Plan-bound authorization
Execution is evaluated against the approved plan and scope before tools are permitted to mutate external systems.
GOVERNED VERIFIED EXECUTION
Separate action
from authority.
DSG ONE separates who proposes, proves, authorizes, executes, and observes AI-agent work. External side effects are bounded by an approved execution contract and evidence-backed verification.
Truth boundary: this site distinguishes implemented code paths and deployment evidence from external marketplace approval, certification, legal conclusions, or independent audit.
01 / PRODUCT
Govern before side effects. Verify before claims.
DSG ONE binds execution to approved intent, separates trust domains, and keeps proof, authorization, execution, and observation from silently collapsing into one agent decision.
Independent proof
Cinema and exact Z3 verification provide a separate proof boundary rather than allowing the candidate generator to validate itself.
Controlled execution
Approved work reaches APIs, MCP tools, browsers, and deployment adapters through bounded execution paths.
Evidence & replay
Execution evidence, hashes, replay state, and post-deploy observations remain inspectable after the action.
02 / CANONICAL ARCHITECTURE
Five trust domains. One authority boundary.
The system that creates a candidate does not approve it. The verifier does not execute it. Monitoring does not mutate production. Select a trust domain to inspect its responsibility.
PROPOSAL → PROOF → AUTHORITY → EXECUTION → OBSERVATION
ALLOW
Approved exact-plan work
Required authorization, plan alignment, capabilities, constraints, execution conditions and evidence support continuation.
WAITING_PERMISSION
Approved work, missing capability
The requested step remains inside the approved plan, but execution pauses until the required permission or capability is provisioned.
BLOCK
Out-of-plan or unsupported work
The system does not convert missing authority, failed constraints, incomplete proof, or out-of-plan actions into success.
QUBO / Ising candidate boundary
Formal encodings can be searched deterministically. Natural-language hints without the required encoding do not become executable proof.
VERIFIED_GLOBAL_OPTIMUM
Exact Z3 verification is a bounded mathematical claim about the encoded problem, not a certification or legal/compliance conclusion.
Remote Browser
Browser sessions are tied to approved work. Verifier-side observation is separated from executor-side mutation, and out-of-scope actions are expected to fail closed.
Baseline / Review / Rollback
Monitoring evidence can support acceptance of the next baseline, hold for review, or rollback recommendation while mutation authority remains separate.
03 / PRODUCT STATUS
What is live now, and what comes next.
The public product surface is a working MVP and architecture/evidence package. Production-provider binding and scaled commercial rollout remain explicit next-stage work.
LIVE NOW
Governance + verification foundation
Canonical authority model, independent Cinema/Z3 proof path, bounded execution design, public documentation, source evidence, and CI/E2E references.
INTEGRATION STAGE
Provider-bound deployment
Production adapters must be explicitly bound and authorized. An unbound provider remains blocked by design and is not presented as deployment success.
NOT CLAIMED
Certification or third-party approval
DSG ONE does not claim marketplace acceptance, independent audit, legal certification, or Google Cloud endorsement without separate external evidence.
03 / IMPLEMENTATION DEEP DIVE
The trust boundary is enforced in code, not only in the diagram.
This section maps the canonical architecture to concrete execution rules: candidate admission, deterministic search, independent proof, promotion authority, multi-lane joins, browser isolation, and post-deploy control.
01 · AGI SIMULATION
Better score ≠ safe candidate
Candidate admission requires composite-score improvement, real-efficiency improvement, and no protected-metric regression.
- P99 latency must not regress
- Error rate must not increase
- Throughput and cost efficiency must not fall
- Audit completeness must not fall
- Human override rate must not increase
- Previously passing SLA and zero-data-loss guarantees must not regress
ADMIT = scoreImproved ∧ efficiencyImproved ∧ noProtectedRegressiondsg-agi-simulation/src/candidate-admission.ts02 · DETERMINISTIC SEARCH
QUBO / Ising needs formal encoding
Integer-based deterministic search can evaluate formally encoded QUBO/Ising problems. Natural-language hints without the required encoding do not become executable solver input.
missing constraints.aimoEncoding → REVIEWShard progress and whole-space proof are separate states; one shard cannot claim a global optimum before the full search boundary is established.
searchComplete ≠ wholeSpaceSearcheddsg-agi-simulation/src/aimo/solver.ts03 · CINEMA FORMAL PROOF
Optimizer result is not the final proof
Cinema uses Z3 to obtain a candidate, applies deterministic tie-breaking, then creates a separate proof obligation asking whether any assignment exists with strictly lower energy.
verifier_objective < candidate_energy- UNSAT → VERIFIED_GLOBAL_OPTIMUM
- better witness found → COUNTEREXAMPLE_FOUND
- timeout → VERIFICATION_TIMEOUT
Only the UNSAT proof path yields verified=true for that bounded encoded problem.
