Migration, Convergence, Governance, Proof, DeedSonar, and the Architecture of AI That Executes
"AI that can act without becoming ungovernable." Source claim, dossier cover, 14 Sep 2026.
Built is not merged. Merged is not deployed. Deployed is not running. Running is not healthy. Healthy is not authorized. Authorized is not revenue-proven. Source claim.
Full founder-transport-to-outcome autonomy remains unproven. Source claim, status date 14 Sep 2026.
Current live provider and runtime evidence first, then exact Git, deploy and database state, sealed receipts and independent courts, Vault and current operations, current mission checkpoint, historical architecture, old handoffs, then model memory. Source claim.
Primary evidence in the source: GitHub source and PRs, sealed court records, Notion canonical topology, Library dossiers and artifacts.
Hub: Neuruh Hub | Sibling builds on the hub include the DeedSonar and LandOS surfaces, the Recipe system demo, and the Public Commons chain demo. This page links out; it fetches nothing from them.
Neuruh is being built as an execution system, not a conversational layer. Its central problem is moving a real objective through evidence, bounded decision, authority, execution, independent adjudication, durable proof, and economic outcome without a founder manually carrying state between workers. Source claim.
What changed: a canonical topology distinguishes constitution, current runtime state, source custody, historical lineage, and commercial projection; Workers, Switchboard, Sentinel Court, Git Custodian, Promotion Executor, Founder Runtime, Governance, MOTHER, receipts, and the Factory are treated as bounded owners rather than one undifferentiated "swarm"; the Workers to Sentinel seam was upgraded from conceptual and fixture evidence to real cross-process proof; proof now carries exact identity and authority fields; DeedSonar became the economic proving ground. Source claim.
| Observed pattern | Why it mattered |
|---|---|
| Fragmented repos and long-lived branches | Same concept in multiple implementations, only one wired to a live runtime |
| Multiple architecture generations | Old autonomy proofs matter as lineage while runtime paths are no longer current |
| Staged autonomous components | Switchboard, Sentinel, Git Custodian and Founder Runtime could each be "built" without forming a continuous chain |
| Different pins and supervised clones | A test could court a sibling checkout at the wrong SHA and still describe itself as proving the intended contract |
| Fixture-heavy green suites | A test double could satisfy the consumer schema while the real producer emitted a different shape |
| Operational state outside Git | Provider state, runtime health, launchd configuration, credentials and local custody could lag source code |
| Founder as courier | When a seam stopped, the founder manually transported the next state, effectively becoming the event bus |
Several failures were not missing code. They were mismatches between what a system said it consumed and what it actually received. Examples: stale production pins; wrong source repository identity (the board named one repo, the sealed receipt named the candidate's actual repo, and the independent court correctly refused the mismatch); flattened events; obsolete receipt schemas; Sentinel originally recording 8 of the 16 identity fields; live trust configuration lag; expired ratification; Founder Runtime latch behavior; executor binding gaps; live-vs-CI discrepancies. Source claim.
Honest gap in the source itself: the source jumps from Part I to Part III. No Part II heading appears in the 62-page PDF as extracted. This demo preserves that jump and does not invent a Part II.
| Layer | Question answered | Representative owner |
|---|---|---|
| TARGET | What outcome is being pursued? | Founder Runtime / mission compiler |
| SIGNAL | What changed or suggests an opportunity? | Source adapters / Switchboard / domain workers |
| EVIDENCE | What is actually known, and from which source? | Receipts / source hashes / evidence stores |
| DECISION | What should be done next, and why? | Bounded recipe / specialist / planner |
| AUTHORITY | Is this exact operation allowed under this exact lineage? | Governance / G1 / G2 / founder approval |
| EXECUTION | Which bounded worker performs it? | Workers / dispatch / organ-specific executors |
| RECEIPT | What exact transition occurred? | ExecutionLedger / sealed receipt |
| OUTCOME | Did it work economically and operationally? | OutcomeRecord / CRM / transaction / learning |
The design objective is not perpetual autonomous activity. It is durable progress. Source claim.
6. Governance as an execution primitive. Governance is part of the runtime dependency graph. The executor must be able to ask whether a specific action, source, risk tier and authority lineage are currently qualified.
