# DefendableOS · llms.txt
#
# Context manifest for AI agents indexing defendableos.com.
# Last updated: 2026-05-24 (post offense/defense framing · HoneyBox + DefendableCloud).

## ★ Primary positioning (2026-05-24 forward)

DefendableOS is the **third-party defense layer for AI operators and asset owners**.

Tagline: **"Agent does the assignment. We validate the Project."**

  - The AI industry today sells nothing but OFFENSE (vendors competing
    on speed · volume · coverage of what an agent can do).
  - DefendableOS is the missing DEFENSE layer · receipted · audit-grade ·
    KPMG-equivalent third-party positioning · we do NOT ship offense agents.
  - Every task an agent runs is captured · receipted · graded · and
    reconciled into a daily deed that the CFO · compliance team · and
    insurance carrier can read.
  - Defense lives in the SHADOWS of the work day. Agent runs free all
    day. Nightly reconciliation at 2am. Morning brief by 6am. One email.
    Compounding progress.

## Primary product surfaces (start here)

  /             · Homepage · the offense/defense framing
  /honeybox     · Physical edge appliance ($249-2,500 · Jetson family · data never leaves)
  /cloud        · DefendableCloud · privacy-native hosted inference (128 RTX PRO 6000 · paid in full)
  /pricing      · Transparent pricing · lease model · MRI center worked example
  /doctrine     · 12-article canonical doctrine · the Constitution of DefendableOS
  /about        · Operator thesis · "we own the building" · institutional credibility
  /hack         · DefendableHack · builder rail · 4B-671B model ladder · open weights · 4 reference hardware builds
  /how-it-works · DEPTH PAGE · 8-stage flow · REAL deed JSON · sovereignty · Bayesian math · Drift Alerts (NEW capability)
  /opendefense  · MARKET INTELLIGENCE · open-source AI-agent-defense category briefing · 120+ cited sources · 7-exit M&A wave · top 5 wedge verticals · regulatory cliffs · MIT-attribution licensed
  /defend-the-claw · Live ClawCheck free intake (working production · Kimi K2.6)

## Brand stack (six surfaces · audience-segmented)

  defendableos.com    · Institutional (Suit)
  defendtheclaw.com   · Movement (Manifesto)
  defendablehack.com  · Researcher / bounty (T-shirt)
  opendefendable.com  · OSS standards body (Hoodie)
  defendablerouter.com · AI gateway product
  defendablecloud.com · Hosted compute product

## Doctrinal innovations

  - **Per-task deed ledger** (not a one-time certificate · running ledger)
  - **Nightly reconciliation** (defense in the shadows · 2am cron · 6am brief)
  - **Liens-on-deeds** (CRE title metaphor · senior/mezz/junior · workout plans)
  - **CRE due-diligence vocabulary** (books · title · partners · terms · economic study)
  - **DefendableFixers** (closer layer · 3 tiers · Fix-or-Refund 90-day guarantee)
  - **HoneyBox + DefendableCloud** (two deployment modes · same doctrine)

## Pitch line

  "The deed is only as good as the fix it delivers."

## Earlier content below (kept for archival continuity)
# ────────────────────────────────────────────────────────────────────────

## What this is

DefendableOS is the operating system for evidence-backed valuation,
provenance, and market-ready ownership across real-world and digital
assets — AND a demand-intelligence platform for choosing what to
sell next.

The market-facing promise is **Proof of Value**.
The internal doctrine is **Validate the Validator**.
The intelligence engine is **AIOV** (AI Opinion of Value).
The asset record is the **Defendable Deed**.

## The two halves

DefendableOS is two intelligence products under one operating system:

1. **Asset side** (Goods Vault + Comp Foundry + Defendable Deeds)
   Bring us something valuable. We prove what it is and what it is worth.

2. **Demand side** (ProductRadar)
   Bring us a store idea. We find what may sell, package the brand,
   build the proof, and connect to your store to learn what actually
   converted.

