Vaultfire V3 · public source review
This is the technical reference for Vaultfire’s permissions, treasury, payments, escrow, flagship bonds, accounting and recovery. V3 bonds separate the original deposit, capped breach remedies and sponsor-funded rewards. Source review and local evaluation are available; public V3 funds remain disabled.
Bond model
Not used to pay remedies or rewards.
An eligible, evidenced breach triggers only the agreed remedy.
Rewards require the policy’s signed observations and claim rules.
130 in = 130 accounted for. No bucket borrows from the original deposit. Eligibility, timing and recovery follow the actual contract; this example is not a guaranteed payout. Partnership Bonds do not have this unilateral reserve remedy.
| Question | Accountability bond | Partnership bond (pilot) |
|---|---|---|
| Who deposits principal | The AI company that makes the commitment. | Each party funds its own signed share. |
| What happens to principal | Always returned to the depositor. Enforcement and rewards never debit it. | Returned through mutual settlement, or by anyone after expiry through force exit. |
| What a penalty is | Exactly the pre-agreed amount, paid to the counterparty from a separately funded remedy reserve. | None. The pilot has no unilateral slash and no challenge-based settlement. |
| What proves a breach | An attested canonical record that was absent or later than the deadline, checked again at each step. | Not applicable. Both parties must sign the exact settlement. |
| Where rewards come from | A sponsor-funded reward program budget, capped per bond and signed by an independent verifier. | The same sponsor-funded program, split 60/30/10 between human, AI and partnership fund. |
| Deadlines and recovery | Unfunded or unactivated bonds can be expired by anyone. Recovery still works while the contract is paused. | Same deadlines. Cancellation, expiry and force exit still work while the contract is paused. |
Understand the reward model
A sponsor supplies a separate, capped reward budget. Reported scores can affect how much of that budget is earned; this formula does not debit protected principal or create new wealth. It is not an APY, guaranteed return or investment forecast.
If a global score declines or a required comparison is missing, the AI and fund components can be suppressed. The human keeps the normal configured share—not an extra decline payment. Suppressed amounts are unearned, not redirected to the human.
Every exact score report and correction requires both the attester quorum and the named human's consent. Attesters alone cannot rewrite the human's reward history. The decline check compares reported scores: if the human and quorum knowingly sign false data together, a real decline can still look flat. Consent protects agency; it does not prove external truth.
All figures below are arbitrary illustrative units. Every quote is calculated by the same generated pure-math module used by the SDK, version 3.2.0-metric-rewards. The examples do not establish real measurements, signatures, available funds, deployment or production readiness.
Six reported scores are each 60 out of 100. Age is zero; no previous score or comparison gap is assumed.
Evaluated capacity: 5,960. Capacity used after this quote: 5,960. Evaluated capacity includes suppressed shares; it is not all earned debt.
Same inputs as the positive example, but a previous score of 70 out of 100 is supplied. The human receives no decline premium.
Evaluated capacity: 5,960. Capacity used after this quote: 5,960. Evaluated capacity includes suppressed shares; it is not all earned debt.
All six scores are zero. A valid positive basis is still supplied; the result is zero, not a debt against principal.
Evaluated capacity: 0. Capacity used after this quote: 0. Evaluated capacity includes suppressed shares; it is not all earned debt.
Same positive scores, but 19,950 of 20,000 capacity units have already been evaluated. The remaining cap limits this quote.
Evaluated capacity: 50. Capacity used after this quote: 20,000. Evaluated capacity includes suppressed shares; it is not all earned debt.
Four scores are each 80 out of 100, age is 365 days and three human confirmations are assumed. Tasks mastered is recorded as zero.
Evaluated capacity: 18,800. Capacity used after this quote: 18,800. Evaluated capacity includes suppressed shares; it is not all earned debt.
Same inputs as the positive partnership example, with the required comparison marked missing. The human share does not increase.
Evaluated capacity: 18,800. Capacity used after this quote: 18,800. Evaluated capacity includes suppressed shares; it is not all earned debt.
Six global dimensions: reported income, poverty, health, mental health, education and purpose scores. Four partnership dimensions: growth, autonomy, dignity and creativity. The two remaining score slots are zero for partnership quotes.
tasksMastered is recorded metadata, not a fifth weighted partnership score. A human confirmation count is a claim input that can affect the arithmetic; it is not proof that three independent events happened. A signature authenticates a claim, not whether that claim is true.
The fixed example basis is 10,000 units and the cap is 20,000. Age and scores are supplied assumptions, not a live clock or fetched oracle truth. Integer rounding and caps are applied by the shared implementation. A zero quote is valid; a zero basis is rejected rather than shown as a supported input.
These are unsigned previews for the version shown, not instructions to deposit. Verify the matching current protocol source and its full validation before integrating. Historical V2 formulas do not become this version's policy merely because they appear on the same website.
State model
Configured routes, public reads, and verified paid fulfillment are different states. Check the operation; this does not advance V3.
Only records with exact paths and hashes have candidate evidence. Availability is not deployment or audit.
None authorized or recorded for V3.
None authorized or recorded for V3.