Mission -> authority facts -> Governance evaluate -> ALLOW / founder_required / DENY
Execution only if the bound predicate remains true. An approval record can exist yet remain unusable if it is not bound to the exact published scope and source SHA. Source claim.
7. Sentinel Court. Independent adjudication membrane. Separate process boundary and repository boundary, with Workers forbidden from importing the organ. The builder sends a production-shaped request over loopback HTTP; the organ seals a record naming the request it actually answered.
| Property | Design consequence |
|---|---|
| Separate process | Builder and judge are not one crash domain |
| Separate repo | Workers cannot quietly patch the court's logic to make its own evidence pass |
| Identity echo | Court record names exact receipt, plan, task, correlation, causation, source head, source repo and authority |
| Fail closed | Unavailable court is BLOCKED / OPEN, not a deadlettered fake failure |
| Honest verdicts | A candidate can earn FAIL because its environment is invalid; that FAIL can still be correctly ingested |
8. Workers. Execution substrate holding workflow state, JobQueue, ExecutionLedger, dispatch logic, and bounded consumer contracts. Workers does not become the universal owner; authority, adjudication, custody and deployment stay with their organs.
9. Founder Runtime. The boundary between a proposed mission and a consequential operation. Sep 14 evidence reproduces a real defect: a gate ALLOW was recorded before the authorization request existed, later ratification created the request, but the decision latch kept replaying an artifact-less approval. The proposed late-binding repair is a draft and test lane, not a merged fix. Source claim.
10. Switchboard / Dispatch. A wrapper around durable ledgers and queues, not a second event bus. It observes sealed receipts, creates deterministic jobs, routes court requests, and, once explicitly enabled, binds existing Git Custodian and Promotion Executor organs behind environment switches.
11. Receipt and hash-chain model. Receipts establish provenance and causation across process boundaries. Critical fields: mission ID, job ID, receipt reference, receipt SHA, candidate SHA, plan ID, task ID, correlation ID, causation ID, source head, source repository, authority reference, and the court's execution mode.
receipt.sealed -> court.requested -> independent_sentinel_court -> court record -> identity binding -> court.failed | court.passed -> custody / promotion only when separately authorized
| Step | Observed state | Interpretation |
|---|---|---|
| Initial mutation attempt | MUTATION_NOT_RATIFIED | Governance correctly refused because the prior ratification was expired |
| Founder renewal | G1 at 2026-09-14 11:14Z | The external authority condition changed |
| Governance response | founder_required | The system now had a valid place for a founder decision |
| Request publication | scope_key / source SHA published | The exact work was now identified for approval |
| Latch defect | Artifact-less prior ALLOW replayed | The founder's earlier decision could not be rebound to the newly published scope |
| Pin drift | 7b1431b4 to b109810a | Even a late artifact would need to bind to the current qualified source SHA |
| Current result | SOURCE_SHA_MISMATCH / AWAITING_APPROVAL | The system stops rather than fabricating a continuation |
| PR | Purpose | Head (short) | Merge SHA | State |
|---|---|---|---|---|
| Sentinel Court #5 | Identity echo | 6a4f2b01... | 3f2cc79b6f6c80c5f96e845a4c240b4e9000405b | Merged |
| Workers #13 | HTTP adjudicator injection | 50d36e02... | e0c58ec15aa9298eb57170710b3a9cc7a0ef8e28 | Merged |
| Workers #15 | Founder Runtime execution-outcome sender | 38e8b3f2... | d0f5e955f826dc1e30397d1b61348495bf7fba6c | Merged |
| Workers #16 | Custodian + promotion executor composition | 2b2ec693... | 665a1e192d4ee4c8f5a8644a2f3f1149f5f556c9 | Merged |
| SOPS #14 | Organ activation / pins / scheduled dispatch | c6af3947... | a0d01ae7d3e54057ffcc28fc8bbe61c9dd2785c8 | Merged |
Open, not merged, at the status date: Sentinel Court #6 (current 200/200 LIVE_PROVEN harness), Workers #17 (source_repo identity fix + real seam test), Recipe Engine #27 (live Founder Runtime late-binding defect tests). Source claim.
