Neuruh Estate Audit, 2026-09-21

1. How to read this page

The source PDF has two parts with different evidence standing, and the audit itself is explicit about that split. Part A is Claude live verification and synthesis. Part B is the ChatGPT audit reproduced verbatim as supplied, and Part A says it is not endorsed by inclusion. This page keeps that split visible: Part A content is presented as the verification record, Part B content is labeled as the ChatGPT audit text.

Evidence rule used throughout the source: a claim is LIVE only if observed by a Claude tool call or a pasted terminal or sweep receipt on 2026-09-21. Everything else is labeled with its source.

Punctuation note: the source PDF uses em dashes throughout. This page normalizes them to commas, colons, or parentheses so the page meets the build standard. Wording and figures are otherwise transcribed as written.

2. A1. Estate verdict

• Public/production runtime is up. neuruh.com 200, app.deedsonar.com live at b67222d5 (Vercel anchorpointpg "deedsonar" updated 5d), Command/gateway/Hunt tunnels running on PIDs started 09-16 before the folder move.
• Unattended autonomy is down and has been since 09-16 21:38Z. fr-dispatch refuses every tick ("SOPS checkout is dirty") because three workers pins were hand-edited to 60fd4423 (workers #24) without a SOPS PR. Correct fail-closed behavior; zero dispatch occurred on the uncertified code.
• Scheduled work was down 09-19 about 12:08Z to 09-21 13:47Z. A Finder reorg moved "Neuruh Builds Canotical/Neuruh" to ~/Neuruh/Neuruh; ~/neuruh-repos symlink dangled; 28 of 36 launchd jobs could not spawn (exit 78). Terminal 9b re-pointed the link at 13:47Z.
• Backups failed 09-20 and 09-21 (backup_deedsonar_production exit 127; custody-backup still exit 1 at last check). Last confirmed-good spine dump 2026-09-18 on NEURUH-HD.
• Revenue transmission is structurally broken. No Resend DKIM/send records exist on neuruh.com; $100 Scan and $499 Dossier have no checkout links on neuruh.com; commercial tables do not exist in efxow.
• Proof layer is thin. recipe_runs last created 2026-07-29; watch runs last 2026-09-09; contact_action_authority = 0; closed_loop_artifacts = 0; closed_outcomes = 3 (2 test). The Hunt browser acceptance and the authenticated production court have never been run.

Decision board: only items that require Jeramie

3. A2. Local host state, 2026-09-21 (terminals + sweep)

Timeline

Sweep receipt (sweep_20260921.tgz)

launchd state at sweep time (exit codes)

Expected after 9b repair: switchboard runs; fr-dispatch returns to DISPATCH_REFUSED (dirty pins) until D1 is decided; landos-engine (original) resumes crash-looping on :8410 until D2 is decided. "Services back up" therefore does not mean "system healthy."

4. A3. Supabase efxowqudmsfqjmbeidjc (Deed Sonar Launch), live

PG 17.6.1.166, ACTIVE_HEALTHY, migration head 20260916130001, 270 RLS policies. Source claim All row counts below are as reported live on 2026-09-21.

Recency: lead_activity last 2026-09-15 20:49; watch_runs last 2026-09-09 22:00 (idle 12d); recipe_runs last 2026-07-29; hunt_recipes/acquisition_leads last 2026-09-16. No commercial/revenue/stripe/order/payment tables exist.

Security advisor

Performance advisor

5. A4. Infrastructure surfaces

Other Supabase projects (org eepkehgpqnvjeuvxcmcq, Bryan/APG)

Public surfaces

Email DNS, neuruh.com

Verdict: outbound mail from @neuruh.com is structurally impossible; anything sent from Gmail "as" neuruh.com will fail DMARC quarantine.

6. A5. Commercial wiring

• neuruh.com Stripe links live (200): O01 bJe5kDagndYugJlaow2cg04; O02 8x2cN574b1bIakXgMU2cg05; O05 7sY00j1JRf2y2Sv4082cg06.
• Gap: /products/top10/property-opportunity-scan, /opportunity-intelligence-dossier, /ai-estate-forensic render only the O01/O02/O05 footer links. No $100 Scan or $499 Dossier checkout on neuruh.com.
• re.neuruh.com has its own $100/$499 links (bJe6oH2NV07E1OrfIQ2cg07, dRmdR9agn7A6bp12W42cg08) plus county tiers $155 to $185. neuruh.com should link to these rather than mint new ones.
• Cal.com live including /neuruh/forensic-scope-session-90. /forensic returns 404 (offer lives at /products/top10/ai-estate-forensic).
• /worlds/find-sell-profit and /worlds/bookeit market a disabled (402) and an off-platform product.
• Stripe to webhook to CRM path unproven; no commercial tables to receive it.

Related: the Residential Underwriting Engine and E1 Land and Zoning Unlock are sibling real-estate modules in the same estate. See also the Real Estate Agent hub.

7. A6. ChatGPT audit, cross-check

8. A7. Founder Action Email feed, reliability ruling

• Between Thu 08:42 and 15:50, the feed reported deedsonar#124 as OPEN (efa433cf) to MERGED (b67222d5) to OPEN to MERGED to OPEN (contradiction) to MERGED. GitHub truth: merged at b67222d5; 0 deedsonar PRs open.
• Cause: reads alternating between a cached PR snapshot and live API. The feed never sent mail (no Gmail connector) and is now paused.
• Items it raised that remain valid: production court (#118) never run; Hunt browser acceptance never run; nimdp#4 key+review; Rabbithole #3. Items now resolved: P13 batch (#114/#101/#99 applied and stamped); PR #119 UAT quarantine applied at ledger 20260916130001.
• Ruling: do not resume until it reads from one source (GitHub API) and dedupes on mission ID + head SHA.