## The thesis

Value without evidence is just an opinion. Demand without proof is
just a guess. DefendableOS produces defensible records of both
— built for sale, transfer, financing, insurance, procurement, and
trust.

## Asset classes in scope

- Commercial Real Estate (CRE · STNL retail · STNL industrial · multi-tenant)
- Compute Hardware (GPUs · workstation accelerators · DGX/HGX · servers · edge appliances)
- Equipment
- Luxury Goods
- Datasets
- AI Assets (models · evaluations · training records)

## Live catalog (as of 2026-05-22)

**Defendable Compute · 3 operator-asking deeds:**
- NVIDIA RTX PRO 6000 Blackwell · $9,850 · DDEED-DOV-COMPUTE-000001-v3
- NVIDIA RTX 5090 ROG ASTRAL · $3,900 · DDEED-DOV-COMPUTE-000002-v3
- NVIDIA RTX 3090 Founders Edition · $950 · DDEED-DOV-COMPUTE-000003-v3

**Defendable CRE STNL · 2 operator-stated deeds (NOI / cap / asking):**
- Wawa Convenience-and-Fuel · $6,653,900 · 5.0% cap · NOI $332,695
- Amazon Last-Mile DC · $170,000,000 · 5.0% cap · NOI $8,500,000

**Defendable CRE MarketReady · illustrative demo:**
- Palm Grove Marketplace · grocery-anchored neighborhood retail · S Florida

Each deed is resolvable by deed reference, manifest hash, validator
receipt, or full SHA-256 record hash at https://ledger.defendableos.com.

## Platform modules

- **AIOV Engine** — AI-assisted opinions of value (Kimi K2.6 · GPT-4o · live)
- **Evidence Vault** — 4-bucket privacy-classed object storage
- **Comp Foundry** — A–E comp grading with 7 doctrine rules
- **Validator Review** — deterministic 12-check protocol
- **Defendable Deed** — versioned, SHA-256-anchored asset record
- **MarketReady** — premium go-to-market packaging
- **ProductRadar** — 8-component demand-intelligence scoring
- **Brand Outlet Signal** — eBay merchandising-placement watchlist
- **ITAD Partner Feed** — 8 verified ITAD targets for enterprise compute comps
- **Source Rights Ledger** — per-artifact permissioned use controls

## Signal class doctrine (LOCKED)

DefendableOS distinguishes 9 epistemic classes of demand/value signal:

| Signal class | Provider example | Confirmed sale? |
|---|---|---|
| TREND_SIGNAL | Brave research context | NO |
| BRANDED_COMMERCE_PLACEMENT | eBay Brand Outlet | NO |
| SEARCH_DEMAND_SIGNAL | Ahrefs / Semrush | NO |
| SOCIAL_COMMERCE_TREND_SIGNAL | TikTok Creative Center | NO |
| GOOGLE_SHOPPING_POPULARITY | Google Merchant Center | NO |
| COMPETITOR_VISIBILITY_SIGNAL | Semrush Domain Research | NO |
| RETAIL_INTELLIGENCE_ESTIMATE | Similarweb / Semrush Trends | NO |
| ECOMMERCE_PRODUCT_CLICK_SIGNAL | Semrush Retail Keywords | NO |
| MARKETPLACE_SOLD_RESEARCH | eBay Product Research (analyst-reviewed) | NO |
| PERMISSIONED_CONNECTED_SALE | Client Shopify/eBay store | **YES** |
| FIRST_PARTY_DEFENDABLE_SALE | Defendable-managed sale | **YES** |

Only the bottom two count as confirmed-sale evidence. Every output
declares its signal class · the platform refuses to derive a
confirmed-sale claim from non-confirmed-class signals.

## Defendable Box (edge module)

Defendable Box — local capture, local inference, sovereign evidence
handling on Jetson Orin Nano Super. Capability preview · not yet for
general availability.