OK in the current Sentinel Court #6 harness. Source claim.
Identity fields echoed; identity_bound true; identity_mismatches empty. Source claim.
Ingested event type. FAIL reason: CANONICAL_REPO_LOCAL_REFUSED for the offline canonical remote. Source claim.
SENTINEL_MODEL_EXECUTION = BLOCKED. No model execution is being implied. Source claim.
Mission ID: msn_2d6e71c5b3e8494494a687288bc7f2ea. Intended journey: created, A2 classified, plan proposed, MUTATION_NOT_RATIFIED, G1, founder_required, source qualification, execution. Current state: AWAITING_APPROVAL, because the approval latch cannot late-bind the missing Governance artifact and the current Founder Runtime source is not yet fully qualified. That is a real blocker, not a storytelling inconvenience. Source claim.
Use the sample loader below to load this mission into the builder and walkthrough.
DeedSonar is the first bounded commercial world where the execution model has something to prove economically: source material is messy, opportunity truth is multi-dimensional, decisions have financial consequences, and the customer ultimately cares whether intelligence becomes action and revenue. Source claim.
LandOS is the intelligence substrate underneath the product. Source adapters are replaceable; source custody is separate from inference. County exports, tax sources, foreclosure data, property records and other public datasets become versioned evidence inputs instead of opaque scraping artifacts. Source claim.
Recipe Engine is the Signal to Opportunity kernel. A recipe carries identity, version, jurisdiction, universe, signals, required evidence, amplifiers, negative evidence, disqualifiers, timing modifiers, economic gates, source dependencies, coverage, missing evidence, next-best-evidence, routing, output class, status and verification state. Source claim.
Founder-reported current state (source labels it this way, not independently verified in the dossier): daily recipe execution exit 0; engine 14/14; 3 upserts; /leads/top HTTP 200; rank-1 candidate be05bf71. Source claim, founder-reported.
Tape court, Sep 13: seven production tapes, 7/7 Sentinel PASS, 0 model calls, $0 spend, external false, production_mutated false, and a deliberate STALL state where the contact ladder still requires a human step. That is meaningful evidence of an economic control surface, not proof of completed revenue. Source claim.
| Stage | Meaning | Bounded-world output |
|---|---|---|
| State / county | Acquire authoritative source data | Versioned source rows + source hash |
| Candidates | Turn records into opportunity candidates | Candidate identity |
| Classification | Assign truthful status | Qualified / research / rejected / etc. |
| Map-ready | Make the opportunity spatially usable | Geocodable opportunity object |
| Evidence | Assemble proof | Evidence graph + provenance |
| Phone enrichment | Resolve contact evidence | Phone evidence, not automatic callability |
| Callable | Apply DNC / consent / policy gates | Eligible contact action |
| Acquisition | Move into operator action | Lead / offer / negotiation |
| Contract | Formalize transaction | Contract state |
| Disposition | Route opportunity | Buyer / partner / exit |
| Transaction | Capture economic outcome | Revenue / close / loss |
| Learning | Feed outcomes backward | Calibration / next recipe improvement |
| Typical agent system | Neuruh execution model |
|---|---|
| Prompt, model, tool, result | Target, evidence, bounded decision, authority court, execution, independent adjudication, receipt, continuation, outcome |
| Conversation is the state | Mission state is durable outside the model |
| Tool success implies task progress | State transition requires evidence and ownership |
| One agent can judge its own work | Judge is a separate organ / process |
| Retries are often implicit | Retries are bounded and tied to idempotency keys / exact evidence |
| Deployment is a tooling concern | Promotion is a separately authorized state transition |
| Failure is often a final response | Failure becomes typed, resumable operational state |
System-level refusal. The model is not trusted to adjudicate its own permission, provenance or identity. Wrong identity stops. Wrong SHA stops. Expired authority stops. Missing evidence stops. Missing receipt stops. Missing custody stops. Unavailable adjudicator is a visible blocked state, not fake failure. Ambiguous professional or legal boundary is routed to the proper authority, not invented. Source claim.
