PRAXIS cockpit became capsule-first, with core health and portable package support.

• Public and authenticated Home use the same Tasklet-style PRAXIS cockpit.
• Context Capsule management gained trust/health posture, .core export/import, clone/share/download foundations, and community metadata.
• Ask PRAXIS can preserve capsule conversation context and attach document/source intent.
• The visible app stays simple while advanced capsule management remains in the context surfaces.

Policy agents moved behind a compact cockpit instead of a default marketplace grid.

• Introduced the policy-agent registry for research memo, sourceability review, red-team review, daily brief, and handover modes.
• Research Memo Agent creates a Memo Studio draft with source and gap receipts through the worker boundary.
• Policy agents stay recommended or drawer-based instead of overwhelming the default surface.
• Run destinations are explicit: Memo Studio, Canvas, Library, and Handover.

Typed Genkit flows became available as safe capsule-aware checks.

• Added Sourceability Check for supported, weak, unsupported, and date-sensitive claims against provided snippets.
• Added Context Capsule path classification against the canonical policy path registry.
• Fallback behavior remains deterministic when provider execution is unavailable.
• The routes classify and review only; they do not mutate capsules, send externally, or claim broader automation.

The public surface was narrowed around Learn, TACITUS, Release notes, and the capsule cockpit.

• Added public Learn, TACITUS, and Release notes surfaces.
• /tour redirects back to Home.
• Tasklet MVP and public-surface audits guard the simplified public entry.
• Historical build prompt packs moved under docs/_archive/build with active pointer stubs preserved.