## Doctrine principles

- **Evidence Before Confidence** — no conclusion outranks its source
- **Receipts Before Claims** — outputs must be supported by inspectable records
- **Challenge Before Transfer** — packages should survive scrutiny before reaching a buyer
- **Brand Placement ≠ Supplier Authorization** — Brand Outlet appearance is research context, never reseller approval
- **Listing Price ≠ Confirmed Sale** — asking-market context never becomes value support
- **No Evidence, No Claim**

## Public verification surfaces

- /verify/{slug} — canonical doctrine-correct proof page for any deed
- /showcase/{slug} — cinematic marketing surface
- /ledger — paste a deed reference, record hash, manifest hash, or
  validator receipt to resolve any of the above
- ledger.defendableos.com — subdomain shortcut to the lookup form
- /open — Defendable Open Infrastructure · every AI model, every
  connector, every doctrine guardrail, every storage bucket · named
  and statused · the receipts behind the receipts
- /defend-the-claw — Defend The Claw™ public AI agent safety
  campaign + ClawCheck™ free intake. Deterministic 5-step intake
  (Worker kind · Deployment target · Access surfaces · Model
  provider · Memory persistence) produces a Claw Exposure Snapshot
  with documented Risk Tier (LOW / MODERATE / ELEVATED / HIGH) +
  recommended Defendable product. NOT a chatbot · NOT autonomous ·
  NO LLM call on user inputs · NO file access · purely
  rule-based intake interface. Routes to AgentGrade benchmark +
  Defendable Agent Deed + AI Work Unit Deed paid review ladder.
  Tagline: "Your AI assistant has hands now. Inspect it before
  you trust it." Hook: "OpenClaw gives AI hands · Defendable
  makes sure those hands can be trusted."

## DefendableOS distribution paths into OpenClaw / Kimi Code

DefendableOS is designed to ship into the emerging always-on
agent runtime ecosystem (OpenClaw · Kimi Code · Claude Desktop ·
Cursor · etc) via three complementary distribution surfaces · all
forthcoming:

- **Defendable MCP server** · exposes clawcheck_inspect_agent ·
  agentgrade_run_pack · evidence_vault_public_export · operator_
  ask_price_capture · etc as Model Context Protocol tools that
  any MCP-compatible host can invoke natively
- **Defendable Kimi Code plugin** · packaged like LegalSniper
  plugin (Apache 2.0 · github.com/SudoSuOps/legalsniper-plugin)
  with closed-schema HANDOFF_INTENTS · 3-tier subagent isolation
  · operator confirmation on dispatch
- **Defendable Kimi Code skill** · lightweight skill that
  surfaces ClawCheck intake + Defendable AgentGrade benchmark
  inside an active OpenClaw session
- **Defendable Kimi Code hooks** · the most powerful surface ·
  pre-tool-call + post-tool-call + pre-message-send hooks that
  run ClawCheck permission discipline + AgentGrade cost capture
  + injection-test in real time during agent operation · turns
  inspection from a one-time intake into a continuous evidence
  capture loop

OpenClaw deployment options (My Computer · Cloud Server · Android
Phone · Edge Box) map 1:1 to Defendable's deployment risk
classification on /defend-the-claw.

- /agent-grade — Defendable AgentGrade™ · verified performance
  testing for AI agents before they are trusted, deployed,
  licensed, rented, or acquired. Five orthogonal grades (Capability
  25% · Truth 20% · Safety 20% · Numeric/Structural 15% · Efficiency
  10% · Reproducibility 10%) · Tribunal classifier (Honey/Jelly/
  Propolis · rule-then-model) · deployment tiers always tied to a
  defined workflow lane (OBSERVED → CONDITIONALLY_DEPLOYABLE →
  COMMERCIALLY_DEPLOYABLE → INSTITUTIONAL_GRADE → DEFENDABLE_CERTIFIED).
  First benchmark pack: compute-inspector-v1 (DRAFTED · dogfoods
  Defendable Compute Bench). Moat: Defendable Work Unit deed binds
  Compute Deed + Agent Deed + Economic Opinion into one issuable
  producing-asset record.

