All outputs are constrained by an explicit evidence standard. The system discloses what is admissible for correlation and audit, and makes equally explicit what is excluded. This ensures reproducibility without exposing tunables or private data.Documentation Index
Fetch the complete documentation index at: https://none-38c466ad.mintlify.app/llms.txt
Use this file to discover all available pages before exploring further.
8.1 Evidence Standard
- Typed signals: every artifact is schema-typed before admission
- Provenance: each signal carries acquisition timestamp, source URL/domain, and runtime context
- Integrity: every output bundle includes verification digests to support replay and tamper detection
- Explainability: linkages are accompanied by rationale vectors listing contributing pivots, rarity adjustments, decay, conflicts, and conflict suppressions
- Replayability: correlation can be recomputed from preserved signals without access to proprietary orchestration logic
8.2 Disclosure Boundaries
The following classes of information are not disclosed:- Tunables and thresholds (exact rarity caps, decay rates, gate constants)
- Collection tactics (orchestration logic, stealth methods, bypass procedures)
- PII and sensitive payloads (user data, credentials, or content)
- Code implementations (algorithms are described by invariants, not source)
8.3 Scope Bands
- Corpus scale: hundreds of thousands to millions of domains
- Refresh cadence: high-activity cohorts refreshed on weekly cycles; lower-activity cohorts refreshed less frequently
- Cluster size: graphs partitioned stably at millions of nodes with modularity-based detection
- Acquisition wall-clock: bulk runs measured in hours to days depending on vertical size
- Cost/compute contours: reported in compute-hours and efficiency metrics, not monetary terms
8.4 Known Limits
- Blind spots: complete per-domain key rotation, aggressive client-fingerprint gating, and per-request bucket generation reduce persistence
- Non-goals:
- Deanonymizing individuals or personal accounts
- Content classification or semantic interpretation of page text
- Circumventing legal process for data access
- Failure modes: infra-only coincidences, CDN hub collisions, and placeholder templates are suppressed by design but remain known sources of attempted evasion