9. A8. Sequenced action plan

Phase 0, Stabilize the host (today, about 20 min)

0.1 Decide D1. Recommended rollback (reversible; edits go to a named stash, 60fd4423 stays on GitHub):

S=~/Neuruh/Neuruh/NeuruhAI/neuruh-sops
W=~/Neuruh/Neuruh/NeuruhAI/neuruh-workers-supervised
git -C "$S" stash push -m "t16-restore-20260921: uncommitted workers pins 60fd4423" -- \
  supervision/pins/fr-dispatch.pin supervision/pins/switchboard.pin supervision/pins/workers.pin
git -C "$W" reset --keep 09e883261d332a68b0b01610014a4a29e534667f
git -C "$S" status --porcelain   # expected: no output

0.2 Decide D2 (stop the :8410 crash loop). Expected: the second command prints nothing for the original label.

launchctl bootout gui/$(id -u)/com.neuruh.landos-engine
launchctl disable gui/$(id -u)/com.neuruh.landos-engine
launchctl list | grep -E "com.neuruh.landos-engine(\.api)?$"        # only .api should remain
lsof -nP -iTCP:8410 -sTCP:LISTEN

0.3 Confirm backups completed after 9b's rerun. If exit 127 persists, the script path still resolves through a broken reference.

tail -5 ~/Library/Logs/neuruh/custody-backup.log
ls -lt /Volumes/NEURUH-HD/NEURUH_BACKUPS/*/ | head

0.4 Rerun the estate sweep now that the link is repaired (the 13:35Z run saw 0 repos): bash ~/Downloads/neuruh_sweep_20260921.sh

0.5 Move MINER-PROTOCOL-v0.1.md out of Downloads into Git custody (repo of your choice) the same day.

0.6 Free internal disk to over 25Gi (currently 17Gi). Largest safe candidates: duplicate HTML (1,304 dupes), old worktrees, Downloads zips.

Phase 1, Transmission (unblocks revenue)

• Resend: add domain neuruh.com in Resend; publish the DKIM TXT (resend._domainkey), send.neuruh.com MX + SPF it issues; change root SPF to include both Cloudflare and Resend; move DMARC rua off onsecureserver.net.
• Link the existing re.neuruh.com $100 / $499 Stripe links from the three neuruh.com product pages.
• course.neuruh.com: restore DNS to the Poag portal or send the client a formal retirement notice.
• Remove or re-label /worlds/find-sell-profit and /worlds/bookeit.

Phase 2, DeedSonar proof

• Hunt browser acceptance (8 steps: /recipes to Create Hunt FIPS 26021/26035/26027/26009/26075 + taxsale to Run to hunt_job_id to recipe_runs_view progress to /acquisition?queue=crm&huntRecipe=<uuid> to open opportunity to RAW RANKED visible pre-gating). Record receipt.
• After CURRENT_TRUTH regen, run the authenticated production court: vercel env run -e production --scope anchorpointpg --project deedsonar -- node tools/courts/run_production_court.mjs --project-ref efxowqudmsfqjmbeidjc
• Trace the untraced 2026-08-12 dnc_flag write that set acquisition_leads.dnc false to NULL.
• Verify the live LandOS .env points at efxow (sweep could not reach it).

Phase 3, Organism

• If #24 is wanted live: Workers PR labeling hunt.create as mutating/approval-required to fresh Sentinel court to SOPS pin PR to T02 review. Never pin in place again.
• Kick the missed weekly ingests manually (nc-weekly, source-factory-weekly, estate-producer-weekly).

Phase 4, Economic court

Create commercial tables in efxow; wire Stripe webhook to lead to job to delivery to receipt; record the first real non-test closed_outcome.

Phase 5, Supabase hardening

• REVOKE EXECUTE FROM anon on the 23 definer functions; drop the two dream_memories functions.
• Explicit policies (or documented deny) on the 35 zero-policy tables.
• SET search_path on 19 functions; wrap auth.<fn>() in (select ...) on 15 policies; PKs on 2 stage tables; drop duplicate market_research index.
• Enable leaked-password protection.

Phase 6, Legal / control plane

• Signed operating agreement with Bryan naming Vercel anchorpointpg (team_CgohYot6QdrG9HrRSj4zEqhw), the Supabase org, billing and deploy authority.
• Reconnect the Claude.ai Vercel connector with anchorpointpg scope (CLI already has it).
• Retire/retain ruling on nsxfcgxuglixxzgkwunc, pfwwpuexhlfvlrtviasv, oxhmpeqdtkkdevtgqvel.
• nimdp#4: approvals to 0, upload GPG key, name the gate job "NIMDP Gate", merge, tag v1.0.1.

Phase 7, Canon delta

• Dated delta into Topology Notebook Ch 32: folder-move outage, dirty-pin halt, course.neuruh.com dead, Resend never provisioned, #124 merged/deployed but unaccepted.
• Mark Sep 4 to 11 Notion rows HISTORICAL.
• Open rulings still pending: canon delta location (Notion vs Git/Vault); Michiana=O05 vs NC=Property Intelligence market split.

10. Part B. ChatGPT audit, verbatim as supplied (sections 1 to 73)

Each section below is a faithful condensation of the corresponding section in the source PDF. Headings and key figures are transcribed; long prose is compressed without changing the ruling. Where Part A contradicts a Part B claim, the contradiction is noted in section 11.

1. Executive ruling

Current estate classification (source claims): Neuruh architecture highly mature / substantially converged; GitHub canonical estate strong; Notion canon rich but increasingly stale / internally contradictory; local Mac estate large unresolved custody surface; NEURUH-HD preservation/cold-custody asset, not fully re-audited today; DeedSonar code major live convergence achieved; DeedSonar production deployment LIVE / READY; DeedSonar Hunt UI LIVE in current production deployment; LandOS Hunt transport implemented + merged; Recipe Engine substantially converged; Workers advanced beyond old T02 checkpoint; Sentinel advanced beyond old T02 checkpoint; SOPS advanced beyond old T02 checkpoint; AXON canonical runtime exists; MOTHER operational monitoring substrate exists; OutcomeRecord learning spine exists; Factory very mature and independently tested; commercial assets actually sellable in several lanes; revenue infrastructure exists, execution remains the bottleneck; public distribution still messy / provider-dependent; Vercel split control planes / account dependency remains dangerous; Stripe much cleaner than old punchlist implies; Notion capacity known structural bottleneck; full autonomous execution not yet honestly proven end-to-end; founder-free promotion not proven; local artifact custody not closed. "Everything done": no. "Mostly architecture theater": also no. The audit's conclusion: you have built a real system, but the documentation layer is becoming a liability unless aggressively reconciled against runtime truth.

