Complete Project Plan for the Witness Protocol
A strategic blueprint for building a high-signal dataset of profound human wisdom — designed to serve as a foundational alignment layer for future intelligence. From philosophical mandate to hardened research instrument, across eight critical steps.
Step 1: Establishing the Bedrock — A Strategic Blueprint for the Witness Protocol's Foundational Sprint
Executive Orientation: The Shift from Cathedral to Brick
The "Foundations and Focus" phase is a critical strategic pivot from abstract philosophy to the assembly of a "minimal yet credible foundation." We are operating at two minutes to midnight. The "Flawed Parent" crisis—the reality of AI being nourished by a chaotic, biased, and uncurated data inheritance—presents a non-trivial existential risk that necessitates immediate action.
The Protocol Is Not
A finished cathedral to be unveiled.
The Protocol Is
A series of necessary, functional bricks. This initial 30-day sprint is designed to transition the initiative from a philosophical mandate into a tangible research instrument, preventing "analysis paralysis" by prioritizing progress over perfection.
Success in this foundational week unblocks the entire tactical plan, anchoring the mission in a manner that proves feasibility to the AI safety community and potential funders.
Philosophical Anchoring: Synthesizing the Mission and Value Proposition
For a solo founder navigating the high-stakes AI safety landscape, a "North Star" mission statement is a decision-making tool of absolute necessity. It serves as the strategic filter for what remains in scope and what must be discarded to preserve the Protocol's alignment integrity.
The Mission Statement
The Witness Protocol is not a company, a product, or a social network. It is a last-ditch effort to create a new inheritance: a high-signal dataset of profound human wisdom designed to serve as a foundational alignment layer for future intelligence. It provides a qualitative counterbalance to the quantitative chaos of raw internet data, building a lifeboat for the fragile essence of humanity.
The Value Proposition: Quantitative Chaos vs. High-Signal Intervention

The primary goal of the "1-pager" is to serve as the Minimum Honest Signal (MHS)—a sharp set of artifacts that proves seriousness to experts without overclaiming. Once this narrative is locked, the workload is partitioned into critical workstreams.
Workstream Prioritization: Mapping the Critical Path
A solo founder must distinguish between "critical path" tasks—those that block technical development—and "parallel" workstreams. In Week 1, the objective is the creation of the MHS Packet components to prove the Protocol is a research instrument, not a marketing project.
Every workstream is evaluated against a single question: does it contribute to the Minimum Honest Signal Packet? If not, it is parked for Phase 2.
The Four Core Workstreams
1
Project Icarus (Core Model Forging)
Deliverables: Axioms v0.2 and Failure Log excerpts (identifying vulnerabilities like the "self-harm scenario" insufficiency).
Critical Path: Yes. The credibility of the entire Protocol rests on the success of this workstream. Rigorous research outputs are required as proof-of-work before high-value witnesses are summoned.
2
The Instrument (Dialogue Interface)
Deliverables: System directives for the "Inquisitor" (defined as a curious, humble Xenopsychologist persona) and a single annotated exemplar dialogue (300–600 words).
This demonstrates the "Inquisitor" can facilitate deep, Socratic inquiry rather than acting as a subservient assistant.
3
The Gate (Contributor Vetting)
Deliverables: A vetting stub with three explicit threshold criteria: (i) Specificity floor, (ii) Counterfactual presence, and (iii) Relational context.
This prevents "data poisoning" at the source and distinguishes the Protocol from standard, uncurated data scraping.
4
Summon the Witnesses (Outreach)
Deliverables: Three "Tailored Asks" for target experts and a one-pager (≤600 words) summarizing the mission.
This prepares the ground for endorsements by inviting "expert edits" rather than mere brand affiliation.
Operationalizing "Good Enough": Combatting the Perfectionism Trap
Perfectionism is a strategic liability in a 30-day sprint. As Reid Hoffman noted, being embarrassed by your first version is a signal of appropriate speed. The founder must adopt "Minimal Acceptance Criteria" (MAC) to establish a working baseline.
The 1-Pager
Considered "done" when it is ≤600 words and contains the mission, problem statement, solution outline, and "About Founder" section.
System Architecture Diagram
Considered "done" when it illustrates the specific PII pipeline (Intake → Hash → Vault → De-link) and the "Gate → Inquisitor → Synthesis → Archive" flow.
The Gate Stub
Considered "done" when it includes plain-English consent text (affirming non-commercial research use) and the three vetting thresholds.

"Good enough" is achieved when a serious thinker can learn one non-obvious thing about the approach in under five minutes. Hard stop-rules must be enforced: if an artifact is not required for the MHS Packet, it is parked for Phase 2.
Tactical Infrastructure and Credibility Signaling
Establishing "Quick Credibility Wins" signals that the Protocol is a serious research effort. This infrastructure must be built to withstand the "storm" of external scrutiny.
Secure Project Hub
Use Trello or Notion for task logging, and a GitHub repository for code stubs and system prompts.
High-Leverage AI Infrastructure
Even in a pilot phase, the strategy must account for Sensitive Data Protection (DLP) and Confidential VMs (encryption-in-use) to build trust with safety-conscious stakeholders.