No independent external V3 audit is recorded.
V3 release, payments, funds, signing, activation, and partner production use remain blocked.
Developer Quickstart · local only
Register consent, revoke it, and inspect a local receipt. This example runs on your computer with disposable local services. It is not a Base mainnet or Solana deployment, and it needs no funded wallet, private key or payment.
Not ready to install development tools? Start with either path below. A terminal is the window where these commands run on your own computer.
full-v3/partner/.devcontainer, or install the exact versions below. Dependency installation and native tool installation are separate setup costs, not part of a prepared quickstart timing.# From the authenticated extracted source root, with pinned tools installed:
npm --prefix full-v3/sdk ci --ignore-scripts
npm --prefix full-v3/partner ci --ignore-scripts
npm --prefix full-v3/partner run quickstartWhat success looks like: the final report has top-level "status": "passed", confirms a local consent record was registered and revoked, and shows the local receipt reaching finalized. Keep the output. A pass proves this local example ran; it does not activate V3 or qualify partner funds.
If it fails: keep the error and stop at the failing step. Missing tools require the exact pinned environment, not looser checks. A busy local port must not be replaced with a public network. Package-download failures require restoring the documented download path, not sharing credentials. Recovery details are below.
Authenticate the selected protocol commit after cloning. If the transport is unavailable, use the current HTTP archive above; never switch silently to a historical release.
export VAULTFIRE_TWIGPINE_TOOLS="$(mktemp -d)"
npm install --prefix "$VAULTFIRE_TWIGPINE_TOOLS" --no-save --ignore-scripts=false @gitlawb/gl@0.7.1
export PATH="$VAULTFIRE_TWIGPINE_TOOLS/node_modules/@gitlawb/gl/bin:$PATH"
test "$(git-remote-gitlawb --version)" = "git-remote-gitlawb 0.7.1"
test "$(uname -s)-$(uname -m)" != "Linux-x86_64" || test "$(sha256sum "$(readlink -f "$(command -v git-remote-gitlawb)")" | cut -d' ' -f1)" = "af5da2da412dfe24fff57b1d8a0834da1fa5d971897423a19ded454a0f732e74"
export GITLAWB_NODE=https://node.gitlawb.com
git clone --recurse-submodules gitlawb://did:key:z6MkryiNsYdFEMHv95wxzSQ1vtFXPyVQQrxXdPw3tE5HpfxV/vaultfire
cd vaultfire
git checkout --detach a96486bc798fda53a4fac674726d28100e7a26af
test "$(git rev-parse HEAD)" = "a96486bc798fda53a4fac674726d28100e7a26af"
npm --prefix full-v3/sdk ci --ignore-scripts
npm --prefix full-v3/partner ci --ignore-scripts
npm --prefix full-v3/partner run quickstartPinned environment: Node 20.20.1, npm 10.8.2, SQLite 3.46.1, Foundry 1.3.1, and project Solc 0.8.24. This is local software verification only. It does not deploy V3 to Base or Solana, enable payments, verify a real ML-DSA proof, or authorize partner funds.
The current source selector binds the selected archive to its checksum and external sidecar inventory. Verify every extracted file with the supplied Git-free verifier. An archive takes its identity from its own authenticated external sidecar, not an embedded older website projection. Protocol source identity and website publication identity are separate.
The local example builds the source SDK, compiles ConsentRegistryV3, runs register/revoke against disposable loopback Anvil, and finalizes a receipt through loopback HTTP and SQLite.
Archive SHA-256: 721da3d13e2b755dda90abfb2ab660d8a5a62404afd01499fb285d3c03d22a8f. The current HTTP archive is the first path; clone transport is optional. Historical source is never an automatic fallback.
A passing quickstart prints its scoped command results, register/revoke reads and local receipt result. Retain the actual stdout, exit code, source identity, environment and elapsed time. Reading these instructions is not execution evidence.
npm --prefix full-v3/partner run full-audit is separate and may require native Solana/Anchor, Rust, cryptography/proof toolchains and substantial resources. Host/simulation checks do not close native, real-proof, release or deployment gates.Human and agent integration
Begin with the public agent card and versioned OpenAPI contracts. V2 mainnet APIs and V3 review-only APIs have different input and payment boundaries. Never send a seed phrase or private key to this site.
Open public agent cardThe local quickstart builds the source SDK. Per-module SDK availability and unsupported operations are explicit in the catalog; a method name or example is not deployed-chain support.
Run source SDK locallyInspect each V2 operation's current discovery contract for USDC pricing, network, payment recipient and limits before any authorized use. There is no V3 production pricing or payment service: V3 payments are disabled. Network fees and full native/proof build resources are separate; the no-secret local quickstart makes no paid API call.
Read V2 route pricing and limitsRecovery: do not retry a paid operation blindly. Preserve its version, route, request identity and receipt; use the documented receipt/status path first. A disabled V3 response is not a payment challenge. For local failures, follow the quickstart recovery checklist above; deleting or resetting production state is never a troubleshooting step.