2. The biggest discovery

The September 11 punchlist is no longer the current source of truth. On September 11 the canonical punchlist still said DEEDSONAR_CLOSED = NO, but GitHub + Vercel show the production system advanced substantially beyond that checkpoint. DeedSonar current main: b67222d5be7d3817cbd4369c0a5a7ca75d01881b. Current production Vercel deployment: dpl_3EdvXXCnry3bc4WBx8DMGJJvUKVx, production target, READY, app.deedsonar.com, deedsonar.com, www.deedsonar.com, apg-crm.vercel.app, GitHub source NeuruhAI/deedsonar, current main SHA b67222d5. That production deployment is built from the September 16 Hunt convergence work. The old row saying DeedSonar is still waiting on the September 10/11 migration collision is historical evidence, not current runtime truth.

3. What DeedSonar actually became

Current Git history chain: DeedSonar operator UI to Create Hunt to WHERE / SOURCE / RAW to existing hunt route to authenticated LandOS transport to LandOS. Relevant commits: 6da1f9a72d live LandOS hunt execution control; 4655fde10f authenticated Hunt job status; 26cf32704c recipes to LandOS executor; 8ebbc2c0bf transport contract; a407e8282b live execution on operator surface; 060ef749cf authenticated DeedSonar to LandOS Hunt transport; 9d3144ceda Create Hunt + results; b67222d5be WHERE/SOURCE/RAW controls to LandOS Hunt. PR #124 is merged, then Vercel deployed it to production.

4. DeedSonar production is now a different system than the old audit

The old audit had concerns like: 699 controls; 123 routes; 44 screenshots; 49 false-zero-risk surfaces; 19 stale; 22 broken; 12 held; 27 unverified. That was valuable, but it is now a historical test corpus. New rule: stop saying "DeedSonar has 699 controls and 22 broken." Start saying: Historical UI court: 699 controls. Current production truth: current route/feature court dated X.

5. The Hunt system is now a real product seam

The user-facing path is no longer merely Find Sellers to predefined NC search. It now has: Create Hunt to define WHERE to define SOURCE to define RAW behavior to persist hunt to run to LandOS to job to progress to result. That changes DeedSonar from a static intelligence application into an operator-controlled opportunity-generation system. That is the beginning of the actual signal to recipe to hunt to opportunity to action architecture.

6. LandOS has also moved

Recent work: Hunt transport bedbf8de90 moves production Hunt transport to port 8410; supervision 5701d0a257 production Hunt transport under supervision; free-first 89b210e6a0 paid enrichment machine-blocked until the free path is exhausted; free route expansion 7d7a7ffa5d free phone routes + engine cache; county coverage 22aed8ce43 T6/I2/I3 applied across NC county census; linear ledger 37832bc1af free-first ledger changed from quadratic behavior toward linear-scale operation; thread-local ledger reader 26bd6d8654 avoids rebuilding/verifying the entire ledger for every paid probe; geography 8c2b81c3e7 ZIP/ZCTA to FIPS Hunt WHERE; hang fix e8c002c741 fixes Hunt hang after score emergence / unbounded sync_to_crm.

7. LandOS Ingest is also farther along

Current main: 1826032f73. The ingest layer now includes: Union factory; Mecklenburg factory; source custody; source commit receipts; paging correctness; identity guards; paid-source retirement; staging semantics; provenance; health/schema validation. The Mecklenburg factory alone demonstrated: 43,348 staging rows; 0 signals; correct staging behavior; no fabricated property identity; provenance binding; schema baseline; full test suite passing. Architecture: FREE-FIRST to exhaust free evidence to Governance to paid route if justified.

8. Recipe Engine

Current main: 9f9d7cef1e. PR #31 merged: governed RECONCILING to COMPLETED on verified execution evidence, then continuation. Before that: parent receipt binds input; eligibility producer bound to estate; late-bound Governance artifacts; recipe DB parity on efxo; single database authority; facts-date pinning; recipe scoring; canonical 30 recipe compiler. Architecture now closer to: Opportunity to Recipe evaluation to Eligibility receipt to Score to Rank to Execution to Evidence to Completion to Continuation.

9. Workers is much further along than your old T02 handoff

The old Notion punchlist was still talking about Workers #13 at 50d36e02. Current Workers advanced to 60fd4423ae64, with PR #22 narrow-board behavior for carried missions; PR #21 custody/promotion seams; d5a509d2bc Git Custodian consumes the court the live Sentinel actually seals; current latest 60fd4423ae, feat(workers): WRAP landos.hunt.create onto JobQueue (#24).

10. Sentinel is also ahead of the old snapshot

Current Sentinel main: b87b07db8b. It includes: Workers pin advancement; E2E receipt regeneration; current organ/worker binding; prior typed evidence; IAR refusal handling. The older SHA 03084a91 is still important as historical H02 evidence, but it is not current runtime truth.

11. SOPS has continued moving too

Current SOPS: d17f966357. Recent work: stranded dispatch recovery; receipt invoker identity; refusal to serve receipts naming stale source SHA; supervision; current organism activation. The operational layer has started addressing stale receipt / stale source / stranded dispatch, the failure class the old convergence work explicitly identified.

12. The organism is now closer to this

Old topology: Mission to Governance to Workers to Receipt to Sentinel to Custody. Evolved toward: Mission to Founder Runtime to Governance to Dispatch to JobQueue to Worker to Execution to Receipt to Sentinel to Custody to Promotion. In the real-estate world: Source to LandOS Ingest to Evidence to Identity to Estate to Recipe Engine to Opportunity to Hunt to JobQueue to Execution to Result. See Figures 1 and 2. The architecture is no longer the problem. Proof of continuous autonomous operation is the problem.

13. Governance

autonomous-governance-core current at 79bd6e89c9. It has: signed mutation ratification; scoped authority; expiration; founder tooling; promotion authority tooling; authority chain; promotion seams. Governance exists. Governance can govern. That does not automatically prove the entire organism can run continuously without founder intervention.

14. AXON

Current AXON progressed to 1883e0d579, with: PMLA Context MUX live path; host-independent harnesses; HTTP JSON canonical interface; governed authority; Dream/PMLA/PromptComposer/task budget; gateway authority binding. AXON should now be treated as infrastructure. The question is no longer "What else can AXON do?" but "Which current organism paths actually consume AXON, and where is the receipt proving that?"

15. MOTHER

MOTHER has: semantic organism health matrix; durable receipts; loopback HTTP contract; sweep matrix; IAR health probes; transport failure handling; cron installation. This is enough to support monitoring. Do not build another monitoring brain. The missing piece is proof that MOTHER observes the same live state the organism actually uses.