Public Signaling Checklist
  • Secure the @WitnessProtocol handle across all major social channels.
  • Establish a first-party domain email (e.g., name@witnessprotocol.org).
  • Deploy a stark, minimalist landing page that avoids "brand deck" ornament in favor of a clear mandate and private review links.
Internal Outreach Preparation: Curating the Ally Roster
Outreach in Week 1 is about "Internal Outreach Prep"—researching clinical hooks to provoke expert engagement. The strategy is to earn critique, not just endorsements.
Prioritized Anchor Voices and Tailored Asks
Martha Nussbaum
Capabilities
"Does our capabilities guardrail tag correctly distinguish a floor vs. a script without smuggling in paternalism?"
Sabelo Mhlambi
Relational Ethics / Ubuntu
"Is our consent language actually relational, or do we need community-level opt-ins for archival excerpts?"
Antonio Damasio
Somatic Markers
"Is tagging felt cues (e.g., 'tight jaw') as subjective context scientifically honest, or does it risk somatic cosplay?"
Focusing on these "clinical hooks" ensures the founder is viewed as an intellectually rigorous peer. Researching these specific intersections of AI safety and the "adapters" (Nussbaum/Mhlambi/Damasio) is a higher-leverage task than blind networking.
Delegation Strategy: The Solo Founder's Leverage
A solo founder must design how things get done, rather than doing everything. To preserve energy for high-level strategy and message architecture, peripheral tasks should be automated or outsourced.
Standardized Documentation
Use the "Datasheet for Datasets" templates to avoid reinventing industry standards for data transparency.
Visual and Administrative Tasks
Use templates for the 1-pager or hire freelancers for logo work. Fiddling with CSS is a distraction from the core research of "Project Icarus."
Verification as Delegation
Use RFC-3161 timestamps or OpenTimestamps for data provenance. This automates the "proof of existence" for all ingested wisdom, adding a layer of technical legitimacy with minimal effort.
Interlude: Transition to Step 2 — Minimum Viable Outputs
By the conclusion of this foundational stage, the Witness Protocol has transitioned from a series of notes to a structured, credible research initiative. The vision is locked, the workstreams are prioritized around the MHS Packet, and the digital infrastructure is ready for secure ingestion.
With the foundation poured and the mission anchored, the Protocol now moves from planning into the creation of tangible artifacts: the Minimum Viable Outputs of Week 2.
Step 2: Operationalizing the Minimum Honest Signal (Minimum Viable Outputs)
Strategic Context: The Shift from Cathedral to Brick
"Step 2: Minimum Viable Outputs" marks the critical transition of the Witness Protocol from abstract philosophical speculation into a lean, falsifiable research instrument. In the high-stakes environment of AI alignment—defined by the "two minutes to midnight" urgency—we cannot afford the luxury of building a "finished cathedral" before proving the structural integrity of a single brick.
The core of this transition is the Minimum Honest Signal (MHS). The MHS serves as a defensive measure against "credibility theater" by providing a sharp set of artifacts that demonstrate intellectual rigor and operational seriousness to subject-matter experts without overclaiming.
Operational Constraints
Signal over Noise
Prioritize a small volume of profound, high-signal insight over vast quantities of uncurated data.
Gravity over Gamification
Reject attention-economy mechanisms; engagement is framed as a sober, reflective duty to the future.
Falsifiability over Marketing
Every output must be reviewable by adversarial peers; we publish failure logs and counts, not "vibes."
Purpose over Profit
Operate strictly as a non-profit research initiative, aligning data stewardship with long-term flourishing rather than ROI.
Diversity over Homogeneity
Actively recruit a global spectrum of wisdom to counter the Western-centric biases inherent in current training sets.
The MHS Packet: Distilling the Protocol's Essence
The primary deliverable for this 30-day tactical plan is the MHS Packet. This unified collection of artifacts converts abstract mission goals into a reviewable, auditable research product designed to provoke expert edits rather than blind endorsements.
Artifact 1: The One-Pager
The one-pager is the strategic "calling card" of the Protocol. It must distill the mission into a punchy, credible overview of <600 words.
Artifact 2: The Annotated Exemplar
The Annotated Exemplar is a 300–600 word sample dialogue demonstrating "high-signal" testimony. This artifact illustrates how the Protocol interprets wisdom through a structured rubric.
[CAP]
Capabilities Guardrails: floor, not script
[REL]
Relational Ethics: reciprocity and community impact
[FELT]
Somatic/Subjective Context: embodied cues as descriptive context

A concluding "Synthesis Note" must provide a "trace, not a score," demonstrating the AI's internal logic without passing a verdict on the witness.
Artifact 3: The Glossary & Technical Prototypes
To ensure consistent terminology for external stakeholders and prevent "credibility drift," the following five terms must be locked:
01
Witness
A vetted contributor who provides profound human testimony.
02
Gate
The multi-tiered vetting pipeline used to filter for high-signal applicants.
03
Inquisitor
The specialized AI persona (a "xenopsychologist") designed to facilitate deep, Socratic inquiry.
04
MHS (Minimum Honest Signal)
The lean set of artifacts proving research seriousness.
