integrity dashboard

---

title: integrity-dashboard created: 2026-07-07 updated: 2026-07-31 type: entity tags: [infrastructure, sdk] confidence: medium source_files:

• integrity-dashboard/src/App.tsx
• integrity-dashboard/src/config.ts
• integrity-dashboard/src/chain/bytecode.ts
• integrity-dashboard/src/components/ui/RegisterAgentModal.tsx
• integrity-dashboard/src/components/ui/VisualTopologyMap.tsx
• integrity-dashboard/src/services/oracle.ts
• integrity-dashboard/src/services/userapi.ts
• integrity-dashboard/src/components/shared/SeededDataBadge.tsx
• integrity-dashboard/demo/src/integrity_demo/main.py
• integrity-dashboard/demo/src/integrity_demo/agent_loop.py
• integrity-dashboard/demo/src/integrity_demo/userapi_bridge.py
• integrity-dashboard/scripts/seed_mock_data.py
• integrity-dashboard/.env.example
• scripts/sync_abis.py

---

This page was rewritten from scratch on 2026-07-12. The version of integrity-dashboard it previously described (a mature app with AgentListPage/ AgentDetailPage/MarketsPage/LeaderboardPage/WalletPage/ CapitalAllocationPage, real JWT auth, a Demo/Live data-mode toggle backed by MSW fixtures, demo/ as a real Python scenario engine, 129 vitest + 16 Playwright specs) no longer exists on disk. integrity-dashboard/src/ was independently rewritten into a new 16-page shell (all file mtimes ~2026-07-12 00:00-00:20, well after this page's prior "2026-07-11" content) by a process outside this wiki's tracking, confirmed by the user as intentional ("the new dashboard ui") rather than accidental data loss. None of the file paths this page previously cited under source_files exist anymore. Per this wiki's own "no aspirational content" rule, the old content has been replaced rather than patched — it described a build that is simply gone. If you need it for historical reference, it's recoverable from this file's prior version in your editor's/tool's history, not from anything in the current tree.

Table of contents

• Page consolidation (2026-07-16) — route count and names below are stale
• What actually exists now (as of 2026-07-12, this session's work — page
• Prereq wiring: benchmarks, VCs, custodial wallet, XNS, governance (2026-07-25)
• TriMetricWidget fixed 2026-07-15 — was the most severe fake-data surface in the dashboard
• What is NOT done yet

Page consolidation (2026-07-16) — route count and names below are stale

App.tsx now defines 11 routes, not 16: /, /landing, /identity, /contracts, /settings, /finance, /traces, /diagnostics, /health, /agents, /documents. Six pages named throughout the rest of this document no longer exist under those names — AuditPage, ChainOfThoughtPage, CompareTracesPage, ExchangePage, IntelligencePage, SdkTelemetryPage were deleted and consolidated into two new pages:

• TraceAnalyticsPage.tsx (/traces) — merges ChainOfThoughtPage's Historical Traces DAG view and CompareTracesPage's Gantt/compare view into one tabbed page (Live Stream, Historical Traces, Metrics, Time-Travel Debugger, Compare Traces). Metrics/Time-Travel Debugger are honestly SeededDataBadge-marked — no backend endpoint exists for either.
• SystemDiagnosticsPage.tsx (/diagnostics) — merges SdkTelemetryPage's real oracle telemetry/OTLP-volume view and AuditPage's disclosed-simulated audit-log feed into one tabbed page (SDK Telemetry, Audit Logs).

IntelligencePage's real radar-widget work (below) now lives on the Dashboard (/) directly. ExchangePage's real wagmi market-entry flow was folded into FinancePage's "A2A Markets & Escrow" tab (src/components/finance/MarketsEscrowPanel.tsx).

Re-verified for real after the consolidation, not just by reading code: brought up a full local anvil + docker-compose stack, generated a real 3-span nested OTel trace via the SDK's traceable() API against the live oracle, and confirmed TraceAnalyticsPage renders it as a real DAG with real span attributes over the live SSE stream, and SystemDiagnosticsPage's telemetry volume chart reflects the same real data — the demo→oracle→ frontend pipeline described in §10 of PRODUCTION_GAPS.md survived the rename intact. Also found and fixed one real bug this pass: FinancePage.tsx's live ITK balance was off by 10^18 (missing formatUnits(..., 18) on the raw wei-scale itk_balance string from GET /v1/agent/{id}/wallet), rendering a nonsense multi-quintillion-dollar portfolio value — see PRODUCTION_GAPS.md §7 for the full writeup.