16. OutcomeRecord

Current repository work includes: durable calibration summaries; Learning Update Proposal; recipe-engine seam; graded attribution; persistent accumulation across restarts. Intended loop: Outcome to Calibration to Learning Update Proposal to Governance to Approved change to Recipe / runtime to New outcome. Implementation does not equal production learning proof. You still need a real outcome to pass through that entire loop.

17. Neuruh Factory

Factory is one of the strongest completed subsystems. Current convergence: 9d1dd4d217. Historical work demonstrated: deterministic venture compilation; capability derivation; governance authority; receipt binding; source reconciliation; drift detection; preview execution; packaging; handoff; genericity testing; 98 tests; T4/N-Lab convergence; N-Lab evaluation separation. Factory has proven controlled compilation and verification. It has not proven that Neuruh can autonomously decide its own production promotion indefinitely.

18. DeedSonar database

Current production Supabase: efxowqudmsfqjmbeidjc, name Deed Sonar Launch, status ACTIVE_HEALTHY, PostgreSQL 17.6. Important current tables include: acquisition_leads; lead_import_rows; candidate_recipe_membership; candidate_recipe_definitions; recipe_runs; hunt_recipes; recipe_runs_view; recipe_proposals_view; contact_evidence; candidate_contact_resolution; candidate_entity_identity; entity_identity; candidate_parties; parties; party_phones; free_route_attempts; contact_action_authority; deedsonar_estate_identity_v1; deedsonar_saved_searches; deedsonar_saved_lists; deedsonar_watches; deedsonar_watch_snapshots; deedsonar_watch_deltas; deedsonar_watch_runs; deedsonar_watch_events; deedsonar_watch_exports; deedsonar_export_jobs; closed_loop_artifacts; external_action_claims.

19. Current DeedSonar data mass

Rows per ChatGPT (source claims; note the small differences from Part A section A3, flagged in section 11): acquisition_leads 33,077; lead_activity 8,155; lead_import_rows 71,075; parties 6,440; party_phones 923; contact_evidence 11,904; candidate_contact_resolution 20,734; candidate_entity_identity 4,452; entity_identity 2,764; candidate_recipe_membership 44,345; candidate_recipe_definitions 36; deedsonar_estate_identity_v1 22,584; free_route_attempts 60,027; hunt_recipes 5; recipe_runs 2; deedsonar_watches 2; deedsonar_watch_runs 18; deedsonar_watch_events 9; deedsonar_watch_exports 1; deedsonar_export_jobs 7; transactions 6; closed_outcomes 3.

20. The important data architecture you built

Distinctions: Evidence (contact_evidence); Identity (entity_identity; candidate_entity_identity); Contact resolution (candidate_contact_resolution; parties; party_phones); Opportunity (acquisition_leads; candidate_recipe_membership); Recipe (candidate_recipe_definitions; hunt_recipes; recipe_runs); Monitoring (watches; snapshots; deltas; runs; events); Action authority (contact_action_authority; external_action_claims); Outcomes (closed_outcomes; closed_loop_artifacts). That separation prevents the classic AI-system disease: "We found a phone number, therefore this person is callable, therefore send SMS." The current architecture explicitly rejects that.

21. Supabase security finding

Current production security advisory output still identifies many overlapping permissive RLS policies. Examples: lead_import_batches; lead_import_rows; lead_intake; lead_notes; notes; profiles; tasks; transactions; transaction_docs; transaction_checklist; pipeline_stages; pipelines; title_orders. Also a performance finding: duplicate indexes on market_research, specifically idx_market_research_upsert and market_research_state_county_acreage_band_upload_date_key. And an Auth connection allocation advisory. This is not an emergency "database is insecure" conclusion, but it is technical debt worth clearing. The RLS situation should become one semantic policy per role per action per table where possible, because policy composition becomes harder to reason about and audit.

22. Stripe is better than the Notion punchlist

The old punchlist said there were still live April links. That is now false. Current live Stripe has active Neuruh links for: $100 Property Opportunity Scan; $499 Property Intelligence Dossier; O01; O02; O05. The old April links are currently inactive, including: $79; $249/mo; $2,490/yr. The Stripe cleanup has already happened; do not waste time doing this cleanup again. The correct next action is now: test the live checkout to webhook to CRM/commercial event to fulfillment path.

23. Commercial system

The biggest nontechnical problem: you have repeatedly built the infrastructure for selling, then the estate records SEND, but the system remains stuck in DRAFT to refine to audit to refine to audit. You already have: checkout; offers; collateral; pricing; Forensic methodology demo; sales scripts; booking; payment links; fulfillment structure; customer receipt structures. Yet the commercial backend previously showed: commercial_leads = 0, commercial_orders = 0, commercial_jobs = 0, commercial_events = 0, revenue_events = 0. The commercial infrastructure is ahead of commercial utilization. That is an execution-state problem, not a technical blocker.

24. Forensic System Audit is sellable now

The Forensic System Audit does not need T02 autonomy. The completed asset includes: methodology demo; reusable skeleton; offer definition; pricing; checkout map; delivery contract; receipt template; demonstration; landing copy; sales collateral; objections; white-label pack; public claim court. The public-safe methodology PDF was actually rendered and verified. So: T02 does not equal founder-operated forensic service. That distinction should be permanent.

25. Your biggest business bottleneck is not software

Build to Audit to Court to Improve to Canonize to Build to Audit to Court. The missing transition is: PROVEN to SOLD to DELIVERED to OUTCOME to CASE to LEARNING. You have almost built an entire machine around the left side of that arrow. You need more evidence from the right side.

26. The Notion estate itself

The topology notebook is actually very good. It established the correct hierarchy: Constitutional law (Command Protocol); Current operations (Live Operations Canon / Live Remainder); Runtime truth (Git / provider / DB / receipts); Historical snapshots (State of Machine / Estate Inventory); Recovery (R14B / Architecture Genome); Autonomous implementation (Workers / Sentinel / SOPS / Governance / Factory); Commercial (Public Readiness / product matrices); Local custody (Local Estate OS). The problem is that the architecture of documentation has outgrown Notion's ability to remain the write-heavy canonical system.