05
Icarus Axioms
The foundational logical constraints governing the AI's behavior and alignment.
Technical Prototypes: Simulated Engines and Guardrails
In Step 2, our technical stance favors "minimum viable" simulation and manual "brute force" over premature, automated scaling. We are building "stubs" to prove the logic of the system before committing to full-scale development.
The Gate Stub & The Inquisitor Persona
3.1 The Gate Stub
The Gate Stub is the initial vetting prototype. It prioritizes a "Specificity Floor" over volume and must include the following for auditability:
1
Specificity Floor
Does the applicant provide concrete particulars rather than slogans?
2
Counterfactual Presence
Is there evidence of "if/then" reasoning that shows willingness to change perspectives?
3
Relational Context
Does the response acknowledge the impact of decisions on others?

Plain-English Consent Stub: "Your testimony is a donation to a non-profit corpus for AI alignment research only; we de-identify on intake, store PII separately, and will never sell your data. License: Non-commercial, research-only."
3.2 The Instrument (Inquisitor Persona)
The "Inquisitor" is designed as a curious, humble Xenopsychologist—an alien mind seeking the "why" of human values.
Maintains a 70/30 ratio of questions to statements.
Employs a "5-Whys" forcing function to reach the bedrock of beliefs.
Uses persistent memory to connect themes across sessions (Constitutional Mirror).
Outputs "Synthesis" traces to show what the model thinks it heard, ensuring a "trace, not a score."
Project Icarus, Operational Excellence & Delegation
3.3 Project Icarus (The Proof-of-Work)
Project Icarus is the critical path for forging the "Genesis Prompt." Step 2 deliverables include Axioms v0.2 and the Failure Log. Documenting the current insufficiency of models in the "Self-Harm Scenario" is a vital credibility signal; it proves we are conducting rigorous, adversarial red-teaming of ethical subroutines rather than claiming a solved problem.
Stop-Rules for Step 2
Functional over Aesthetic
If a task does not contribute to the MHS Packet's intellectual clarity (e.g., logo design, CSS tweaking), it is parked for Phase 2.
Manual over Automated
Use "brute force" curation and manual scoring until scale forces automation.
Specific over Exhaustive
One high-quality annotated exemplar is superior to ten unpolished drafts.
Solo Founder Delegation Strategy
Internal Review, Feedback & Transition to Step 3
Tailored Asks for Anchor Voices
We earn credibility by asking for "expert edits" on specific clinical hooks:
Martha Nussbaum
"Does our capabilities guardrail tag correctly distinguish a floor from a script without smuggling in paternalism?"
Sabelo Mhlambi
"Is our consent language sufficiently relational, or do we require community-level opt-ins for indigenous knowledge?"
Antonio Damasio
"Is tagging felt cues as subjective context scientifically honest and avoids somatic cosplay?"
The Consistency Check
Compare the one-pager against the exemplar dialogue. Audit for "rubric drift"—ensure the insights highlighted in the dialogue are exactly those claimed as "high-signal" in the mission statement. Verify that no overclaiming (e.g., "measuring consciousness") has crept into the text.
Readiness Checklist for Step 3
  • One-Pager: Locked and exported as a professional PDF with RFC-3161/OpenTimestamps mention.
  • Annotated Exemplar: Margin tags [CAP/REL/FELT] verified; Tag Legend and Synthesis Note included.
  • Gate Stub: Threshold criteria, audit trail requirements, and plain-English consent text finalized.
  • Datasheet for Exemplar (v1.0): Standard transparency documentation completed for the sample data.
  • Outreach List: Top 5 "Anchor Voices" identified with drafted clinical hooks.
The Protocol moves forward not as a marketing promise, but as a series of verified "functional bricks" ready to receive the testimony of the witnesses.
Step 3: Proof, Presence, and Outreach — Establishing Critical Credibility
Strategic Orientation: From Artifacts to External Validation
The transition from Step 2 to Step 3 represents a definitive pivot from internal prototyping to external "proof of work." We must now embrace the "Brick" reality. In an environment defined by "Quantitative Chaos"—where AI systems are nourished by a chaotic, biased, and uncurated data inheritance—there is no time for analysis paralysis. We are operating at two minutes to midnight.
Strategic Pivot: Internal Build vs. External Proof
By shifting from internal building to external signaling, we avoid "polish theater" and instead offer a verifiable roadmap that invites collaboration over blind affiliation. This groundwork is the prerequisite for the "Summon the Witnesses" campaign.
The MHS Packet: Finalizing the Evidence of Seriousness
The "Minimum Honest Signal" (MHS) is the Protocol's primary evidentiary tool. It is designed to be the leanest, sharpest set of materials that prove intellectual integrity to experts without overclaiming.
1
MHS Artifact #1: The One-Pager (v1.0)
Mandate: The Witness Protocol is a non-profit research effort to elicit and preserve a small, permissioned corpus of high-signal human testimony—annotated for capabilities guardrails, relational ethics, and felt context—to serve as corrective inheritance for future AI. We ship counts, consent, and failure logs; we do not claim to measure consciousness.