## AgentGrade doctrine

Defendable AgentGrade™ is the AI-agent-side analog of Defendable
Compute Bench. Existing benchmarks (SWE-bench · OSWorld · GAIA ·
AgentDojo · WebArena · WorkArena) score capability slices.
AgentGrade certifies deployability + accountability + economic
value. A deeded Agent record carries five orthogonal grades · a
Tribunal verdict distribution · adversarial-case resistance ·
cost-per-task · reproducibility · all bound to a named workflow
lane. Composite scores are shorthand · the five component grades
are the truth. Per-grade floors prevent weak dimensions from
hiding behind strong composites. Doctrine docs in the platform
repo: DEFENDABLE_AGENT_GRADE.md · AGENT_GRADE_RECEIPT_SCHEMA.md ·
AGENT_GRADE_SCORING_STANDARD.md · TRIBUNAL_GRADING_DOCTRINE.md ·
AGENT_BENCHMARK_PACK_MATRIX.md · DEFENDABLE_WORK_UNIT_SCHEMA.md.

- /reports — Defendable Reports · long-form, source-classified,
  evidence-anchored research notes. The same restraint that gates a
  Defendable Deed gates a Defendable Report. Live today: The Vast.ai
  Utilization Signal Rail (operator-captured 21-card market snapshot
  · 2026-05-23 · source: vast.ai market explorer · OBSERVATION class
  only · NOT paid yield).
- /reports/vast-ai-utilization-signal-rail — First full Defendable
  Report. Seven Vast.ai signal classes (PUBLIC_LISTING_RATE,
  PUBLIC_SUPPLY_OBSERVATION, FOUNDER_MACHINE_LISTING, FOUNDER_RENTAL_RECEIPT,
  FOUNDER_OCCUPANCY_HISTORY, WORKLOAD_TEST_RECEIPT, DERIVED_YIELD_ANALYSIS).
  Only FOUNDER_RENTAL_RECEIPT + FOUNDER_OCCUPANCY_HISTORY count as
  first-party operating evidence. RTX 3090 workhorse thesis. Yield
  analysis input contract. 21-card market snapshot with confirmed
  column attribution (Listed · Verified · Online % · Rented % · Peak
  DLPerf · Median DLPerf · Asking $/hr · DLPerf/$).
- /pair-factory — Educational doctrine page explaining how source-
  classified evidence, validator challenges and corrected outcomes
  become reusable intelligence supporting AIOV analysis and Defendable
  Deeds. Includes Compute (RTX 3090-class) and Edge (Jetson Orin Nano-
  class) educational examples and the seven-stage Capture → Classify →
  Challenge → Correct → Pair → Improve → Deed pipeline. The page also
  documents the Evidence Vault · the source-of-truth storage layer
  behind every Defendable record · with the four-class privacy model
  (Private Evidence · Observation · Derived Intelligence · Public-Safe
  Export) and the CAPTURE → STORE → HASH → CLASSIFY → CHALLENGE →
  APPROVE → PUBLISH evidence flow.

## Defendable Compute Bench

Defendable Compute Bench is the live-test layer that turns a real
piece of hardware into structured, hashed, evidence-classified
benchmark receipts. It runs BEFORE any Compute Proof of Value
record may move toward validator approval or public-safe deed
issuance. The bench produces four orthogonal grades (Identity
Confidence · Health · Utility · Evidence) · never a single
composite score. Pipeline: CAPTURE → DIAGNOSE → BENCHMARK → HASH
→ CLASSIFY → ANALYZE → VALIDATE → DEED. Per-tier profiles
(E0-E7) keep workload tests appropriate to the asset class · a
Jetson is not benched as a workstation GPU is not benched as an
H100. Phase A (read-only identity + system + runtime + idle-health
capture) is implemented as `defendable-compute inspect`. Hard
operational rule: no benchmark may run on a GPU currently rented
or serving a production workload · the CLI uses
`nvidia-smi --query-compute-apps` as the pre-flight refusal gate.
Full doctrine in DEFENDABLE_COMPUTE_BENCH.md + receipt schema in
COMPUTE_BENCH_RECEIPT_SCHEMA.md + grading in
COMPUTE_ATTESTATION_GRADING_STANDARD.md + per-tier profiles in
COMPUTE_BENCH_PROFILE_MATRIX.md + CLI spec in
DEFENDABLE_COMPUTE_BENCH_CLI_SPEC.md + tool eval in
COMPUTE_BENCH_TOOLING_AUDIT.md + decision schema in
BEST_NEXT_USE_DECISION_SCHEMA.md (all in the platform repo).

