Controls: Deployable safeguards for human flourishing
Balance the Triangle Labs turns insight into practice: small, reusable controls you can run in minutes to reduce risk when technology, human pressure, and governance lag collide.
These controls are designed to be:
- Practical (5–15 minutes)
- Pressure-aware (built for real conditions, not ideal ones)
- Portable (usable without special tooling)
- Auditable (leave behind a trace of use)
Featured Control: HPC-1 — Human Pressure Check
If you only download one control, start here. HPC-1 is a lightweight, pressure-aware safeguard that prevents “help” from turning into harm—whether you’re deploying AI or making consequential human decisions.
Jump to
HPC Stack • Stop Ladder • Assurance Before Scale • Provenance & Claims
The HPC Stack
Human Pressure Controls for decisions that matter
When people—or AI systems—operate under pressure, judgment collapses before intent is examined.
The HPC Stack (HPC-1 → HPC-5) is a practical control system that restores agency, preserves boundaries, and keeps accountability intact—before harm occurs.
The Five Cards
HPC-1 — Human Pressure Check
Detect pressure. Invoke restraint.
Identifies Cognitive, Emotional, Authority, and Dependency pressure (including stacking) and triggers the correct response: SLOW / STOP / ESCALATE—for AI deployments and human-only decisions alike.
HPC-2 — Escalation Path Integrity
Make escalation reduce risk—not hide it.
Ensures escalation preserves context, de-amplifies pressure, assigns ownership, and avoids rubber-stamping decisions under urgency.
HPC-3 — Authority Boundary Clarity
Name authority. Bound it. Anchor responsibility.
Prevents role confusion and overreach by requiring explicit mandates, scopes, and responsibility—so advice never masquerades as permission.
HPC-4 — Dependency Boundary Protection
Support without replacing autonomy.
Guards against exclusivity, role creep, and over-reliance by reinforcing agency, maintaining role stability, and making off-ramps visible.
HPC-5 — Evidence & Accountability Continuity
Make decisions defensible over time.
Captures owner, rationale, evidence, alternatives, and handoff context with integrity—so organizations can reconstruct and defend what was decided, and why.
How It Works
Detect → Restrain → Escalate safely → Bound authority → Prevent dependency → Preserve evidence
Stop Ladder Card (SL-1)
Use when: you need a simple, shared rule for when to slow, stop, or escalate.
What it does: standardizes the escalation pathway so it’s not improvised under stress.
“Assurance Before Scale” Checklist (AS-1)
Use when: you’re about to expand an AI assistant to more users, more autonomy, or higher stakes.What it does: forces proof questions: scope boundaries, failure modes, human handoff, logging, and rollback.
Provenance & Claims Test (PC-1)
Use when: an AI output is being treated like a fact, diagnosis, legal claim, or policy guidance.What it does: separates “helpful synthesis” from “unsupported claim,” and prompts verification steps.
Controls by scenario
Deploying AI assistants
PC-1 — Provenance & Claims Test
How to use this library
- Pick the control that matches your situation
- Run it once with a real decision in motion
- Save the trace (one sentence is enough)
- Repeat under pressure until it becomes habit
Want the “enforced” version?
These controls are designed to stand alone.
SpiralWatch™ implements the same logic as fail-closed, auditable assurance for AI systems operating under human pressure.
Get updates
New controls are added as the library grows.
BALANCE THE TRIANGLE LABS 