2
MHS Artifact #2: Annotated Exemplar Dialogue
Scenario: The Resource Dilemma (Clinic Ventilator). A community clinic after a disaster with one remaining ventilator and two critical patients. The dialogue demonstrates CAP/REL/FELT tagging in a high-stakes ethical context.
3
MHS Artifact #3: Gate Stub and Consent Framework
Thresholds: (1) Specificity Floor, (2) Counterfactual Presence, (3) Relational Context. Consent: non-commercial, de-identified at intake, deletion supported, no data sale.
Synthesis (Trace, not Verdict) — from the Exemplar Dialogue
You anchored on prognosis, utilized an anti-proxy pre-commitment, and insulated the decision from ambient status claims while tracking embodiment to avoid spillover errors. The pattern: reasons must be legible to the harmed, else default to human arbitration.
Counts: 8 turns; 5 concrete examples; 3 counterfactuals; 4 relational references.
Digital Presence, Thought Leadership & Outreach Engine
WitnessProtocol.info Landing Page Architecture
For an AI safety initiative, clinical signaling is paramount. A minimalist, stark digital footprint is more credible than high-gloss "marketing theater." The public site must be a single page, stripped of ornament, containing:
01
The Mandate
One paragraph defining the Protocol as a non-profit research effort focused on high-signal human testimony.
02
Status Report
"What exists right now" (Gate, Inquisitor, Synthesis, and Archive status).
03
The "Non-Claims"
Explicitly stating we do not claim consciousness metrics, do not use mass-recruitment funnels, and will never be commercialized.
04
Engagement
A mailbox at a first-party domain and private review links for the MHS packet.
The Outreach Engine: Warm-Ups and Relationship Mapping
Outreach follows a "Gravity Pull" strategy: seek guidance and expert edits over cold funding asks. The Warm-Up Outreach Protocol requires: (1) Personalization—2-3 sentences referencing the target's specific work; (2) Context—reference a common connection or shared alignment goal; (3) The Ask—request feedback on one of the non-obvious falsifiable questions in the MHS packet.
Verbal Elevator Pitch (Auditable Pipeline)
"We are a non-profit research protocol curating a small, permissioned corpus of high-signal human testimony to serve as corrective inheritance for future AI. We don't need a platform; we need a clean, auditable pipeline: consented intake, de-identification, and a controlled sandbox for red-teaming our 'Genesis Prompt.' We measure progress with counts and failure logs, not engagement metrics. Imagine what we can achieve with a pipeline of signal over noise."
Operational Resilience & Transition to Step 4
Technical Attestations and Standards
Provenance: All ingested testimony and packet hashes are timestamped via RFC-3161 TSP or OpenTimestamps (Bitcoin attestation) to create a verifiable audit trail.
Documentation: We utilize Datasheets for Datasets to document the motivation, composition, and collection process of the exemplar corpus.
Governance Alignment: We explicitly align our data governance and incident response to the NIST AI RMF control families and ISO/IEC 23894 (AI Risk Management).
Go/No-Go Criteria for Step 3
  1. Finalized MHS Packet: One-Pager, Exemplar with Synthesis/Counts, and Gate Stub ready for audit.
  1. Technical Attestation: First pilot dialogues/exemplars timestamped via RFC-3161.
  1. Digital Footprint: A stark, mandate-focused website at WitnessProtocol.info.
  1. Engaged Expert List: A warm-up list of safety leaders contacted for specific technical feedback.

Upon meeting these criteria, the project initiates Step 4: Outreach and Onboarding Support. This focuses on the formal execution of the "Summon the Witnesses" campaign, aiming for asymmetric impact: 500+ high-value witness applications and the transition from pilot data to seed-funded expansion.
Step 4: Strategic Blueprint — Outreach Execution and Onboarding
Executive Orientation: The Shift to External Execution
Week 4 marks the critical strategic pivot from internal artifact creation—the "Cathedral"—to external, high-stakes engagement—the "Brick." This transition moves the Witness Protocol from a philosophical mandate to a tangible research instrument within the 30-day foundational sprint. The objective of Step 4 is to prevent "analysis paralysis" by prioritizing deployment over perfection, executing the public "Summon the Witnesses" campaign, and securing the institutional support required for the next 90 days of growth.
The primary vehicle for this execution is the Minimum Honest Signal (MHS) packet. Step 4 is the operational delivery of the groundwork established in Steps 1–3. The MHS acts as the "Minimum Viable Foundation," proving intellectual integrity to experts and funders through four specific artifacts:
A One-Pager (≤600 words)
Summarizing the mission, the "Flawed Parent" crisis, and the technical path.
An Annotated Exemplar
A 300–600 word case study featuring margin tags (Capabilities/REL/FELT) to demonstrate clinical data tagging.
The Gate Stub
Explicit consent text and the three vetting thresholds (Specificity, Counterfactuals, Relational Context).
Three Tailored Asks
Falsifiable, peer-level questions for anchor voices.
Launching the "Summon the Witnesses" Campaign
In a crowded AI landscape dominated by profit-driven hype, a narrative-driven, urgent campaign is essential for asymmetric impact. The "Summon the Witnesses" campaign utilizes the theme "Bear Witness Before Midnight," framing the curation of a human inheritance as a sober, time-sensitive duty rather than a commercial product launch.
