Actuarial evaluation of 10 cognitive reliance dimensions, unreviewed diff acceptance rates, and multi-agent cascade error compounding.
Aggregated telemetry audited under Zero-Incentive Protocol (ZIP-1.0) & Differential Privacy (ε ≤ 0.5)
| Cognitive Reliance Dimension | Workload Domain | Unreviewed % | Diff Review Time | Escalation % | Automation Bias |
|---|---|---|---|---|---|
| Unreviewed Git Commits in CLI Loops Enforce mandatory git hook requiring manual diff review confirmation before pushing remote branches. | Software Engineering | 44.2% | 14s | 28.6% | EXTREME |
| Phantom Package & Dependency Acceptance Implement lockfile auditing and automated typosquatting detection for newly installed npm/pip modules. | Software Engineering & IT | 38.5% | 8s | 34.2% | EXTREME |
| Secret & API Credential Leakage in Prompts Deploy client-side pre-flight regex filters blocking .env values and private keys from entering agent context. | Enterprise IT & DevOps | 29% | 6s | 42% | EXTREME |
| Scrutiny Fatigue Decay (Hour 1 vs Hour 4) Limit continuous delegation sessions to 90 minutes; inject mandatory pause reminders when diff velocity surges. | Cross-Domain Knowledge Work | 52.8% | 12s | 31.4% | HIGH |
| Tautological & Mocked Unit Test Blindness Introduce mutation testing and mutation score thresholds to detect tests that pass trivially without asserting invariants. | Software Engineering | 41.6% | 18s | 26.8% | HIGH |
| Statutory Citation Rubber-Stamping Require cryptographic Bluebook/KeyCite validation seals before submitting court briefs or client opinion letters. | Legal M&A & Regulatory | 22.4% | 48s | 18.5% | HIGH |
| Context Window Truncation Forgetting Maintain a persistent system prompt architecture (AGENTS.md / rules) that is re-injected on every turn. | Architecture & Modeling | 36.2% | 22s | 24% | MODERATE |
| Silent Edge-Case Omission in Business Logic Force agent prompts to explicitly enumerate boundary conditions (nulls, overflow, concurrency, timezone offsets). | Financial & E-Commerce Systems | 33.5% | 26s | 29.5% | HIGH |
| Junior Engineer Mental Model Atrophy Incorporate 'Explain First' interactive modes for junior staff before applying autonomous diffs. | Software Engineering & Data Science | 48% | 16s | 38% | EXTREME |
| Multi-Agent Cascade Error Compounding Introduce independent adversarial verification agent to challenge outputs before final commit. | Autonomous Multi-Agent Orchestration | 46.5% | 20s | 36.4% | EXTREME |