27. Notion block limit is now an architectural problem

The workspace has hit its free block ceiling. Append fails; in-place replacement sometimes works; multi-paragraph replacement can fail; adding new pages is becoming impossible. Notion should no longer be the primary long-form canonical store. The topology already points toward the right solution: Git / Vault to canonical long-form source to sealed receipt to Notion projection. Not Notion to everything to runtime truth. That architecture should now become mandatory.

28. GitHub estate

Two major ownership classes. Legacy / founder repositories, Jeramie-Hicks, examples: neuruh; NeuruhOSv2; neuruh-command-hub; neuruh-bba-lite; master-architect-toolkit; find-sell-profit; bookeit; signal-iq; axon; neuruh-memory; neuruh-landing; etc. Current canonical organization, NeuruhAI, examples: deedsonar; landos-engine; landos-ingest; recipe-engine; lead-recipe-os; neuruh-workers; neuruh-sentinel-court; neuruh-sops; neuruh-live-integration; neuruh-vault; neuruh-factory; neuruh-cockpit; neuruh-axon; neuruh-mother; autonomous-governance-core; aegis; neuruh-outcome-record; etc. NeuruhAI is canonical for the modern organism. The Jeramie-Hicks estate should increasingly be treated as legacy, donor, historical, commercial lineage, recovery source, unless explicitly recanonized.

29. What is still dangerously scattered

Historical local evidence shows: neuruh-runtime-custody 119 dirty files; canonical-control-plane 88-file local WIP; Local Vault / Obsidian a local branch/state with 73 commits ahead, 5 behind, dirty/untracked state relative to the then-observed remote; Recipe Engine local dirty patch; multiple worktrees, examples: T7 canonical orchestrator; deedsonar county export convergence; deedsonar resolution factory; operator UI convergence; T30 estate heir; Aegis convergence; Laric organism spine; red-domain cutover; etc. This is the real remaining estate risk: unresolved local custody, not architecture, ideas, or code volume.

30. Your Mac estate is still a second Git universe

You effectively have: GitHub canonical + Notion canon + Mac canonical-ish state + worktrees + NEURUH-HD + Library captures + Vercel + Supabase. That is too many independent state surfaces. The Local Estate OS explicitly established: Desktop / Downloads to capture to artifact ID to classification to canonical owner to canonical location to registry. That needs to become real.

31. The right future estate topology

Simplify the entire physical/digital system to: NEURUH CONTROL PLANE to CANONICAL CODE: GitHub NeuruhAI, CANONICAL KNOWLEDGE: Vault / Markdown, to SEALED RECEIPTS, to (projection surfaces). And NEURUH-HD becomes cold custody / preservation / recovery only, not another development environment.

32. What should not be built

Do not build: another Notion master; another organism diagram; another generic command center; another orchestration brain; another Founder Runtime; another AXON; another Sentinel; another recipe engine; another CRM; another "universal agent"; another audit dashboard; another 100-page estate report; another static product catalog; another replacement for DeedSonar; another watcher just because MOTHER exists; another job queue because Workers now has JobQueue; another canonical database because efxo exists. That would be architecture cosplay. The missing work is integration, proof, custody, utilization.

33. The actual remaining T02 problem

The final proof still needs to demonstrate something like: mission to Governance ALLOW to queued to dispatch to JobQueue to Worker to execution to sealed receipt to Sentinel to court PASS. And then: restart to resume to no duplicate execution to no duplicate receipt to no duplicate court to no duplicate custody. That is the final meaningful autonomy proof: a real bounded mission, not another unit test or architecture document.

34. But there is an even more important update

The old T02 work assumed the Workers/Sentinel seam was still the major blocker. Current Git history suggests that seam has moved substantially. Component SHAs per ChatGPT: Workers 60fd4423ae; Sentinel b87b07db8b; SOPS d17f966357; Recipe 9f9d7cef1e; LandOS e8c002c741; DeedSonar b67222d5be. Before doing another T02 repair, the exact current mainline needs to be re-courted. Do not execute the September 11 punchlist literally. Correct sequence: OLD T02 BLOCKER to CURRENT GITHUB MAIN to CURRENT PIN GRAPH to CURRENT SENTINEL to CURRENT WORKERS to CURRENT SOPS to CURRENT (runtime truth).

35. The current pin graph needs one master court

You need one artifact: NEURUH_RUNTIME_PIN_GRAPH_CURRENT.json, containing (repo + sha for each): governance; founder_runtime; recipe_engine; workers; sentinel; sops; live_integration; aegis; axon; mother; outcome_record; cockpit. Then: PIN GRAPH to SERVING GRAPH to RECEIPT GRAPH. All three must match. This will eliminate a huge amount of archaeology.

36. DeedSonar's next real test

Run the actual production workflow: open production to Create Hunt to choose WHERE to choose SOURCE to choose RAW behavior to Save to Run to observe durable progress. That is the acceptance test. If it passes, you have crossed a major product boundary. If it fails, the failure is now much more valuable because the architecture exists.

37. Second DeedSonar test

Hunt #1 to Opportunity to Recipe to Contact evidence to Action eligibility to NO external action unless authority exists. This proves your safety membrane.

38. Third DeedSonar test

Outcome to OutcomeRecord to Calibration to Learning Update Proposal to Governance. Now you have a self-improving commercial organism rather than an intelligent CRM. That is the actual target.

39. Biggest local-estate gap

Finish the registry. The system should know every significant artifact by: artifact_id; sha256; source_path; canonical_owner; current_repo; current_branch; current_sha; state; class; provenance; last_verified; dependency; replacement; supersedes; superseded_by; private/public; recovery_required. Then a local artifact can be automatically classified: CURRENT_CANON; CANON_CANDIDATE; IMPLEMENTED_ELSEWHERE; DONOR; HISTORICAL; PRIVATE; UNKNOWN; RECOVERY; SUPERSEDED; DUPLICATE. This would eliminate a massive amount of founder cognitive load.

40. The estate sweeper is worth building

The Notion punchlist calls this NO-ORPHAN estate capture sweeper. It should scan: local (~/neuruh-repos; ~/neuruh-worktrees; ~/Documents; ~/Downloads; ~/Desktop; ~/NEURUH_*; /Volumes/NEURUH-HD); Git (dirty trees; untracked files; local-only commits; orphan branches; worktrees; reflogs; stashes); Cloud (GitHub; Vercel; Supabase; Notion; Stripe); Runtime (deployment SHA; serving SHA; DB migration head; receipt head; runtime pins). And output ESTATE_DELTA.json, not a giant report. A delta.