04 · PROMOTION AUTHORITY
Simulation cannot promote itself
The promotion boundary requires the approved plan hash, baseline/candidate commit binding, allowed paths, plan alignment, passed constraints/tests/build, objective improvement, required evidence, and independent Cinema proof.
candidateAuthority = SIMULATION_ONLYpromotionAuthority = DSG_CONTROL_PLANEselfPromotionAllowed = falseRequired evidence includes commit, metric, test output, and build output. Candidate commit and Cinema proof binding must agree before promotion can be authorized.
promotion-gate.ts05 · MULTI-LANE EXECUTION
No evidence, no join
Parallel lanes move through an explicit state machine and must satisfy join requirements before the wave can complete.
QUEUED → RUNNING → READY_TO_JOIN → JOINED → COMPLETEDJoin checks require a joinable state, commit SHA, evidence, and no blockers. Missing evidence in one required lane can block the wave rather than silently merging partial success.
BLOCKED / FAILED remain explicit statesControl Plane multi-lane runtime06 · CONTROLLED BROWSER
Executor and verifier have different powers
Remote Browser sessions bind to an approved plan, approved step, and agent identity. Mutation-capable controllers are separated from verifier-side observation.
agent_verifier → extract / screenshot onlyPlaintext passwords, OTPs, API keys, passkeys, private keys, secrets, and MFA codes are not intended to travel through the model/evidence path. Delegation uses opaque references such as secret_ref and otp_ref.
Remote endpoints require HTTPS, reject localhost/private-address targets, validate DNS against SSRF-style private resolution, and seal session tokens with expiring AES-GCM protection.
api_v1/remote_browser.py07 · POST-DEPLOY TRUTH
Deploy success is not the end state
Production evidence feeds a monitoring decision while mutation authority remains separate.
PASS → COMMIT_NEXT_BASELINEREVIEW → HOLD_REVIEWBLOCK → EXECUTE_ROLLBACKMonitoring is observation-only; the canonical authority owns execution. An unbound provider or disabled production-deploy capability fails closed instead of being represented as deployment success.
PRODUCTION_TARGET_UNBOUND → BLOCKCurrent rollback adapter scope is bounded to declared providers such as AWS, GCLOUD, and DOCKER with HTTPS endpoint requirements.
post-deploy control / deployment adaptersTECHNOLOGY / CODE MAP
Implementation stack
Canonical governanceTypeScript / Node.jsState & promotion gatesTypeScriptDeterministic simulationTypeScriptQUBO / Ising searchTypeScript + BigIntFormal verificationPython + Z3 SMTVerification serviceFastAPICryptographic bindingSHA-256 canonical dataRemote-session securityAES-GCMAgent protocolMCP + REST/OpenAPICI orchestrationGitHub ActionsRuntime / deploymentAzure Cinema path + provider-neutral control pathEvidencehashes + commit binding + proof records
ARCHITECTURAL PRINCIPLE
Proposal, proof, authority, execution, and observation are separate trust domains.
AGI may propose. Cinema may prove. Monitoring may report. Executors may act. None of those roles silently grants itself canonical authority.
03 / ENGINEERING EVIDENCE
CI/CD and E2E are part of the product truth.
DSG ONE separates test-surface coverage, real E2E path coverage, deployment evidence, and instrumented source coverage instead of treating them as interchangeable claims.
44workflow files in current repository snapshot
40primary pytest modules
E2EAzure · Browserbase · Copilot CLI
NOT MEASUREDnumeric line / branch coverage
Core CI
API · Z3 · Marketplace · Revenue
Path-scoped workflows compile, run unit/contract/integration suites, enforce deterministic governance assertions, validate packages, and keep channel-specific failures visible.
Production E2E
Real external boundaries
Current E2E paths include isolated Azure Cinema→Z3 proof/replay, plan-bound Browserbase production execution, and authenticated Copilot CLI + DSG MCP flows.
Deployment evidence
Run-scoped, revision-scoped
Deployment receipts, GitHub Actions artifacts, client evidence, proof hashes, and production probes support specific runs. An old PASS is not evidence for a later revision.
COVERAGE TRUTH
Do not infer a percentage from test count.
Test-surface coverage DOCUMENTEDE2E path coverage DOCUMENTED + EXECUTABLEInstrumented line coverage NOT MEASUREDInstrumented branch coverage NOT MEASURED
04 / BUSINESS MODEL
Governed execution infrastructure for AI systems.
DSG ONE is delivered as B2B infrastructure for teams that need controlled external execution, deterministic verification, and evidence-backed outcomes.
Who pays
AI teams, SaaS operators & agent platforms
Organizations operating AI agents or automated workflows that need plan-bound authorization, controlled side effects, proof receipts, audit evidence, and replayable execution history.
How DSG ONE is delivered
API + CI + MCP + Marketplace channels
Current delivery surfaces include the Direct API, GitHub/CI integrations, agent and MCP integrations, marketplace packages, and enterprise deployment paths. Package availability does not imply external marketplace approval.