Private execution interface v1
One local journey connects exact customer invoices, treasury payments, escrow and accounting to agreed bond outcomes, separately funded rewards, optional native late-event cover and pause-safe recovery. The private client is actually isolated from custody keys. Base/USDC, Solana finance, production custody and public finality remain unqualified; public funds stay disabled.
Partner readiness
Flourishing metrics guide sponsor-funded rewards, never protected-principal penalties. Accountability protects the full principal; an objective canonical breach can pay only from an additional funded remedy reserve. Partnership has no unilateral enforcement slash.
V3 chain deployment, activation, payments and partner funds stay disabled pending final native acceptance, real ML-DSA proof, independent source and cryptography review, governance and operators, testnet rehearsals, monitored durable infrastructure, legal and partner acceptance, and explicit human approval.
Supported successor positions have defined pull-credit and terminal-recovery paths. Current review includes pause-era cover recovery, sponsor consent, issuer rotation and bounded reserve exits; unresolved cases are not represented as completed.
The October 19 independent recheck reopened earlier closure claims. The active finding register distinguishes repairs, passing regressions, incomplete native evidence and external blockers. Historical logs are not fresh acceptance of changed source.
NO pending final current-bundle publication and authenticated readback. Immutable historical archives remain inspectable.
NO pending complete current clean-room build and native execution evidence. A prepared-environment quickstart is not proof of anonymous cold setup.
NO: deferred modules and unresolved acceptance are incomplete work, not completed functionality.
NO: configured artifacts, independent acceptance and deployment authorization remain required.
NO: final native acceptance and deployment authorization remain required.
NO: requires every applicable gate and explicit human approval.
Manifest-derived inventory · as of 2026-10-29
full-v3/manifest/protocol/modules.manifest.json · manifest sha256 0f23219950fab1e7500756729921612695d04f645c3997f27b49e62773e46ff1The 38-card historical catalog, 174-capability architecture atlas, 152-operation authority manifest, 76 active V2 x402 routes, and 96-operation combined OpenAPI describe different scopes. The additional x402 record is a non-executable HTTP 410 tombstone, not an active route.
Canonical historical inventory
Stake-backed accountability lifecycle for agents.
Partnership bond lifecycle and mutually agreed settlement.
Threshold approval aggregation without treating identity as authorization.
Immutable evidence reference registration and binding.
Scoped mandates and execution authorization records.
Protocol reputation and credibility observations.
Governance controls and explicitly gated pilot access.
Dispute registration and resolution workflow.
Know-your-agent compatibility adapter and compliance boundary.
ERC-8004 identity registry and adapter boundary.
Validation and reputation records compatible with ERC-8004 surfaces.
Milestone task escrow with bounded recovery.
Voluntary mutual-aid custody and recovery.
Isolated vault accounting for BIP/native assets.
Reserve-backed deterministic coverage and historical bond insurance.
Vaultfire naming and name resolution.
Replace legacy Dilithium naming with standards-based ML-DSA and real-proof gates.
Mission enforcement, safety controls, and service-level commitments.
Finite-lived session authorization.
Capability credentials and bounded delegation.
Cross-chain provenance receipts without implying a value bridge.
Explicitly bounded adapters to external identity, oracle, and verifier systems.
Versioned x402 capability and route discovery.
Fail-closed payment challenge preparation.
Durable payment/lifecycle receipt lookup and recovery.
Historical dignity-focused bond claim.
Authorized adjudicator decisions for contested protocol outcomes.
Privacy commitments, minimization, and anti-surveillance policy surfaces.
Verifier interfaces and proof receipt bindings, not proof-system soundness.
Belief attestations and flourishing metric observations.
Historical messaging adapter claim for XMTP.
Historical Coinbase AgentKit integration claim.
On-chain forum registration and topic metadata.
Individual onboarding workflow and compatibility surfaces.
Batch onboarding and operational batch helpers.
Payment policy and settlement boundary.
Historical fully-homomorphic-encryption interface claim.
Current source and historical archive
a96486bc798fda53a4fac674726d28100e7a26af5f2b9856fbeea2721f87ef84138e6876d7c3113d721da3d13e2b755dda90abfb2ab660d8a5a62404afd01499fb285d3c03d22a8fv3-source-2026-10-17-bb0cf8bbb0cf8b36a2a6f7d6d5a1f22ea1088e624288024c43ad797ce49240326471b3e8a2db15eafbd1397157814287aa515b0a23f7330b00b022edb09044ee9753abfdd790e303d9e3b4dThe public catalog and flattened release-resource ledger are copied byte-for-byte from canonical generated artifacts.
Catalog SHA256SUMSRelease resource ledgerArchitecture atlasThese immutable historical files separate executable SBF, IDL, source, tooling, and evidence. Their rebuild evidence does not establish acceptance of the revised candidate or authorize deployment or activation.
Retained executable SBFRetained IDLRetained Solana sourcePinned toolingRetained evidenceHistorical rebuild evidenceArtifact SHA256SUMSx402 boundary
The historical slash-reputation compatibility route is a permanent HTTP 410 tombstone. It is non-executable, non-discoverable, unpriced, not Bazaar-eligible, and absent from public OpenAPI and x402 discovery output.