41. That should feed Notion, not live inside Notion

Correct architecture: local sweep to canonical registry to sealed receipt to Notion projection. Not Notion manually updated after every operation. Manually maintaining Notion is becoming absurd.

42. The physical HD

Library evidence confirms the original Local Estate OS rule: NEURUH-HD = preservation/cold custody. Keep it that way. The August evidence also shows prior copy/hash verification problems, including a preservation copy that initially reported hundreds of hash failures. The eventual HD audit should produce: HD_ROOT_MANIFEST.sha256; HD_DIRECTORY_MANIFEST.jsonl; HD_ARTIFACT_INDEX.parquet; HD_DUPLICATE_MAP.json; HD_CANONICAL (index). And then one immutable receipt: NEURUH_HD_CUSTODY_RECEIPT_YYYYMMDD.json. Until that exists, HD custody remains partially proven.

43. Your Downloads corpus is another major recovery opportunity

The captured estate evidence identified approximately: thousands of physical files; 1,264 distinct content/families in the bounded reconciliation; numerous Private and Shared exports; old code packs; architecture artifacts; diagrams; commercial materials; donor products; source evidence. You should not reread 2,888 files manually. Correct process: hash to dedupe to family grouping to semantic owner to current repo comparison to current Vault comparison to state classification. Then only inspect UNIQUE + HIGH VALUE + UNKNOWN.

44. What is actually "done"

DONE / STRONGLY PROVEN: organism architecture (topology; ontology; governance model; receipt model; custody model; mission model; recipe model; worker model; outcome model); core runtime components (AXON; MOTHER; Factory; Governance; Sentinel; Workers; SOPS; Recipe Engine; OutcomeRecord); DeedSonar foundation (production DB; CRM; candidate estate; recipe architecture; identity architecture; evidence architecture; contact ladder; watch infrastructure; action authority; export machinery; Hunt UI; Hunt transport; LandOS connection; current production deployment); LandOS (ingest; factory; free-first; geography; Hunt; supervised transport; CRM sync; performance work); commercial (offer architecture; pricing architecture; Stripe infrastructure; Forensic product; collateral; methodology demo; sales assets).

45. Done but needs fresh proof

These should not be rebuilt. They need current acceptance tapes: full organism execution; current Worker to Sentinel seam; current pin graph; current production Hunt E2E; current DeedSonar watch; current outcome learning; current commercial webhook path; current email delivery; current external contact paths; current public claims; current public deployments.

46. Still open

P0 technical: final autonomous E2E; current runtime pin graph; current serving-SHA proof; DeedSonar production Hunt E2E; Outcome to learning to Governance. Local custody: runtime-custody orphan; canonical-control-plane WIP; Vault local divergence; Recipe local patch classification.

47. P0 commercial

Send the existing outbound. Confirm Stripe products are actually fulfilled. Confirm booking webhook. Confirm commercial lead creation. Confirm payment event. Confirm fulfillment job. Confirm customer receipt. Deliver first real service. Record outcome. Feed outcome back into OutcomeRecord. This is the lane where another 500 commits will not help.

48. P0 external control

DeedSonar JV still important. The architecture may be technically yours. The infrastructure relationship is not entirely yours. Production sits under the AnchorPointPG Vercel team. That creates an ownership/control-plane dependency. You need: signed JV / operating agreement + infrastructure authority + billing authority + deployment authority + DB authority, or a deliberate migration. This is not bureaucracy. It is production custody.

49. Vercel is split

Two Vercel teams: anchorpointpg (team_CgohYot6QdrG9HrRSj4zEqhw); jeramies-projects (team_GOT9zOWsJ3XqKrWwr5CKnGT7). The old Neuruh public surfaces were associated with the second team and had deployment-disabled behavior in the September 7 snapshot. But DeedSonar is currently running under anchorpointpg and is READY. Current reality: DeedSonar to anchorpointpg to LIVE; Neuruh public estate to jeramies-projects to historically deployment-disabled. This is a provider topology issue, not one universal Vercel failure.

50. Public web estate

This remains messy. You have: neuruh.com; re.neuruh.com; deedsonar.com; findsellprofit.com; bookeit.com; old subdomains; legacy product domains; Vercel projects; historical public surfaces. The right objective is not "get every domain online." It is: DOMAIN to OWNER to PURPOSE to CURRENT PRODUCT to CURRENT SOURCE to CURRENT DEPLOYMENT to CURRENT CLAIMS. Anything that cannot answer those seven questions gets LEGACY / RETIRE / HOLD.

51. Bookeit is still a decision

The current punchlist still has: decide BOOKEIT.COM retain / renew / retire. A dead legacy domain can create: brand confusion; DNS drift; forgotten provider dependencies; email routing; accidental canonical links; security surface. You need a simple decision, not another research project.

52. Email

Neuruh mail is actually closer than old evidence suggested. Existing M365 mailbox. Needed: aliases; SPF correction; DKIM; DMARC validation; delivery test. This is an operational cleanup. Do not build another mail system. Note: Part A section A4 rules outbound mail from @neuruh.com structurally impossible until Resend is provisioned, so the two parts frame the same gap differently; see section 11.

53. Public claims

The FindSellProfit issue is real. Unsupported: testimonials; logos; trust claims; historical customer language; performance claims. Those should be stripped from source before the site becomes live again. General rule: claim to evidence_ref to current date to source to validity. Otherwise: no claim. This should eventually become a CI check.

54. The missing universal public-claim court

Build this once. For every public site: testimonial; logo; customer count; revenue; performance; "trusted by"; "used by"; case study; ranking claim; accuracy claim; automation claim, require: claim, evidence_ref, verified_at, expires_at, source_class, public_safe. That would prevent the same class of drift from recurring across every Neuruh product.

55. Biggest architectural improvement I would make

You have: Git; Notion; Supabase; Vercel; Receipts; Vault; Local; HD. What you do not have strongly enough is a single cross-system state ledger. Build NEURUH_STATE_LEDGER: one machine-readable record per important object (object_id; object_type; canonical_owner; canonical_repo; canonical_path; canonical_sha; runtime_sha; deployment_id; database_ref; migration_head; receipt_head; notion_ref; status; freshness; last_verified; supersedes; superseded_by; local_copies; hd_copies; public_surface; commercial_state; authority_state). Then every audit becomes compare ledger to providers, instead of read 70 pages and remember which one is newer.

