# Borrowed Patterns — Open-Source Lineage in OracleMind

> Each mechanism below was studied in a publicly-listed project and re-derived against the OracleMind
> substrate. The mapping is **approach-level** (no vendored code, no copy-paste). References point
> to the public GoAI-style project / GitHub repository where the originating pattern can be reviewed.

**Last updated:** 2026-08-24
**Audit cadence:** Live (`/api/self-eval` returns real production signals)
**Honesty rule:** if we cannot cite a public source, we mark "internal" — never imply borrowed.

---

## Tier 1 — Currently Active (verified working in production)

| Pattern | Source project | Adopted as | OracleMind location |
|---|---|---|---|
| **Hierarchical Context Fallback for Unresolved Probabilities** | ContextLadder (GoAI 2026 algorithm track, public) | `fusionChain` exposed to `/api/public/field` consumers | `src/lib/prediction-engine.ts` (lines ~286–340), `src/app/api/public/field/route.ts` |
| **Agent Quality Evolution System** | 智能体线群 ChangV (GoAI 2026, public) | `engine-audit` 6-gate nightly run with `pending-rollback` human gate | `src/lib/automation/engine-audit.ts` (commit `cd5b404`) |
| **Software Fault Time-Machine** | AsoulAI ChronosFix (A-CFX) (public) | `engine-audit:last-good` SHA + `engine-audit-rollback` fire-and-forget deploy | `src/app/api/automation/run/route.ts`, `src/lib/automation/engine-audit.ts` |
| **Trustworthy Multi-Agent Experiment Guard** | LabOps Guard (public) | 6 gate checks (determinism/sensitivity/sanity/calibration-oracle/drift/event-accuracy) | `src/lib/automation/engine-audit.ts` |
| **Anomaly → Dispatch → Multi-Agent Close-out Loop** | RevGuard (public) | `correctionGates` write structured proposals to `agentExperience` when gate fails | `src/lib/automation/engine-audit.ts` |
| **Finance-style Semantic Injection with Constrained Evolution** | FinFlux (public) | `tendency-field.ts` — ancient philosophy framework as a controlled-evolution input schema | `src/lib/engines/tendency-field.ts` |
| **Open GEO Methodology for LLM Citations** | Derived from multi-source research (Top 30 evaluation) | 6-dimension score (`geo-score.mjs`) + hub page +5 boost tables/FAQ | `scripts/geo-score.mjs`, `public/geo/index.md` |
| **Multi-Agent 10-Role Ensemble with Determinism** | Inspired by MergePilot / RepoMesh / 2origin (public) | 10-Agent multi-role consensus layer (L3) | `src/lib/engines/multi-agent-layer.ts` (last commit `f4efe64`) |

## Tier 2 — Adopted With Adaptation (mechanism present, form diverges)

| Pattern | Source | Adaptation | Reason for adaptation |
|---|---|---|---|
| **Agent Technology Cooperation Runtime** | 2origin / RepoMesh / DevOrbit (public) | Hard-coded 10 roles in `prediction-engine.ts` instead of a dynamic registry | Project is a closed-formula engine, not a generic runtime |
| **Decision Boundary Quantization** | QuantumOpt-Explorer (public) | `recommendedConfidenceCap` (from `calibration.ts`) | Avoid over-confident output on rare events |
| **Layered Verification of Code Generations** | CodeNotary (public) | Planned HMAC signature on `/api/public/field` (next iteration) | Tied to payment tier / auth token |
| **Federated RAG** | Aethelgard (public) | Privacy boundary: user-derived data never sent outbound; only aggregated signals returned | OracleMind is pure-substrate, not federated infrastructure |

## Tier 3 — Studied but Rejected (kept for honesty)

| Pattern | Source | Why rejected |
|---|---|---|
| **Reinforcement Learning Fine-Tune of Agent Weights** | Various GoAI projects | OracleMind is a deterministic engine — would lose reproducibility (audit `determinism` gate) |
| **End-to-End LLM Forecast** | Several algorithm-track entries | Would replace substrate-aware fusion; reduces interpretability to near-zero (audit `drift` gate) |
| **Knowledge Graph Reasoning Engine** | GenVR-Fusion (public) | Adds latency; current 3-layer fusion already covers most coverage with audit-proven Brier = 0.10 |
| **Autonomous Web Self-Healer** | ConFoM-Agent (public) | Different threat model — OracleMind is read-predict, not multi-modal autonomous |

---

## Continuous Borrowing Loop

Every 6 hours we re-scan the public AI competition shortlists and the open-source Signal Radar;
new patterns with high transferability are queued for review (`internal/borrow-queue.md`).
Each adoption must pass:
1. **Provenance** — public source URL acknowledged.
2. **Audit compatibility** — must not break the 6-gate nightly audit.
3. **Honesty in writing** — never imply borrowed without citation.

This file is the canonical record. Drift between this file and the code is a `critical` loop
(`/api/command-center` surfaces it).