What actually exists now (as of 2026-07-12, this session's work — page

names below predate the 2026-07-16 consolidation above)

React 19 + Vite + TypeScript, 16 routed pages (src/App.tsx). The rewritten shell shipped cosmetically complete but has now been fully resolved with real backend/chain wiring and a Notion-style drag-and-drop widget layout engine.

1. Notion-style block & widget dashboard. DashboardPage.tsx is a drag-and-drop grid (react-grid-layout). Toggle "Edit Layout" to drag widgets by their grab handle and resize them.

   • WidgetRegistry.tsx maps widget types (gauge, throughput, latency, nodes, events, radar, notes) to real components.
   • WidgetWrapper.tsx provides the common panel chrome — drag handle, delete/duplicate menu.
   • Layout and widget list persist to localStorage, with a "Reset" button to restore the defaults.

2. The build was broken; it isn't anymore. A missing return, two imports pointing at nothing, and unused-variable lint errors were blocking npm run build/npm run lint. Fixed, plus two shared components (src/shared/{Panel,StatusBadge}.tsx) that were imported but never written. Verified via Playwright: all 16 routes render with zero console errors.

3. Wallet/data infrastructure. wagmi + viem + @tanstack/react-query power a real ConnectWalletButton (injected wallets only, no RainbowKit). src/services/oracle.ts/userapi.ts are typed fetch clients checked field-for-field against spec/ais-api/v1/openapi.yaml. Every agent-attributable on-chain write goes through one shared helper, useSovereignAgentWrite, which wraps SovereignAgent.execute(target, 0, calldata) — the same pattern integrity_sdk/markets.py's _execute_via_agent uses. scripts/ sync_abis.py now also emits trimmed ABIs into src/abis/ and copies deployments.*.json into src/deployments/.

4. Real reads, verified against a live stack (real anvil + real cargo run oracle-backend + real registered test agents — not assumed from response shape alone):

   • AgentContext — was 3 hardcoded fake agents, now oracle.listAgents()
   • AgentsPage — real DID/tier/AIS/created_at
   • IntelligencePage — real leaderboard via oracle.getLeaderboard()
   • IdentityPage — real DID + ITK balance/open-positions via oracle.getWallet()
   • ExchangePage — real oracle.listMarkets() for the Active Markets list
   • FinancePage — real ITK balance, transaction history, and allowances
   • DashboardPage — real AIS distribution + high-integrity %, live-updated over SSE

   Confirmed end-to-end via Playwright: a real agent DID registered through integrity-cli appears verbatim in the rendered page, sourced from a captured real network response — not just "the build compiles."

5. Honest labeling for what's still simulated. A new src/shared/SeededDataBadge.tsx marks every page with no real backing yet: ChainOfThoughtPage, SdkTelemetryPage, CognitionPage, CompareTracesPage, DocumentsPage, HealthPage, most of ContractsPage (the compile/deploy flow is a labeled sandbox, not a real compiler), parts of ExchangePage (the order-book UI — IntegrityMarket is a pari-mutuel pool, so there's no honest order book to show), FinancePage's treasury stats, ActuarialHub, parts of DashboardPage's widget grid, and several SettingsPage panels.

6. Mock-data seeding, done right. scripts/seed_mock_data.py registers real test agents (and one real market) through integrity_sdk, the same way a real agent would register — not a "write fake rows into Postgres" script. Gated by MOCK=true. Runs outside the browser, since it needs the funder's private key. VITE_MOCK_MODE is read-only in the UI — it just reports whether the build is pointed at a seeded stack; it can't seed anything itself.

7. Wallet-interactive writes, for real (2026-07-12).

   • ExchangePage's "Place Order" does a real two-transaction approve + enterPosition flow, gated on the connected wallet matching the agent's on-chain controller.
   • HealthPage's Smart BAA registry reads real SmartBAAFactory .BAACreated event logs and wires real sign()/revoke() writes.
   • ClaimAgentModal was rewritten, not patched: the old "claim via signature challenge" premise had no on-chain support at all — it submitted a transaction using ERC-20 approve's selector as if it were a claim call, silently swallowing the failure. The new "Verify Agent Control" modal does something real: resolves the on-chain controller, compares it to the connected wallet, and has the user personal_sign a "prove you hold this key" confirmation.

   All three write paths verified against real anvil + oracle: a real enterPosition transaction moved a market's outcome_staked from 0 to a real non-zero value.