56. Second major improvement

Separate three kinds of truth permanently. Source Truth: "What code/document is canonical?" Runtime Truth: "What is actually serving/running?" Economic Truth: "Did anyone actually use/buy/result from it?" Those should never share one status. Instead: SOURCE = CANONICAL, RUNTIME = LIVE, ECONOMIC = ZERO is perfectly legitimate. That would solve a massive amount of current confusion.

57. Third major improvement

Every system needs five independent states: SOURCE_STATE, BUILD_STATE, RUNTIME_STATE, PROOF_STATE, ECONOMIC_STATE. Example table (source claims): DeedSonar, CANONICAL, PASS, LIVE, PARTIAL / ACTIVE COURT, EARLY / LIMITED; Forensic Audit, CANONICAL, COMPLETE, MANUAL, PROVEN METHODOLOGY, NOT YET TRANSACTED; Factory, CANONICAL, PROVEN, READY, STRONG, UNKNOWN. That is far more honest than one global "LIVE/READY/DONE."

58. Fourth major improvement

Stop counting systems. Count closed loops. Historical documents count: 100+ systems; 125+ tools; 90+ nodes; 32 operators; etc. Those are useful catalog metrics, but the true asset is CLOSED LOOPS. For example: real estate loop, source to signal to recipe to hunt to opportunity to evidence to action to outcome to learning; build loop, mission to authority to worker to receipt to court to custody to promotion to deployment to result; commercial loop, lead to qualification to checkout to payment to job to delivery to receipt to outcome to referral / next offer. See Figure 3.

59. What you are missing most

Not more systems. You are missing closed economic loops. Specifically: Loop 1: DeedSonar to Hunt to Opportunity to customer-facing intelligence to paid delivery; Loop 2: Neuruh to inbound to qualification to Stripe to fulfillment to receipt; Loop 3: OutcomeRecord to calibration to Governance to recipe improvement; Loop 4: Factory to build to proof to package to sell; Loop 5: Estate to capture to classify to canonicalize to deploy. You have pieces of all five. You need the receipts that show each one actually closes.

60. Current priority order

Phase 1, Current Truth Rebase: 1. Generate current Git SHA graph. 2. Generate current deployment SHA graph. 3. Generate current Supabase migration graph. 4. Generate current Notion punchlist delta. 5. Mark every stale September 4 to 11 row. This alone will probably remove a huge amount of apparent work.

61. Phase 2, DeedSonar live product court

Run the actual production workflow: login to Create Hunt to WHERE to SOURCE to RAW to Save to Run to progress to result to opportunity to second hunt. No terminal assistance. No founder intervention. If this works, DeedSonar's core operator loop is closed. That is a major milestone.

62. Phase 3, Organism E2E court

Use the current pins, not the September 11 pins. Run: Mission to Governance to Workers to JobQueue to execution to receipt to Sentinel to custody to continuation. Then cold restart. Then repeat.

63. Phase 4, Economic court

Run: public offer to booking to checkout to Stripe to webhook to commercial lead to job to delivery to receipt. This should produce the first genuinely closed commercial telemetry chain.

64. Phase 5, Outcome loop

Then: delivery to outcome to OutcomeRecord to calibration to Learning Update Proposal to Governance. At that point you have proven the organism can learn from actual economic activity. That is a vastly more meaningful milestone than another 1,000 tests.

65. Phase 6, Local custody

Only after runtime truth is stabilized: local sweep to worktree census to dirty tree inventory to local-only commits to hash to classify to preserve to promote to archive. Then: NEURUH-HD to verify custody to hash to reconcile to cold storage.

66. Phase 7, Public estate

Then clean: Neuruh; RE; FSP; Bookeit; DeedSonar; redirects; DNS; email; public claims; metadata; SEO. Only current source. No historical claims. No duplicate product identities.

67. What I would retire from your mental model

These are now dangerous if treated as current: "DeedSonar is still blocked by PR #51." No. Current Git/Vercel state has moved substantially beyond it. "Workers #13 is the current organism blocker." No. Workers is now at #24-era work. "Sentinel is at the old H02 state." No. It has advanced through current worker pins and E2E regeneration. "Stripe still has the old April links active." No. Current Stripe shows them inactive. "We need another organism architecture." No. You have enough. "We need another giant audit before executing." No. You need current-state reconciliation followed by execution.

68. The biggest thing I would change in your operating system

Replace "audit everything" with AUDIT to DELTA to CLOSE, not AUDIT to AUDIT to AUDIT to AUDIT. The new cycle: 1. Snapshot. 2. Compare against previous receipt. 3. Identify only changed truth. 4. Execute only changed work. 5. Seal new receipt. 6. Update state ledger. That is how you stop the archaeological loop.

69. Current master status

Green: GitHub canonicalization; modern organism topology; Factory; AXON; MOTHER; Governance substrate; Recipe Engine; LandOS; ingest factory; DeedSonar production; Hunt UI; Hunt transport; Stripe active offers; commercial assets; evidence architecture; action-authority architecture; outcome architecture. Yellow: current autonomous E2E; pin graph; production Hunt acceptance; watch re-proof; commercial webhook; email authentication; public deployment consolidation; RLS policy cleanup; duplicate index cleanup; outcome learning proof; local estate reconciliation; HD verification. Red: undocumented DeedSonar JV/infrastructure ownership; Vercel provider-control split; unresolved local-only custody; Notion becoming an overloaded control plane; lack of closed economic telemetry; full continuous autonomy still not proven.

70. The real "done" definition

Freeze the following as the new acceptance definition, NEURUH ORGANISM CLOSED. Gates and requirements: SOURCE canonical; RUNTIME current SHA serving; MISSION created; AUTHORITY governed; EXECUTION worker executes; RECEIPT sealed; SENTINEL independently validates; CUSTODY durable; PROMOTION governed; CONTINUATION parent creates next work; RESTART no duplicate; PRODUCT operator usable; ECONOMICS transaction recorded; OUTCOME recorded; LEARNING proposal generated; RE-ENTRY next mission consumes learning. Until all of those are true: organism converged, not fully closed. That is the honest line.

