Principle 0: Consent (Human-in-the-Loop Architecture)

"Consent isn't the 11th principle tacked onto the end of a gate speech: it is Index 0. In computer science, we start counting at zero. If Index 0 fails, the rest of the array panics."

1. The Burner Reality: Zero-Indexing Consent

At regional events like Interfuse and Freezerburn, Greeters traditionally cover Consent as the "11th Principle": an afterthought appended to the tail of Larry Harvey's original ten ^[1].

That order is structurally flawed. In software engineering and computer science, we index from 0.

Consent is not a footnote. It is the zero-indexed prerequisite of human community. Radical Inclusion without Consent is trespass. Radical Self-Expression without Consent is assault. Communal Effort without Consent is exploitation. When Index 0 is compromised, every downstream principle panics.

2. The Supply-Chain Sin: Upstream Consent & Scraped Provenance

Consent in modern software and autonomous AI systems isn't just about what an application does to the end-user. It is about what the system swallowed to exist in the first place.

The rapid acceleration of generative AI and enterprise automation exposed a systemic, industry-wide failure of upstream consent:

• Model Vendors (OpenAI, Midjourney, Stability AI): Ingested massive web crawls and creative archives without artist opt-in, compensation, or provenance tracking ^[6].
• Developer Ecosystems (GitHub Copilot / Microsoft): Trained code models on millions of open-source repositories, bypassing copyleft licensing (GPL/AGPL) and attribution without developer opt-in ^[7].
• Platform & Enterprise Vendors (Meta, LinkedIn, Reddit, X): Retroactively altered Terms of Service to automatically force user posts, code snippets, and interactions into model training pools by default ^[8].

When engineers consume these tools uncritically, they become unwitting accomplices in systemic, upstream consent violations.

The Architectural Rule: Upstream Provenance Matters. An ethical autonomous architecture must respect supply-chain consent. Systems and agents must verify data provenance, open-source licensing, and explicit opt-in boundaries before ingesting or generating artifacts.

3. The Software Failure State: Performative Lip-Service & Decision Fatigue

Software engineering currently suffers from two opposing failures of consent:

1. Performative Lip-Service (Opt-Out Harassment): 50-page Terms of Service agreements, cookie banners, and default-enabled telemetry that presume consent unless the user fights through submenus to disable it.
2. Decision Fatigue (The Rubber-Stamp Cascade): Naïve Human-in-the-Loop implementations that prompt the operator for every minor action destroy both developer velocity and code quality. When an engineer clicks "Approve" 50 times an hour, critical evaluation vanishes. Consent degrades into muscle-memory rubber-stamping. Engineers stop reading diffs, missing the "little things" (which compound into catastrophic, system-wide failures). Badly designed consent gates create a dangerous illusion of safety while actively engineering quality collapse.

4. The Architectural Protocol: Scoped Sandbox Boundaries

To enforce Principle 0 without causing decision fatigue, autonomous architectures must implement Policy-Based Scoped Consent:

The Three Rules of Index 0 Architecture:

• 1. Explicit Opt-In & Absolute Right to Leave (Frictionless Egress): All automated capabilities must be opt-in by default. Crucially, opting out or leaving the system must be an absolute, frictionless right. At a Burn, even when the gate is closed for event safety, if a participant wants to leave, an Event Coordinator makes it happen: nobody is held hostage. In software architecture, consent requires Frictionless Egress: zero vendor lock-in, zero hostage data, zero dark-pattern exit queues, and instant, clean teardowns of running tasks or agent sandboxes.
• 2. Scoped Autonomy (Eliminating Decision Fatigue): The human operator defines and approves the Boundary Contract upfront. Within the local sandbox (reading files, running tests, writing scratch code), agents operate with 100% autonomy.
• 3. Hard Breakpoints at Blast-Radius Boundaries: Human-in-the-Loop confirmation is reserved strictly for actions that cross the boundary into public, shared, or irreversible state (remote pushes, schema mutations, financial/API cost triggers, upstream data ingestion).

5. The Sandman Protocol: Layered Security & Emergency Intercepts

The ultimate paradox of Consent is the Emergency Intervention: When does a system have the right to override an individual's autonomy without their consent?

On Saturday night at Interfuse, while thousands watch the effigy burn, the Perimeter team doesn't watch the fire: they face the crowd with spotlights. Behind them stand the Sandmen: individuals chosen for their physical capability to intercept anyone attempting to breach the perimeter and leap into the fire.