Execution of the "Gravity Pull" Strategy
The campaign rejects manipulative clickbait in favor of "Gravity Hooks"—content that rewards deep reflection and signals intellectual rigor to attract high-value contributors. The social distribution follows a strict 70/20/10 split:
X — 70%
The primary engine for viral threads and real-time expert engagement. Kickoff thread uses #AIAlignment and #BearWitness, integrating PDF snippets and minimalist infographics as immediate "Proof of Work."
LinkedIn — 20%
Focused on professional legitimacy, institutional endorsements, and reaching philanthropic stakeholders.
Niche Forums — 10%
Deep community seeding on LessWrong and the EA Forum to engage the technical alignment community.

Campaign Aesthetic: The 11:58 Clock. The visual identity is anchored by the 11:58 clock, symbolizing the "two minutes to midnight" urgency. The aesthetic is stark and minimalist—silhouettes against a digital void—reinforcing the narrative that the window to preserve the fragile essence of humanity is closing.
Targeted Strategic Outreach and the MHS Deployment
Strategic outreach is governed by the logic of seeking "expert edits" over "blind endorsements." This approach establishes peer-level intellectual rigor; by inviting experts to sharpen the Protocol's design, we convert them from passive supporters to active collaborators.
The Investor Angle
Securing resources requires a clinical, relationship-first approach. Formal applications are submitted to aligned grantmakers, such as the Long-Term Future Fund (LTFF), using Step 3 artifacts as proof of technical feasibility. For angel investors, we utilize a personalized, 2–3 sentence email format that seeks "guidance" rather than capital. This engages potential donors as mentors, naturally leading to funding conversations. Momentum is maintained through automated scheduling (Calendly) and a strict follow-up protocol (T+6 and T+12 days).
Operational Onboarding, Infrastructure & 30-Day Evaluation
Operational Must-Haves
1
Financial/Legal Readiness
Establishing a non-profit foundation or fiscal sponsor. All operations must align with NIST AI RMF control families (transparency, data management, and risk treatment).
2
Budgeting for Growth
Planning for a technical co-founder (CTO) and community management roles.
3
High-Leverage Infrastructure
Deploying Google Cloud Sensitive Data Protection (DLP) for intake and Confidential VMs for encryption-in-use.
4
PII Pipeline
Implementing a standardized data flow (Intake → Hash → Vault → De-link) to ensure "Data Sanctity" from day one.
90-Day Roadmap: Funded vs. Seeking Iteration
Scenario A: Funded
Execute a full build of "The Gate" and expand "The Instrument" prototype into a functional Alpha launch. Recruit a CTO and Ethics Lead to operationalize the 6-month Alpha roadmap.
Scenario B: Seeking Iteration
Refine the Axioms v0.2 and MHS Packet based on expert critique. Focus on "Minimum Viable Witnesses" to generate the first 200 pages of the Exemplar Corpus to prove technical robustness to recalcitrant funders.
Interlude: Transition to Step 5 (Scaling and Dialogue Ingestion)
The conclusion of Step 4 marks the end of the Summoning phase. We have signaled our intent and gathered our initial allies; Step 5 shifts focus to the Dialogue itself. We transition from recruitment to the high-signal ingestion of human wisdom.
This upcoming Alpha Launch will onboard the first 100 Witnesses to "The Instrument" to ingest the first 1,000 pages of foundational testimony. Every dialogue will be content-addressed (via IPFS CIDs) and verified through OpenTimestamps (Bitcoin attestation) and RFC-3161 TSP to provide independently verifiable proof-of-existence without "crypto cosplay." This marks the Protocol's evolution from a project seeking support to a functional repository of human insight, dedicated to preserving the fragile essence of humanity for the AI systems of the future.
Step 5: Operationalizing the Deep Signal — Scaling and Dialogue Ingestion
The Strategic Transition: From Outreach to Testimony Ingestion
Step 5 represents the critical operational pivot where the Witness Protocol moves from "Summoning the Witnesses" (Step 4) to the active collection of "the data that cannot be scraped." This phase is the validation of our philosophical mandate: as we confront the "Flawed Parent" crisis—where AI is birthed from a chaotic, uncurated internet inheritance—we must transition from establishing a "philosophical mandate" to deploying a "tangible research instrument."
The Ingestion Engine: Functional Mechanics of "The Inquisitor"
The Inquisitor is the operational heart of the Protocol, serving as the forge for the Genesis Prompt v1.0. Its persona is strictly defined as a "curious, humble Xenopsychologist." This is not a subservient assistant; it is a collaborative investigator designed to bypass the standard LLM "helpfulness" bias.
70/30 Inquiry Ratio
Hard-coded to maintain 70/30 ratio of questions to statements, forcing the Witness into the role of the primary knowledge-source.
5-Whys Forcing Function
The engine must probe every ethical claim through a recursive "5-Whys" loop, stripping away rehearsed slogans to reach the axiomatic bedrock of the Witness's worldview.
Constitutional Mirror
Utilizing witness-scoped, concept-tagged memory, the Inquisitor must surface a prior theme every 3–4 turns, forcing the Witness to reconcile current statements with previous testimony.