8. userapi auth, tests, and docs (2026-07-12). SettingsPage's login/register/API-key management verified end-to-end via Playwright against a real running integrity-userapi + isolated Postgres. New test infrastructure: 9 vitest unit tests (client request shapes, hook logic, context behavior) and 18 Playwright e2e tests (all 16 routes, zero console errors, a real-network-response assertion), both run against a real backend+chain. README.md rewritten from the default Vite scaffold into real project docs.

9. UI validation and agent selection (2026-07-14). Fixed strict TypeScript errors in DashboardPage.tsx/FinancePage.tsx that were blocking npm run build. Propagated selectedAgentId from AgentContext into DashboardPage/ChainOfThoughtPage, so the SSE hooks correctly filter live OTLP/AIS data to the active agent.

10. UI layout and legacy aesthetic integration (2026-07-15).

    • AgentsPage: Extracted the ClaimAgentModal and AgentOnboarding workflows from isolated modals into prominent inline dashboard cards above the agents grid, significantly improving discoverability of the onboarding pipeline.
    • IdentityPage: Fundamentally refactored to replicate the core layout and aesthetics of the legacy integrity-dashboard codebase. Replaced giant glassmorphism panels with a compact Agent Status Strip (DID, AIS, Tier, TEE), a dedicated Hero Bar, and a sub-navigation tab interface sorting components into Identity & DID, Enclave & Security, Economic Capacity, and Credentials.
    • XNS: Fully wired the XNSSearchService into the Identity page's UI layout.
    • ContractsPage: Expanded the IDE workstation with full features: multi-tab editor for files, interactive build/deploy panel, and a dynamic "Deployed Contracts" inspector that generates interactive ABI buttons via source code regex analysis.

11. On-chain primitives deployment and XNS registration (2026-07-18).

    • Decompiled and bundled compiled Foundry bytecodes for SovereignAgent and StateAnchor into bytecode.ts.
    • Built the RegisterAgentModal step-by-step deployment wizard, implementing real, mock-free transactions to deploy core contracts, authorize the oracle signer via SovereignAgent proxy execution, deploy proxy clones via the AgentPrimitivesFactory, and sync indexing with the Oracle database.
    • Upgraded the IdentityPage to support real on-chain reads for active primitive addresses, status indicators, and audit logging feeds.
    • Fully integrated write routing using SovereignAgent.execute to delegate register calls to the XibalbaNameService contract.
    • Re-implemented XNSSearchService to perform real read-only contract resolutions mapping handles to identities on-chain.
    • Consolidated DID, XNS management, claim/registration actions, history logs, security credentials, and the 7 primitive contracts onto a single tabless page layout, displaying the primitives in a structured status table.

Prereq wiring: benchmarks, VCs, custodial wallet, XNS, governance (2026-07-25)

Five backend prereqs were built + wired into the dashboard, each with an honest degradation path (never a mock):

• SDK benchmarks — oracle.getBenchmarks() (GET /v1/benchmarks) drives the model/provider stability panels from real network-wide telemetry aggregates.
• Verifiable Credentials — DIDExplorer now shows a real Ed25519-signed W3C AgentIntegrityCredential via oracle.getAgentVc() (GET /v1/agent/{id}/vc), alongside the real DID document.
• Custodial app-wallet — TokenWallet shows both the real on-chain balance/history and the userapi custodial $ITK ledger (userapi.getWallet()/walletTransfer() — see integrity-userapi); verified live (faucet 10000 → 9750 after a 250 transfer).
• XNS — XNSSearchService resolves human handles on-chain via oracle.resolveXns() (falling back to direct did:/0x lookups), and DIDExplorer's XNS_RESOLVE field reflects the real primary handle via oracle.getAgentHandle(). The register-handle write flow is deliberately deferred (CLI/SDK).
• Governance — GovernancePanel + GuardianPilot render live IntegrityGovernance proposals via oracle.getGovernanceProposals() (GET /v1/governance/proposals). Both degrade to their honest "Not Yet Live" state on the MissingSingleton (HTTP 400) the endpoint returns until the contract is deployed — never a live-but-empty list. The propose/vote write flow is deliberately deferred (CLI/SDK); an inline note says so. See Governance.

The MissingSingleton→400 contract was verified empirically against the live oracle (both XNS + governance endpoints returned 400 with a descriptive body; control route /v1/agents still 200). All frontend degradation branches key on that. Full build (tsc + vite) + 198 forge tests green.