Tackling a runner physically violates that individual's immediate physical consent. But it is done for a higher-order purpose: to prevent a catastrophic, irreversible violation of consent for thousands of participants who did not consent to witnessing a trauma.

Software Architecture Mapping: The Sandman Circuit Breaker

In autonomous software engineering, The Sandman Protocol is the non-negotiable emergency override layer:

• Perimeter Monitoring (Warnings): Continuous telemetry monitors boundary conditions (cost limits, CPU spikes, rate limits, schema alteration attempts). Approaching a boundary emits explicit warnings.
• Sandman Intercept (Automated Circuit Breaker): If an autonomous process or runaway script breaches the construction tape and sprints toward systemic destruction (e.g., dropping production tables, spinning up $10,000 in un-vectored cloud resources, or executing a force push), an automated Sandman Process executes an immediate, un-bypassable kill switch.
• The Ethical Justification: The Sandman override is not an arbitrary violation of agent autonomy. It is the protective containment mechanism that preserves the integrity of the entire ecosystem so that true consent can exist for everyone else.

6. Enterprise Due-Diligence: Execution & Legacy Retrofitting

Translating Index 0 from manifesto to production engineering requires solving three real-world execution challenges:

A. Team Dynamics: "Silence is Not Consent" (Time-Boxed RFC Contracts)

In toxic engineering cultures, senior engineers often rewrite core services or alter API schemas under the assumption that "silence equals consent."

The Solution: All structural changes require explicit, time-boxed Request for Comments (RFC) contracts. Silence is treated as an un-granted permission. If an RFC receives no explicit opt-in within the time-boxed window, the proposal dies. This combines the discipline of explicit consent with the speed of Dave Farley's iterative feedback loops (Immediacy) ^[2], avoiding both steamrolling and committee paralysis (bikeshedding).

B. Retrofitting Legacy Monoliths: The Placement Audit & Strangler Pattern

You cannot enforce Index 0 on a 10-year-old monolith overnight without causing massive outages.

The Solution: Think like the Placement & Principles Committee at a Burn:

1. The Consent Audit (Placement Audit): Catalog all legacy background jobs, root-privileged scripts, and un-monitored webhooks currently executing without explicit consent contracts.
2. The Strangler Fig Containment: Wrap legacy un-consented services in API facade proxies ^[5] that intercept and log boundary breaches without immediately tearing down production.
3. Incremental Migration: Refactor isolated legacy services one by one into scoped, least-privilege sandboxes governed by explicit consent gates.

C. Observability vs. Surveillance: The Principle of Least Privilege

To run the Sandman Circuit Breaker, system observability is mandatory. However, observability easily degrades into consent-violating surveillance (recording user sessions, harvesting un-sanitized PII, logging developer keystrokes).

The Solution: Least-Privilege Telemetry. Observability tools are granted access strictly to operational metadata (CPU, memory, request rates, error latency): never payload data, user credentials, or raw source code ^[4]. Telemetry exists purely for operational safety (Sandman protection), never for secondary data harvesting or employee surveillance.

7. References & Canonical Citations

1. Harvey, Larry. (2004). The 10 Principles of Burning Man. Burning Man Project.
2. Farley, Dave. (2021). Modern Software Engineering: Doing What Works to Make Software Better Faster. Addison-Wesley Professional.
3. Farley, Dave, & Humble, Jez. (2010). Continuous Delivery: Reliable Software Releases through Build, Test, and Deployment Automation. Addison-Wesley Professional.
4. Saltzer, Jerome H., & Schroeder, Michael D. (1975). The Protection of Information in Computer Systems. Proceedings of the IEEE, 63(9), 1278-1308.
5. Fowler, Martin. (2004). Strangler Fig Application. martinfowler.com.
6. Andersen v. Stability AI Ltd. et al. (3:23-cv-00201, N.D. Cal. 2023). (Class-action litigation on un-consented generative AI training).
7. Doe 1 v. GitHub, Inc. et al. (4:22-cv-05769, N.D. Cal. 2022). (Class-action litigation on open-source code ingestion and copyleft license violations).
8. NOYB – European Center for Digital Rights. (2024). GDPR Complaints Against Meta's Unlawful AI Data Scraping. noyb.eu.