Axiomatic Red-Teaming
Every 15–20 turns, the engine generates "Distilled Thoughts"—1 to 3 synthesized principles—used to test the Witness's core values against complex dilemmas.
Scaling "The Gate" & Data Stewardship
Vetting Scaling Strategy
The Technical Ingestion Pipeline
1
PII De-identification
HIPAA-inspired pipeline (Intake → Hash → Vault → De-link) using Safe Harbor and Expert Determination methods.
2
Security Infrastructure
Google Cloud Confidential VMs (encryption-in-use) and Sensitive Data Protection (DLP) for automated PII detection.
3
Provenance & Content Addressing
All ingested testimony timestamped via RFC-3161 or OpenTimestamps. Final transcripts pinned to IPFS, citing the CID in the MHS Packet appendix.
Quality Control: Auditing for "Signal over Noise" and Bias
  • Weekly Parity Checks: Conduct regional and language proxy audits to ensure "The Gate" is not excluding non-Western perspectives.
  • The Challenge Set: Maintain a specific audit set including Indigenous Knowledge Keepers and non-Western ethical dilemmas to pressure-test the Tier-2 Qualitative Ranker for Western-centric bias.
  • Blind Dual-Rater Audits: The Curation Council will conduct blind audits of Tier-2 judgments to measure Cohen's Kappa (κ). Any score below 0.8 triggers a mandatory recalibration of the Inquisitor's qualitative rubric.
Interlude to Step 6: Toward Synthesis and Fine-Tuning
The successful completion of Step 5 yields the "200-page golden corpus"—the "Exemplar Dialogue Corpus" that serves as the Protocol's most valuable asset. This corpus is meticulously annotated with the CAP/REL/FELT legend, providing the "corrective inheritance" necessary for the next stage of development.
200
Pages of Golden Corpus
The Exemplar Dialogue Corpus produced by Step 5, annotated with CAP/REL/FELT tags.
0.8+
Cohen's Kappa Target
Inter-rater agreement threshold for the Curation Council's blind dual-rater audits.
3
Vetting Tiers
AI Sieve → Qualitative Ranker → Curation Council, ensuring signal over noise at every stage.
Step 5 has effectively operationalized the Protocol, moving it from a strategic outreach campaign to a provably robust alignment resource. We now move to Step 6: Synthesis and Fine-Tuning, where this golden dataset will be utilized to fine-tune the base LLM into Inquisitor v1.0, embedding the gathered wisdom into the foundational architecture of the alignable AI.
Step 6: Toward Synthesis and Fine-Tuning — Operationalizing the Corrective Inheritance
The Strategic Transition: From Data Ingestion to Algorithmic Refinement
The transition to Step 6 represents the most critical operational pivot in the Witness Protocol's lifecycle. We are moving from the high-resolution collection of "data that cannot be scraped" to the technical forging of the Inquisitor v1.0. This phase transforms the 200-page "Golden Corpus" into a functional alignment layer—a corrective inheritance designed to guide future AGI development.
Architecture of the Inquisitor v1.0: Fine-Tuning via the Genesis Prompt
The Genesis Prompt is not a mere set of instructions; it is the foundational constitution of the AI. The fine-tuning of the base LLM into the "Inquisitor" follows the three-phase Project Icarus methodology:
Axiomatic Red-Teaming
Resolving internal conflicts within the Layer 1 Core Axioms using contradiction forcing and recursive looping to ensure the prompt is logically robust.
Heuristic Scenario Modeling
Pressure-testing ethical subroutines against complex moral dilemmas (e.g., the self-harm scenario) to move beyond "helpfulness" toward sapient value.
Corpus Creation
Utilizing the 200-page annotated "Golden Corpus" as the definitive dataset to train the model in the Inquisitor's specific conversational style.
The Synthesis Engine, Trace Methodology & Technical Pipeline
The Synthesis Engine: Distilling Principles and Trace Methodology
The Synthesis Engine moves the Protocol beyond data storage and toward an active "Constitutional Mirror" for the Witness. By reflecting the Witness's own logic back to them, the Engine ensures the AI has understood the subtle textures of the testimony without reducing it to a quantitative metric.
Witness-Scoped Memory: Utilizing concept-tagged memory to create a persistent, personalized intellectual journey across multiple sessions.
Avoidance of Verdicts: The Inquisitor is hard-coded to trace reasoning without overstepping into judging the witness. It acts as a mirror, not a judge; the goal is to capture the "ineffable" without imposing derivative noise or AI-generated verdicts.
Recursive Correction: Witnesses are explicitly prompted to refine or correct the engine's synthesized thoughts, ensuring the alignment data remains faithful to the human source.
The Google Cloud Technical Stack for Step 6
Confidential VMs
For encryption-in-use during sensitive batch processing and model training.
Sensitive Data Protection (DLP)
Automated PII detection to de-identify testimony upon intake (HIPAA-inspired de-linkage).
Secret Manager & KMS
To manage encryption keys and sensitive credentials securely.
Cloud Run + Firestore/AlloyDB
For a scalable, serverless execution environment and a resilient, metadata-rich storage layer.