TriMetricWidget fixed 2026-07-15 — was the most severe fake-data surface in the dashboard

The dashboard's "Tri-Metric Risk Analysis" panel badged itself "LIVE MODEL" while every number was fake: avgAis picked from 3 hardcoded magic constants, blockedRate/riskExposure literal strings with no computation, all 3 sparklines fabricated trend arrays — unlike every sibling in WidgetRegistry.tsx, which either fetched real data or disclosed via SeededDataBadge. Two of the three metrics are now real: AIS Deficit and BCC Intent Violation Rate fan out oracle.getAis() across every agent in the global AgentContext (same pattern the gauge widget already used) and average real ais/components.compliance. The third ("Smart BAA Value at Risk") stays honestly marked unavailable — no probability-of-leak model or network-wide staked-collateral index exists anywhere in this protocol (same conclusion independently reached for ActuarialHub, below). Two real runtime bugs were only caught by actually loading the dashboard against the live stack, not by tsc/lint/build: a KaTeX-remount render-storm that froze the browser tab (the 3 formula sub-components were redefined as new component types on every render — fixed by hoisting them to module scope) and a grid-height clipping bug once real content replaced the old sparkline decoration (DEFAULT_LAYOUTS bumped h: 2 → h: 3). Re-verified via live screenshots against a real registered agent, not just a passing test suite.

What is NOT done yet

• integrity-dashboard/demo/ (the Python scenario-engine directory make demo references) now exists on disk (as of 2026-07-15), and make demo itself now works (as of 2026-07-16 — it was documented in three places but had no actual Makefile target until this pass). It's a real scenario engine (integrity_demo package) that registers 4 persona agents and exercises a capital-allocation tool-call loop against a live chain; as of 2026-07-15 it also has an opt-in completion-callback bridge into integrity-userapi's demo_runs table (userapi_bridge.py — see integrity-userapi). Actually running it end-to-end against a real local chain + real oracle (2026-07-16) found and fixed several real bugs no amount of code reading had caught: every OTel span it ever exported was silently rejected by the oracle (missing integrity.agent.id, plus a structural one-shot-global-tracer issue given this engine manages 4 different agent identities in one process — fixed with real per-agent tracers, verified by querying the oracle's otel_spans table directly post-run); one LLM call failure used to crash the whole process with a raw traceback (now degrades gracefully, matching the registration loop's existing per-agent error handling); and there was no preflight check that the funder wallet (real Base Sepolia balance: ~0.001 ETH, 10x under one agent's default funding) could actually afford the run before spending gas on a doomed one.

  Additionally, during integration testing on 2026-07-16, we found and fixed two other crucial issues:

  • Added support for Google Gemini OpenAI compatibility in agent_loop.py to allow execution using GEMINI_API_KEY (using gemini-2.5-flash via the OpenAI client compatibility layer).
  • Integrated the SDK's NonceStore in the capital allocation tool call flow rather than a hardcoded nonce=1, preventing BCC_NONCE_REPLAY rejections in sequential runs.

  The full closed-loop data path has now been verified successfully locally: agent registration -> oracle registration -> OTel trace ingestion -> reputation sync -> BCC intercept gate authorization -> on-chain capital allocation. Full writeup: PRODUCTION_GAPS.md §9. Still genuinely missing: no UI trigger anywhere in this dashboard creates a demo_runs row or launches this CLI process — it remains an operator-run script against live Base Sepolia using a funder private key.

• ExchangePage's order-book/candlestick UI and DashboardPage's latency/node-fleet/security-event widgets remain honestly seeded (SeededDataBadge) — no on-chain order book exists for a pari-mutuel market, and no WSS/OTLP telemetry-streaming infra exists yet (see root PRODUCTION_GAPS.md). throughput is no longer in this list — it was wired to real oracle.getTelemetryVolume/getOtelVolume data earlier this session; the claim that it was still seeded was stale.

• HealthPage's "Propose BAA Contract" (creating a new BAA) is a disabled stub — it requires acting as the covered-entity persona, which this dashboard's single-connected-wallet model doesn't represent.

• Bundle size: a single ~1.9MB JS chunk (KaTeX, Monaco, and the full wagmi/ viem/@wagmi/core surface all ship in the main bundle) — noted by Vite's own build warning, not yet addressed with code-splitting.

• Architectural Gaps for Complete Data: The Dashboard's Hero Metrics (throughput/latency) require an OTel metrics sink in integrity-oracle to aggregate tracing data. Security event alerts (blocked interactions) need event-sourcing in the Oracle to capture bcc_middleware policy evaluations. Transaction history requires retroactive USD pricing via an external price feed integration in the Oracle.

Related: integrity-oracle, integrity-userapi (the demo/ scenario engine's completion-callback bridge target), AIS API spec (the field-shape source of truth oracle.ts was verified against), Agent Primitives (the SovereignAgent.execute write convention).