Commercial model
Free capped activation → governed usage → paid / enterprise
The operating path starts with capped free activation, then governed usage and commercial plans where the required billing and entitlement gates are verified. Enterprise deployment remains a separate delivery option.
What customers pay for
Governance, not model tokens alone
Value is concentrated in approved-plan binding, deterministic verification, controlled execution, evidence collection, replay, proof receipts, and production governance around AI actions.
Execution contract
Proposal → Proof → Authority → Execution
The system proposing work is separated from the verifier, canonical authority, executor, and observer. This reduces the risk of an agent authorizing or validating its own side effects.
Truth boundary
No proof, no commercial claim
Internal package readiness, CI success, or a deployment receipt is not represented as marketplace acceptance, third-party certification, or independent audit. External approvals remain separate evidence gates.
05 / USE THE PRODUCT + VERIFIED EVIDENCE
Do not trust “completed.” Verify what happened.
DSG ONE checks whether recorded AI-agent execution conforms to an approved plan, remains inside its declared scope, includes the required evidence, matches replayed outputs, and supports a verifiable proof receipt.
HOW TO USE IT
Run the production verification flow
- Open the 3D Evidence & Conformance Console.
- Confirm the runtime reports READY.
- If required, activate a free API key in the Console.
- Click Run full flow.
- Read ALLOW, REVIEW, or BLOCK and inspect the Proof Receipt.
WHERE TO SEE THE RESULT
Decision, evidence, replay, and receipt
The Console shows plan alignment, constraint verification, recorded execution, evidence coverage, replay match, the final decision, and machine-readable receipt hashes without requiring the user to search server logs.
PRODUCTION RUN EVIDENCE · 2026-08-27 06:41:45 UTC
Bounded sample flow verified end to end
Plan alignmentTRUEExact verificationVERIFIED_GLOBAL_OPTIMUMEvidence4 / 4 content verifiedReplay matchTRUEReceipt hashVERIFIED ON READAgent verdict acceptedFALSE
plan_01m10z7zgrgegen3pcxat2exec_01m10z8094pdy67kzrsptrproof_01m10z80rtpx6qstcwfmweProof hash
7baac0c87bbf64ce335cdc2f396d636fef79b022b13c32b4d163e3b904ae694eReceipt hashf788b7b4bdb6efac119922616cee8220c730267087369396ee2cde3e791dc50fFail-closed evidence: the first attempt without a valid API key stopped at plan creation with UNKNOWN_KEY and issued no decision or proof. After explicit free-key activation, the authorized flow completed and the receipt was read back with a matching recomputed hash.
TECHNOLOGY USED
Inspectable components across the verification boundary
Runtime APIPython · FastAPI · OpenAPI 3.1Exact verifierZ3 SMT · deterministic proof obligationExecution contractApproved plan · alignment · constraintsEvidence bindingSHA-256 canonical hashesAgent integrationMCP · JSON-RPC 2.0 · RESTProduction hostingAzure Container Apps · Azure static product surface
Proof boundary: this production receipt proves conformance for one bounded sample request using the raw plan, actions, and evidence submitted to DSG ONE. It does not certify every AI agent, an unobserved real-world outcome, legal compliance, SOC 2, ISO certification, or independent audit. The current record store reports memory / ephemeral and durable=false.
05 / COMPANY & CLOUD PLAN
Founder-led, working MVP, built for governed production adoption.
DSG ONE is a founder-led software startup developing governed execution infrastructure for AI-agent systems. This public profile separates what is available now from the next cloud-scale milestones.
Company profile
DSG ONE · dsg.pics
Founder-led B2B software product. Primary customer groups are AI teams, SaaS operators, and agent platforms that need controlled external execution and verifiable evidence.
Current product stage
Working MVP + public architecture
Available now: governed architecture, code-backed control boundaries, independent Z3 verification, CI/E2E documentation, public interactive architecture, and a production-domain information site.
Contact
Talk to the founder
Founder: Thanawat Suparongsuwan
Email: t.dealer01@dsg.pics
Website: dsg.pics
PLANNED GOOGLE CLOUD USE
Scale the verified execution path on Google Cloud.
This is the intended technical roadmap, not a claim of completed migration or Google endorsement.
Cloud Run / GKEContainerized verification services and governed worker orchestrationVertex AIModel access behind DSG ONE approval and evidence boundariesSecret ManagerOpaque credential references outside model and evidence payloadsCloud Logging + BigQueryRun-scoped evidence, monitoring, replay analysis, and operational reporting
05 / TEAM
Built from the control problem outward.
TS
Founder / Product Builder
Thanawat Suparongsuwan
Developing DSG ONE around deterministic governance, controlled agent execution, and evidence-first product behavior. Contact: t.dealer01@dsg.pics
PUBLIC PRODUCT PRINCIPLE
No proof, no claim.
DSG ONE treats proposal, proof, authority, execution, and observation as separate trust domains. Claims on this site are scoped to the implementation, evidence, and external status actually established.