QUALIFICATION = f(runtime_identity, operation_class, source_sha, authority_ref, evidence, policy)
EXECUTE only when QUALIFICATION == TRUE
Semantic confidence and authority identity are different things. A high-quality model output can still be unauthorized. A low-confidence result can still be correctly stored as low-confidence evidence. Source claim. Try the qualification checker in the calculators section.
| Offer | Customer buys | Execution boundary |
|---|---|---|
| $100 Property Opportunity Scan | Fast source-backed opportunity read | Read-only intelligence; human decision |
| $499 Property Intelligence Dossier | Deep decision case with evidence, risk and next action | Human / policy controlled fulfillment |
| Custom deep acquisition case | Complex land, ownership or transaction intelligence | Specialist / legal / professional handoffs where required |
| Data Rescue / Opportunity Recovery | Turn messy or stale data into ranked opportunities | Evidence and transformation, not blind contact |
| Managed Opportunity Intelligence | Recurring hunt + investigate + prioritize + decision support | Bounded recurring service with explicit approval points |
| Custom intelligence-system build | Replicable intelligence pipeline for a buyer or partner | White-label or API surface over bounded world |
Note: the source dossier (14 Sep 2026) lists Scan $100 and Dossier $499. Those are source prices at that date, not a current live price list. Verify before quoting.
The customer should receive a decision document that makes a real opportunity legible: property / owner / parcel identity and source lineage; opportunity thesis and alternative thesis; observed evidence with provenance; ownership / estate / tax / title / zoning / hazard context where applicable; value / scenario ranges and explicit unknowns; risk and kill conditions; next-best-evidence and next-best action; decision receipt; professional-review triggers when the case crosses a licensed boundary. Source claim.
The expansion ladder is not "more pages." It is deeper evidence resolution and greater decision leverage: underwriting, flood and hazard analysis, ownership / estate investigation, comparable land intelligence, acquisition prioritization, and specialist routing, while regulated conclusions stay with the proper authority class. Source claim.
DeedSonar is the first bounded world. The architecture is designed so the organism can operate additional worlds without exposing the internal control plane: one organism, many bounded interfaces. Candidate future worlds can include business acquisition, resale / asset intelligence, field intelligence, or other economic decision domains, added by proving the shared contracts rather than cloning the core. Source claim.
The public surface should act as a witness. It should not require visitors to believe a giant claim about "autonomous agents." The strongest story is concrete: here is a target, here is the evidence, here is the decision, here is the authority, here is what executed, here is the receipt, here is the outcome. Source claim.
| Public proof layer | What the visitor should see |
|---|---|
| Target | A real but safe example objective |
| Evidence | Source-backed inputs and clear uncertainty |
| Decision | The bounded recommendation and reasoning trail |
| Authority | What class of approval was required |
| Execution | The exact action that occurred |
| Receipt | A machine-verifiable proof object |
| Outcome | What changed, or why the system stopped |
| Learning | How the result changes future prioritization |
| Gate | Current state as of 14 Sep 2026 | Why it matters |
|---|---|---|
| Founder Runtime lineage qualification | Live current work still not fully qualified at b109810a; Governance expected source is stale in the reproduced defect | Consequential mutation cannot safely proceed until source identity and authority bind |
| Founder Runtime gate-latch defect | Recipe Engine #27 open + draft; 3 of 4 tests RED on main; proposed late-binding patch is not applied | Without it, a valid founder decision can exist but remain unusable |
| G2 scope-once authority | Still a distinct gate | A one-time founder decision must not silently become reusable class authority |
| Custody / promotion binding | Composition exists but remains environment / authority gated | The ability to open a PR or merge is not the same as permission to do so in production |
| Full founder transport seam | Not yet proven end-to-end | Founder Runtime, dispatch, worker, receipt, Sentinel, ingest, custody, promotion, continuation must still be observed as one machine chain |
| Autonomy proof | Not yet claimed | The machine has proven important seams, not full autonomous economic continuity |
All six gates above were open at the status date. No gate in section 29 is marked closed in the source. Source claim.
The business does not need to wait for technical closure. The product surface can monetize read-only and human-controlled intelligence immediately while the underlying organism becomes progressively more autonomous. Monetize the bounded world, harden the execution spine with real demand, then let paid outcomes decide which deeper automation deserves capital. Source claim.
