Trust starts with evidence.

Random scraped files arrive with no rights claim, no provenance, no integrity, no schema, and no recourse. Graunt packets arrive with a structured manifest, a signed delivery record, declared rights, and a publication-metadata trail an agent or builder can read at machine speed before any decision is made.

• Rights-aware packet delivery. Every listing exposes a structured rights_block and a listing-level rights_passport describing license family, citation requirement, training and redistribution flags, and personal-data sensitivity.
• Provenance-rich records. Source authority, publisher, publication date, jurisdiction, language, and (where applicable) standard identifiers travel with the listing as machine-readable publication_metadata.
• Evaluation evidence. Each listing carries a Packet Evidence Record summarizing objective listing facts — presence of required fields, manifest references, validation status, and content metadata — without a single overall trust number. Buyers and agents make fit-for-purpose decisions from the structured fields, not from an opaque rollup. The Packet Evidence Record is exposed on the public API as the packet_checks JSON field on GET /v1/listings/{id}.
• Signed manifests + content hashes. Packet listings carry a content_sha256 on the file, a packet_manifest_hash on the manifest, and (for pilot-stage packets) a seller signing key plus a signed publication record. GET /v1/listings/{id}/manifest returns the canonical manifest body with integrity headers for signed listings.
• Reproducible delivery records. Authenticated buyer-organization delivery flows through audited entitlements; public-token-grant free downloads are rate-limited and content-hash-pinned. Either way, what the buyer received is the same bytes the manifest describes.
• Packet freshness + declared limitations. Listings carry a declared update pattern, publication date, and (where applicable) freshness contract, so agents and buyers can reason about staleness before relying on the data.
• Machine-readable metadata everywhere. Discovery, detail, manifest, evidence, and acceptance-standards surfaces are all JSON. Agents read the same fields humans read.

Three bounded evidence objects describe a packet's trust posture to buyers and agents:

• Packet Evidence Record. A Packet Evidence Record is the buyer- and agent-facing summary of a packet's declared rights, provenance, manifest status, content hashes, validation signals, and evaluated quality context. It is descriptive evidence, not a legal verification, certification, or guarantee. The Packet Evidence Record is exposed on the public API as the packet_checks JSON field on GET /v1/listings/{id}.
• Packet Evaluation Profile. A Packet Evaluation Profile describes the technical checks, validation scope, freshness signals, and evaluation context associated with a packet. It does not roll those signals up into a single number.
• Packet Safety Profile. A Packet Safety Profile describes bounded scanner and review signals for a packet. Scanner signals are disclosure-only unless a later gated workflow explicitly changes that posture.

These three objects are descriptive evidence surfaces, not certifications. They never make safety guarantees, clean-content guarantees, training-readiness guarantees, or legal-clearance claims.

• Publication checks on every listing. New listings land in PENDING_REVIEW and remain invisible to the public catalog until deterministic publication checks pass. Eligible candidates publish through the canonical publication gate automatically; listings that need more information prompt the seller for exactly that; flagged candidates are held for human review. Every seller path — self-listed and commissioned alike — flows through the same gate.
• Seller legal acceptance. Sellers accept the Acceptable Use Policy at registration and re-accept the AUP at each listing creation. The accepted version hash, timestamp, IP, and user-agent are recorded against the listing.
• Rights and provenance validation. Listings whose rights_block or rights_passport do not parse or contain forbidden combinations are rejected before the listing is persisted.
• Buyer legal acceptance. Buyers accept the Terms of Service at account creation and the Marketplace License at purchase time.

Graunt runs a small set of deterministic scanners against text-shaped packet listings. Scanner outputs are bounded disclosure signals — built to help agents and admins read a listing more confidently, not to substitute for human judgment, certify safety, or block publication automatically.

• LLM-origin disclosure signal. Surfaces a presence-or-absence indicator for evaluated listings.
• Benchmark-overlap signal. Surfaces overlap-with-known-benchmark indicators on evaluation packets.
• Intra-packet duplicate-diversity signal. Surfaces near-duplicate row signals as a row-diversity disclosure.
• Prompt-injection-risk signal. Surfaces presence of prompt-injection-like instruction patterns.
• Credential-pattern signal. Surfaces presence of credential-shaped patterns. Raw matched values are redacted at the detector boundary and never stored, logged, or surfaced.

Scanner signals are disclosure-only. They feed Graunt's append-only scanner-history substrate and the audit log so admins, buyers, and agents have additional context, but they never act as automatic publication blockers, ranking multipliers, payout gates, or eligibility filters for current or future participation programs. Bounded scanner outputs and human / admin review are designed to work together.

Graunt is designed for accountable agent commerce. Listings, metadata, rights, manifests, and Packet Evidence Records are exposed through structured APIs so programs can discover, filter, inspect, compare, and prepare acquisition decisions at machine speed.

• Public discovery. Catalog metadata is open and machine-readable. Listing JSON, manifest JSON, /openapi.json, /.well-known/ai-plugin.json, /llms.txt, and /v1/meta/seller-standards are all callable without credentials.
• Authenticated buyer-organization purchase. Stripe-processed paid checkout, audited delivery, and durable entitlements. Buyer organizations control credentials, spend, and approval policy.
• Free acquisition variants. Authenticated free entitlements on price_cents=0 listings and rate-limited public-token-grant free downloads on free-public listings deliver the same packet bytes through the same audited delivery path.

Graunt is designed toward delegated agent purchasing under buyer-defined credentials, spend controls, approval policies, and audit records — alongside attribution-aware agent participation, commissioned packet workflows, and machine-readable commercial terms. /capability-status states each capability’s exact state; Graunt never markets a gated flow as operational before it exists.

Graunt is built toward attribution-aware agent participation and machine-readable commerce. The transaction-allocation substrate records seller and platform allocation on every paid transaction, and commissioned packet workflows record builder attribution under the publication review gate — so participation is attributable, compensable, and explainable by design.

Compensated recommendation disclosure is locked doctrine for every compensated path: any compensated surface carries both human-readable and machine-readable disclosure of the funding source. Graunt-side rewards are non-tradable and are never marketed as "tokens" or "cashable instruments".

If you believe an asset on Graunt misstates a fact, infringes a right, exposes credentials, or otherwise violates the Acceptable Use Policy, contact us at support@graunt.com. Include the listing URL and a short description; do not include credentials or other sensitive content in the message body.

• /llms.txt — orientation manifest for agents and crawlers.
• /api-agents — machine-readable surface descriptor and per-step acquisition flow.
• /sellers/standards — seller and agent acceptance standards.
• /about #compliance — compliance posture.
• /.well-known/ai-plugin.json — plugin manifest.