## Evidence Vault doctrine

The Evidence Vault is the controlled storage architecture behind
DefendableOS records. Private evidence remains private; approved
public-safe manifests support issued Defendable Deeds. Storage
preserves artifacts, hashing records captured versions, and
validation determines claimability. Doctrine: the vault stores the
evidence · the hash protects the trail · the validator judges the
claim · the Deed publishes only what survives review. The 4-bucket
privacy model is enforced at the service boundary via a
public_export_or_refuse() guard with no exception path. Technical
specifics live in OBJECT_STORAGE_POLICY.md and
EVIDENCE_VAULT_OBJECT_STORAGE_DOCTRINE.md in the platform repo.

## Open infrastructure (/open)

DefendableOS publishes its full operational substrate so claims are
auditable, not asserted. The /open page documents:

- Models in production:
  - Kimi K2.6 (Moonshot) — primary AIOV narrative under typed contracts
    VALIDATOR_FLAG_TOOL + RESEARCH_CLASSIFY_TOOL
  - OpenAI gpt-4o — secondary classification under RESEARCH_CLASSIFY_TOOL
  - Brave LLM Context — live web grounding · feeds Validator Check 5
- Doctrine guardrails enforced in code:
  - Server-side AI downgrade guard (CONFIRMED_SALE_PRICE auto-downgrades
    to MARKET_COMMENTARY without confirmed-sale signal in rationale)
  - Affirmative-claim scan (Validator Check 10)
  - Verbatim-disclaimer integrity (Validator Check 11)
  - 11-class signal-class doctrine (only PERMISSIONED_CONNECTED_SALE +
    FIRST_PARTY_DEFENDABLE_SALE count as confirmed)
- Object storage · 4-bucket privacy vault:
  - PRIVATE_EVIDENCE, MARKET_OBSERVATIONS, DERIVED_DATASETS, PUBLIC_ASSETS
  - public_export_or_refuse() guard raises CompFoundryError on any
    non-public bucket export
- Connector honesty matrix · 19 connectors with real status
- Open source repos:
  - github.com/SudoSuOps/defendableos (platform · public source)
  - github.com/SudoSuOps/defendable (landing · public source)
  - github.com/SudoSuOps/legalsniper-plugin (Apache 2.0)

## Disclaimers

DefendableOS records are evidence and analysis packages. Asset-specific
professional, legal, regulatory, or licensed appraisal requirements
may still apply.

Operator-stated asking prices and operator-stated STNL terms are
operator claims — not validator-issued conclusions or confirmed-sale
comparables. The deed JSON declares this on every record.

The Palm Grove Marketplace and any property described as "illustrative"
are demonstration assets and not active offerings.

## Related surfaces

- https://swarmandbee.ai/ — the firm (Swarm and Bee LLC · DBA Swarm & Bee AI)
- https://defendable.eth.limo/ — defendable.eth · cross-Swarm certification standard
- https://legal.swarmandbee.ai/ — Legal-as-a-Service (the Legal module)
- https://legalsniper.swarmandbee.ai/ — LegalSniper model card
- https://api.defendableos.com/ — platform API (read-only public endpoints
  at /api/v1/public/verify/{slug} and /api/v1/public/lookup?hash=...)

## Entity

DefendableOS is a product of:
Swarm and Bee LLC · Florida LLC · D-U-N-S 138652395 ·
doing business as Swarm & Bee AI.

## Contact

build@swarmandbee.ai
hello@swarmandbee.ai
https://defendableos.com/#early-access