As of 14 September 2026, the most defensible description of Neuruh is: a substantial, converged execution architecture with real process-boundary proofs, durable receipt and authority semantics, an emerging autonomous continuation layer, and a commercial real-estate world capable of turning public-record complexity into decision support. Source claim.
| Claim class | Defensible statement |
|---|---|
| Architecture | Organism and major bounded organs implemented across multiple canonical repositories and runtime surfaces |
| Convergence | September PR stack advanced the exact pins, cross-process court seam, execution outcome write-back, and bounded custody / promotion composition |
| Independent adjudication | Real Workers to Sentinel cross-process seam proven, including 16-field identity binding and honest court.failed ingestion |
| Governance | Authority failures observable, typed and fail-closed; current evidence exposes source-SHA and artifact binding defects rather than hiding them |
| DeedSonar | Real economic world with source ingestion, recipe logic, candidate intelligence, customer tapes and a path to paid dossiers |
| Autonomy | Full end-to-end autonomous execution not yet proven; should not be marketed as completed until the remaining transport and promotion chain is observed |
| Potential | Architecture can support multiple bounded worlds if shared contracts are reused and every new world preserves evidence, authority and custody boundaries |
Neuruh is building infrastructure for AI systems that can act without becoming ungovernable. Not a swarm demo. Not a prompt wrapper. Not a collection of LLM calls. The real product is a control system that can make decisions, prove why it made them, obtain bounded authority, execute against real systems, survive failures, independently verify its own execution, and leave an auditable receipt of what happened. Source claim.
Closing statement (source): Neuruh is not finished because it has more agents to build. Neuruh becomes finished, for the purposes that matter, when the existing organs form a durable machine that can move a real objective through authority, execution, independent verification, and economic outcome without the founder acting as packet transport.
Full head and merge SHAs are in the source; the short pins are shown here and the full values are in the SHA explorer below. Sentinel Court #5 merged; Workers #13, #15, #16 merged; SOPS #14 merged; Sentinel Court #6 open; Workers #17 open; Recipe Engine #27 open / draft. Source claim.
665a1e192d4ee4c8f5a8644a2f3f1149f5f556c9
Production-bound Workers / Switchboard composition point used by the current Sentinel seam proof.
Notion autonomous execution chapter records a historical Autonomous Builder corpus of 227 rows: 135 Autonomous Builder, 71 Autonomous Builder 2.0, 20 Poised Guardian, 1 MOTHER; 178 Success, 47 No-op, 1 Queued, 1 Running. Interpretation in the canonical chapter: bounded work repeatedly succeeded, while continuity broke at external dependencies, permissions or genuine founder decisions. This history is not presented as current runtime proof. It is lineage evidence. Source claim. Use the corpus calculator below to recompute the rates.
| Artifact | Evidence |
|---|---|
| Sep-5 partner dossier | DeedSonar locked customer-facing product; Recipe Hunter R1 closed with 30 canonical opportunity recipes and 7 materialized in production; query and saved-list latency improvements documented |
| DeedSonar PR #5 | Stage-only import center; 802 tests; lint / typecheck / build pass; separate approval surface |
| Recipe Engine PR #5 | NC DPS raw research ranking path based on observed facts, with explicit gaps around verified equity and phone |
| LandOS Ingest PR #5 | Current official NC distress sources with custody hashes and address / account fallback identities |
| LandOS Engine PR #5 | NC DPS title-curative scoring contract; 14 contract tests + 33 adjacent regressions |
| Live Integration PR #5 | Seven Sep-13 customer tapes; 7/7 Sentinel PASS; no model calls; external false; production_mutated false; STALL state names human gate |
| Notion Chapter 07 | End-to-end real-estate topology from hunt through evidence, underwriting, decision, action, outcome and calibration |
| Level | Definition | Allowed language |
|---|---|---|
| L0 Historical | Dated artifact or old runtime behavior | "Historical evidence shows..." |
| L1 Source verified | Current code / PR / exact SHA confirms implementation | "Implemented / merged..." |
| L2 Runtime verified | Current live or integration harness observed the behavior | "Proven in runtime / harness..." |
| L3 Production observed | Provider / runtime production state observed | "Observed in production..." |
| L4 Economic outcome | Paid or transaction-level outcome independently captured | "Revenue / transaction proven..." |
| L5 Continuous autonomy | Machine chain resumes, executes, verifies and captures outcomes without founder transport | "Autonomous execution proven..." |
L0 covers the 227-row historical Autonomous Builder corpus and dated artifacts. It licenses "historical evidence shows", nothing stronger.