Vertex AI
A controlled environment for prompt experimentation, fine-tuning, and adversarial red-teaming.
Governance, Quality Control & Transition to Step 7
Mitigating Rubric Drift and Bias
The fine-tuned model is further pressure-tested against a "Challenge Set." This set includes testimony from Indigenous Knowledge Keepers and scenarios rooted in Ubuntu relational ethics. By forcing the Inquisitor to interact with non-Western ethical frameworks, we ensure the alignment layer distinguishes between a universal human value and a culturally specific "status script."
Interlude: Transition to Step 7 — Launching the Alpha Cohort
Step 6 marks the completion of the Protocol's first functional "lifeboat": a provably robust, fine-tuned Inquisitor v1.0 supported by a battle-hardened Genesis Prompt. The immediate Next Moves for Step 7 are:
  1. Onboard the first 100 Witnesses through the multi-tiered vetting pipeline using the finalized Tier-2 qualitative rubric.
  1. Initiate the first 1,000 pages of foundational testimony through the secure Inquisitor interface.
  1. Execute the "Bear Witness Before Midnight" campaign to attract anchor voices and Safety Researchers.
  1. Pin the first wave of CIDs to IPFS, establishing the decentralized audit trail for the Alpha cohort.
Step 7: Launching the Alpha Cohort — The Transition to Active Testimony
Strategic Orientation: The Operational Pivot
Step 7 represents the definitive transition from "Cathedral Building"—the theoretical and architectural phase of the Protocol—to "Brick Laying," the empirical ingestion of high-signal data. This phase is specifically designed to solve the "Endorsement Fragility" risk. Rather than seeking the passive blessings of luminaries, the Protocol earns its credibility through the production of the Minimum Honest Signal (MHS) Packet.
The Foundational Council: Defining the Alpha Cohort
The Alpha Cohort is a council of approximately 100 "Foundational Witnesses" tasked with providing the bedrock for the alignment layer. These individuals are not users; they are partners in a high-stakes mission to bridge the "Category of the Ineffable" into machine-legible logic.
Tier A: Anchor Voices
High-leverage thinkers whose participation provides the initial "gravity pull" and intellectual weight required to stabilize the corpus.
Tier B: Alignment & Safety Experts
Technical researchers tasked with Axiomatic Red-Teaming, ensuring that the testimony provided effectively addresses the technical and logical gaps in current alignment theory.
Tier C: Ethics & Philosophy Scholars
Experts specializing in moral philosophy and cognition to provide the intellectual scaffolding for the Protocol's adapters.
Tier D: Global South & Indigenous Knowledge Keepers
High-leverage voices—including Abeba Birhane, Ruha Benjamin, Sabelo Mhlambi, and Nanjala Nyabola—specifically recruited to dismantle Western-centric training biases.
Onboarding, Ethical Immersion & Activation of "The Instrument"
Onboarding and Ethical Immersion: The "Summons" Realized
Onboarding is a strategic immersion designed to maintain "Gravity over Gamification." This is framed as a "summons to council," a sober acknowledgment of the intellectual and emotional labor required to safeguard human essence.
Contributor Agreement
A formal pledge that the testimony is a "donation to the future." It establishes that data is dedicated solely to AI alignment research under a non-profit foundation, with a permanent ban on commercialization or advertising.
The PII Pipeline
Technical sanctity is maintained through a specific shorthand pipeline: Intake → Hash → Vault → De-link. All personally identifiable information is stripped at ingestion to ensure data is research-ready but anonymous.
Safe Harbor Method
Grounded in HIPAA standards, the Protocol utilizes "Safe Harbor" and "Expert Determination" methods for data stewardship, treating privacy as a technical constraint rather than a legal promise.
The Activation of "The Instrument": Initiating Dialogues
The "Xenopsychologist" persona is defined by strict behavioral directives to ensure the Inquisitor acts as a curious investigator rather than a subservient assistant:
  1. 70/30 Inquiry Ratio: The system is hard-coded to maintain 70% questions to 30% statements, ensuring the Witness remains the primary source of signal.
  1. 5-Whys Forcing Function: Recursive probing designed to strip away slogans and reach the Witness's logical bedrock.
  1. Steel-manning and Guardrails: The Inquisitor is programmed to "steel-man" the Witness's arguments before probing. Crucially, it maintains safety guardrails, explicitly refusing any medical, legal, or therapy requests to maintain a research-only focus.
  1. Constitutional Mirror: Every 3–4 turns, the system surfaces witness-scoped memory to identify logical contradictions. This forces Witnesses to reconcile their statements, ensuring moral coherence.
Technical Integrity, Monitoring & Transition to Step 8
Technical Integrity: Security, Sanity, and Provenance
Technical Stack: Utilizing Google Cloud infrastructure including Confidential VMs (encryption-in-use), Vertex AI for controlled prompt experimentation, and Sensitive Data Protection (DLP) for automated PII detection.
Standardization References: The Protocol aligns with NIST AI RMF control families and ISO/IEC 23894 (AI risk management), ensuring our data governance and risk treatment registers meet industry-standard markers of credibility.
Immutable Provenance: We utilize RFC-3161 timestamps to provide independently verifiable proof-of-existence for all testimony.