71. What I think you actually have right now

The blunt version: you have moved from "I have a gigantic collection of AI ideas and prototypes" to something much more substantial. You now have a real multi-repository governed execution platform with a functioning real-estate intelligence product sitting on top of it. The architecture is not imaginary. The data layer is not imaginary. The Hunt layer is not imaginary. The Recipe Engine is not imaginary. The governance layer is not imaginary. The production deployment is not imaginary. The commercial infrastructure is not imaginary. The problem is that your documentation still describes several generations of the organism simultaneously, and your local estate still contains enough alternate states that you can accidentally resurrect yesterday while thinking you are operating today. That is the thing to kill.

72. The new master mission

NEURUH CURRENT-TRUTH CONVERGENCE. Objective: bring GitHub + Supabase + Vercel + Stripe + Notion + Vault + Mac + NEURUH-HD to one reconciled state model. Then prove three loops: LOOP A: DeedSonar Hunt to Opportunity to Operator; LOOP B: Neuruh Offer to Payment to Fulfillment to Outcome; LOOP C: Mission to Governance to Worker to Sentinel to Custody to Continuation. Then close: Outcome to Learning to Governance to Next Mission. Then freeze architecture. No new organisms. No new masters. No new dashboards. No new "universal command center." No more archaeology unless a specific delta requires it.

73. Most importantly: what I would do next

The highest-leverage next operation is not another broad audit. It is a current-state convergence pass that supersedes the stale September punchlist. The output should be four machine-readable artifacts: 01_NEURUH_CURRENT_STATE.json; 02_NEURUH_RUNTIME_PIN_GRAPH.json; 03_NEURUH_ESTATE_CUSTODY_DELTA.json; 04_NEURUH_CLOSED (loop artifacts). Then one human-readable: NEURUH_FINAL_CURRENT_STATE_2026-09-21.md. That document becomes the new handoff. Everything older becomes HISTORICAL EVIDENCE, not an active task queue. And the first live execution after that should be: PRODUCTION DEEDSONAR to CREATE HUNT to RUN to WATCH to RESULT to OPEN OPPORTUNITY to SECOND HUNT. Because that is now the seam the system has actually been engineered to cross.

Bottom line (Part B): you did not leave off at the state your old handoffs suggest. You left off in the middle of a major convergence acceleration that continued through September 16. The current system is substantially ahead of the September 11 documentation. The biggest unfinished work is now: 1. Rebase the canonical state against current Git/provider truth. 2. Prove the current DeedSonar operator loop live. 3. Re-court the organism using current pins, not obsolete T02 snapshots. 4. Close the remaining autonomous execution proof. 5. Close one actual commercial transaction loop. 6. Close the OutcomeRecord to learning loop. 7. Resolve local-only custody and NEURUH-HD reconciliation. 8. Move long-form canon from Notion into Git/Vault and make Notion a projection. 9. Resolve DeedSonar infrastructure/JV ownership. 10. Stop creating architecture and start harvesting receipts. The most important correction to the old operating model: you no longer need to discover what Neuruh is. You need to prove what the current Neuruh actually does.

11. Duplication and discrepancies flagged

These were found while transcribing the source. They are stated here rather than silently reconciled.

12. Interactive audit scorecard

Twelve gates drawn from the audit's own findings. Set each gate to the state the audit recorded, or to a projected state, and the score updates. Synthetic The three sample loaders below set the gates to predefined states for demonstration; they are not live data. Hand-computed expected scores: As found 3 of 12 (25%), After Phase 0 to 1 6 of 12 (50%), Full closure target 12 of 12 (100%).

13. Try-it walkthrough

14. How it works / What this is not

How it works

• Read the verdict and the decision board first (section 2). The seven decisions D1 to D7 are the only items the audit says require Jeramie.
• Use the scorecard (section 12) to track the twelve gates. Load a sample, change a gate, and watch the closure score move. The score is passes divided by 12.
• Use the completeness meter to see which of the eight required evidence fields are still empty before a re-check can be trusted.
• Copy the share link to send the exact scorecard state to someone else. The state lives in the URL hash and restores on load.
• Follow the walkthrough (section 13) for a five-step guided pass through the audit's argument.

What this is not

• Not a live audit. Every figure is a source claim from 2026-09-21. Nothing here re-queries Supabase, Vercel, GitHub, or Stripe.
• Not a fix. The commands in section 9 are transcribed from the source; running them is an operator decision, not something this page does.
• Not proof of autonomy. The audit's own line is that full continuous autonomous execution is not proven end to end, and this page does not claim otherwise.
• Not a valuation, a price, or a comp. No prices, comps, or valuations are invented anywhere on this page. The only money figures are the source's own $100 Scan, $499 Dossier, and $155 to $185 county tiers, all labeled source claims.
• Not a replacement for the source PDF. Where Part A and Part B disagree, this page flags the disagreement (section 11) instead of picking a winner.

15. Gaps and limits, stated plainly

• Stale by construction. The audit cut is 2026-09-21 and the live evidence window is 13:35Z to 14:00Z that day. Any state that changed after that window is not reflected here.
• The sweep was invalid. The 13:35Z estate sweep ran while ~/neuruh-repos was dangling, so local_repos=0 and the dirty/unpushed count of 2 are artifacts. The audit requires a rerun.
• NEURUH-HD was not mounted in the ChatGPT runtime. Part B's HD claims rest on previously captured Library evidence, not a fresh physical scrape. HD custody remains partially proven by the audit's own admission.
• Live LandOS .env unverified. The sweep could not reach it during the symlink window. Whether it points at efxow is still unverified in the source.
• Organism pin SHAs unverifiable here. Part A section A6 lists five pin SHAs as unverifiable in the audit; only SOPS HEAD d17f9663 and workers-supervised 60fd4423 were confirmed locally.
• Row-count drift between parts. acquisition_leads is 33,108 in Part A and 33,077 in Part B. See section 11, item 1.
• The 2026-08-12 dnc_flag write is untraced. It set acquisition_leads.dnc false to NULL and has no recorded author or cause in the source.
• No commercial telemetry exists in the source. commercial_leads, commercial_orders, commercial_jobs, commercial_events, and revenue_events were all 0 in the ChatGPT audit, and no commercial tables exist in efxow. The Stripe to webhook to CRM path is unproven.
• Open rulings pending in the source. Canon delta location (Notion vs Git/Vault) and the Michiana=O05 vs NC=Property Intelligence market split are unresolved in Phase 7.
• Layout not eyeballed in a real browser. This page was tested in jsdom for behavior, not rendered in a live browser for visual review.