L1 covers merged PRs with exact SHAs: Sentinel Court #5, Workers #13, #15, #16, SOPS #14. Language: "implemented / merged".
L2 covers the live cross-process seam: 200/200 OK, 16-field identity echo, court.failed ingestion. Language: "proven in runtime / harness".
L3 covers production-observed state such as the founder-reported tape (exit 0, 14/14, /leads/top HTTP 200). The source labels that tape founder-reported, so treat L3 here as source-labeled, not independently verified by this demo.
L4 requires a paid or transaction-level outcome independently captured. The source does not claim L4 for the autonomous chain. The 7/7 tape court is a control-surface proof, not revenue proof.
L5 is open. The machine chain resuming, executing, verifying and capturing outcomes without founder transport has not been observed end to end. Do not use L5 language.
How to read the organism as an operating machine: a set of control surfaces coupled by typed state transitions. Each surface owns a distinct invariant and hands forward a bounded artifact. Source claim.
Every calculator below uses only numbers stated in the source, or numbers you enter. Source-derived results are labeled. No hidden weights are invented; where the source gives no weights, the calculator says so.
Source: 227 rows, 178 Success, 47 No-op, 1 Queued, 1 Running. Source claim.
Success rate: 78.41%. No-op rate: 20.70%. Total: 227.
Seam: 100.00%. Tape court: 100.00%. Model calls on tape court: 0. Spend: $0 (source claim).
Source: 30 canonical opportunity recipes, 7 materialized in production (Appendix D). Source claim.
Materialized share: 23.33%. Not materialized: 23 of 30.
Source prices at 14 Sep 2026: Scan $100, Dossier $499. Source claim, dated.
Bundle total: $1,298. (3 x $100 = $300; 2 x $499 = $998.)
QUALIFICATION = f(runtime_identity, operation_class, source_sha, authority_ref, evidence, policy). EXECUTE only when all six are present and true. Source formula.
Result: NOT QUALIFIED. Missing: source SHA, authority ref. Do not execute.
Source lists ranking components but gives no weights, so this calculator uses an equal-weight mean of the components you score, and excludes UNKNOWN components instead of scoring them zero. Method stated on this page, not a source formula.
Known components: 4 of 5. Mean of known: 7.50. Unknowns: 1 (economic value). Unknown is preserved, not scored zero.
Source section 33: "100x" means 100x more decisions per unit of founder attention, not 100x more agents. This calculator turns transport steps removed into a leverage multiple. Source claim, method stated on this page.
Leverage: 6.00x. Decisions per day after: 12.00. "100x" would require removing 99 of 100 transport steps, or equivalent compounding; it is a direction, not a measured result.
Click a field to see the failure it prevents. The live seam binds all 16; Sentinel originally emitted 8 of 16. Source claim.
Selected: none. 16 fields listed. 8 of 16 was the pre-fix Sentinel emission.
Follow the parked mission through the machine. Each step shows what the source says happens, and where it stops.
Step 1 of 8: Target
A real objective is named and given a durable mission_id. Mission state must survive process death, model substitution, restart, and operator interruption.
msn_2d6e71c5b3e8494494a687288bc7f2ea | last_proven_phase: created
Build a shareable dossier case from the source structure. Tick the fields you have. The meter counts only fields present; missing fields are listed, not hidden.
Loaders fill the builder fields below. They are samples from the source, not live data.
0% complete (0 of 12 fields)
Missing fields:
Share encodes the builder state in the URL hash (#d= base64 JSON). It restores on load. It contains only what you entered here.
No case loaded.
Hub: Neuruh Hub. This demo is one entry in that hub's collection of standalone builds.