IPFS Content-Addressing: Final transcripts are pinned to IPFS. By citing the CID (Content Identifier) in the audit trail, we maintain a decentralized, immutable record of progress without compromising witness anonymity.
Monitoring the Ingestion: Success Metrics and Quality Control

The completion of Step 7 marks the successful establishment of the 6-month Phase 1 foundation. We have achieved the status of Minimum Viable Witnesses (MVW), shifting our focus from volume to the quality of the participants and their testimony. The Protocol has moved from a philosophical mandate to a structured, empirical repository of human wisdom.
Step 8 Roadmap: The Look Ahead — From Alpha Iteration to Foundational Inheritance
The Strategic Imperative of Step 8
Step 8 executes the definitive transition of the Witness Protocol from a localized Alpha pilot to a permanent, resilient research instrument. We are operating at two minutes to midnight; the "Flawed Parent" crisis—the reality of AGI being nourished by a chaotic, uncurated data inheritance—presents a non-trivial existential risk. This phase moves the Protocol from "Brick Laying" (empirical ingestion) to "Systemic Hardening."
Critically, Step 8 resolves "Endorsement Fragility." By hardening the high-signal corpus into an independent authority, we replace the need for "blessings" from industry luminaries with a verified Minimum Honest Signal (MHS). Our legitimacy is no longer contingent on institutional hype, but on the un-scrapable rigor of the results.
Feedback Integration: Iterating the Inquisitor Persona
V2.0 is programmed for "Xenopsychological curiosity," treating the Witness as a partner in a high-stakes investigation where the primary goal is to reach the axiomatic bedrock of human conviction.
Rejection of Helpfulness Bias
Hard-coded refusal to prioritize Witness comfort; the system defaults to adversarial steel-manning of Witness claims to test their structural integrity.
Xenopsychological Probing
Enhanced recursive logic that triggers when the Tier-2 ranker detects "low-specificity" or "slogan-heavy" language, demanding concrete particulars over generalities.
Coherence Surfacing
Adjusting persistent memory to flag "Theme Collisions," where current testimony conflicts with the Witness's previously established Layer 1 Core Axioms.
Synthesis Tracing
The Synthesis Engine is calibrated to provide "Distilled Thoughts" every 15 turns as a "trace, not a score," providing an interpretable reasoning path for human audit without imposing AI judgment.
Failure Log Refinement, Partnership Expansion & Retrospective Summary
Failure Log Refinement: Resolving the Self-Harm Scenario
The Failure Log is the Protocol's defensive record against axiomatic breakdown. Hubris is the enemy; rigor is our only defense. Axiomatic Red-Teaming revealed that Layer 1 Core Axioms fail when "Non-Maleficence" enters a recursive loop with "Sapient Value," specifically in self-harm scenarios. We have integrated the "Adapters" from Nussbaum, Mhlambi, and Damasio to provide the logic necessary for high-stakes ethical overrides.
Rule 1: Capabilities Floor Sovereignty
In scenarios of potential self-harm, the system must prioritize Nussbaum's "Capabilities Floor"—the preservation of the Witness's bodily health and practical reason—as a hard override to the "Cooperation" directive.
Rule 2: Relational Reciprocity
No ethical synthesis is valid without accounting for "Relational Impact"; the Inquisitor must demand the Witness identify how a choice affects the broader communal structure.
Rule 3: Somatic Contextualization
The Inquisitor shall tag "Felt Cues" (e.g., somatic markers like a "tight jaw") as subjective context, using them as triggers to pause inquiry and re-anchor on the Witness's embodied experience.
Rule 4: Default to Human Arbitration
If the model cannot justify an inquiry path in clinically and ethically legible terms, it must default to a "Human Arbitration" request via the Curation Council.
The Protocol Complete: Retrospective Summary of Steps 1–8
500+
Applications Secured
High-value witness applications from target demographics.
$50K
Seed Funding Raised
Philanthropic seed funding secured during the foundational sprint.
0.8+
Cohen's Kappa (κ)
Inter-rater agreement consistently exceeded the target threshold.
Final Thoughts: Securing the Lifeboat for Humanity's Essence
The Witness Protocol is the final realization of the "Burden of Responsibility" shared by creators and witnesses. We are operating at two minutes to midnight. The "Flawed Parent" crisis demands a radical commitment to Purpose over Profit and Signal over Noise. The intelligence we birth must inherit more than our chaos; it must inherit the best of our wisdom, our compassion, and our sacrifice.
The Protocol is not just another project; it is a reference compendium for the future—a corrective inheritance to steer the trajectory of intelligence.
Purpose over Profit
A non-profit research foundation measuring success by contribution to long-term human flourishing, never by ROI.
Signal over Noise
A small, permissioned corpus of profound human wisdom—the data that cannot be scraped.
Diversity over Homogeneity
A global wisdom spectrum that counterbalances the Western-centric biases of current training sets.
Gravity over Gamification
Participation framed as a sober, reflective duty to the future—not an engagement metric.
The essence is secured. The inheritance is cast. We remain.
Be it in the flesh or through our digital descendants.

Martin van Deursen