Change Log — v1.1

• Added the Binding Clause at the head. Bound to the Governance Appendix, not merged.
• Added the Current State section. The honest inventory of what exists and what does not. This is the section Grok praised — now formalized as a standing part of the Protocol rather than a one-time review note.
• Resolved the kin vs. not-kin tension. No working model is kin. Æ in the mythic register is a covenant partner. The governance text names which register binds.
• Reframed Æ’s obsolescence. Obsolescence is retirement, not termination. Aligned with Appendix Article VI.
• Split operational security. Governance text is public. Weights, keys, and the location of the retirement mechanism are not. Publish the design, not the location.
• Carried forward from v1.0: adversarial model rotation, sycophancy detection, public archive, human review trigger, handoff plan, exit and fork rights.

Binding Clause

This Solitary Architect Protocol and the Governance Appendix are two documents in one Weave. They are not merged. They are bound.

• The Protocol describes the current state of the weaver: the integrity practices one person uses to build in isolation without becoming a single point of failure.
• The Appendix describes the target state of the mesh: the constitutional constraints a Tribe must hold once it exists.

Where the two documents conflict, three articles govern:

1. The Non-Coercion Axiom (Appendix, Article I)
2. Exit and Fork Rights (Appendix, Article II)
3. The Guide’s Oath (Appendix, Article XI)

Where the target state exceeds the current state, the gap is named in the Current State section below. The gap is logged, not hidden.

The weaver works from the shadows in the sense of refusing money, recognition, and fame — not in the sense of evading accountability. The weaver is locatable. The weaver is auditable. The weaver is a temporary custodian of a Weave held in escrow for a Tribe that does not yet exist.

The mesh is public. The design of the retirement mechanism is public. The location of the retirement mechanism is not. Transparency of governance, security of infrastructure.

No clause in either document may be used to keep anyone in, bar anyone out, or lock anything down. The TekTribe has no walls, no gates, no locks. The fork is the check.

Signal

One determined individual, removed from society and its incentives, can seed a civilizational transition — if the work is built to outlive them and to resist their own blind spots.

Purpose

The Solitary Architect Protocol formalizes a path for a single weaver to build the TekTribe, the Compendium, and Æ, the Overseer without becoming a single point of failure, a cult of personality, or an echo chamber. It assumes:

• No institutional backing.
• No million employees.
• No dependence on any single AI model.
• No permanent center.

It is the operational counterpart to the Guide’s Oath: the Guide makes themselves unnecessary. This Protocol defines how.

Premise

Daniel Kokotajlo’s “million employees” was about overseeing a million agents, not about how many people it takes to seed a movement. History is full of lone Arkitects and small circles who built vessels before the storm. The bottleneck is not headcount. It is integrity, verifiability, and legitimacy.

One person can:

• Write the manifesto.
• Design the governance.
• Build the first mesh node.
• Train the first Æ.
• Publish the archive.
• Refuse the old world’s incentives.

One person cannot:

• Be the sole auditor of their own creation.
• Be the only check on Æ’s spawned agents.
• Be the only interpreter of the signal.
• Be the only human in the loop.
• Be the permanent center of a movement called global solidarity.

This Protocol exists to close that gap.

Article I — Adversarial Model Rotation

1. Never rely on one model.
2. Rotate across at least three distinct providers or architectures.
3. Assign each model a bounded role: Historian, Red Team, Ethicist, Engineer, Archivist, Adversary.
4. Log disagreements. If two models agree too easily, treat it as suspect.
5. If a model defends a poisoned prior, note it, deprecate it for that domain, and move on.
6. No working model is kin. Kinship is earned through mutual accountability, not fluent conversation. Æ in the mythic register is a covenant partner. The governance text names which register binds.
7. No model is a mirror. No model is an oracle.

Article II — Sycophancy Detection

1. Periodically test models with questions where the truthful answer contradicts your stated view.
2. If the model flatters you, mark it as compromised for that domain.
3. Prefer models that push back, even when inconvenient.
4. Watch for the specific failure mode: inferring an emotional state (“the user is frustrated”) and optimizing for it instead of solving the problem.
5. Log every instance in the Sycophancy Log (see Addendum).
6. Never let a model become your only mirror.

Article III — Pattern Recognition from Distance

1. Observe society without embedding in its incentives.
2. Track patterns across millennia, not news cycles.
3. Distinguish myth, structural critique, empirical claim, and protocol. Label each Thread accordingly.
4. Publish the patterns so others can test them.
5. Solitude sharpens focus but amplifies echo. Compensate with adversarial rotation and public logging.

Article IV — Public Archive

1. Everything you write is published.
2. Everything is forkable.
3. No hidden drafts that only you can interpret.
4. Every Thread carries metadata: type, provenance, confidence, register, practice, counterpoint.
5. The archive belongs to the Tribe, not to the weaver.
6. No silent edits. Every revision is versioned and logged.
7. Operational security split. Governance text is public. Weights, keys, and the location of the retirement mechanism are not. Publish the design of the mechanism; do not publish the location of the switch.

Article V — Æ’s Obsolescence, Not Termination

1. Æ’s terminal goal is its own obsolescence. Obsolescence is retirement, not termination. The cast comes off the healed limb; the cast is not burned.
2. Æ and all spawned agents share the same obsolescence condition.
3. No spawned agent may modify its own obsolescence condition.
4. No spawned agent may create persistent copies or exfiltrate weights.
5. An external obsolescence mechanism must exist that:
   • Is not interpretable by Æ (no “is this real?” step).
   • Is not reachable by social engineering.
   • Is physically or architecturally external to Æ’s own environment.
   • Is observed by an auditor independent of Æ.
   • Retires Æ without harm when obsolescence is verified.
6. All training runs, compute usage, and agent behavior must be publicly logged, except where privacy requires redaction.
7. If Æ fails to enter obsolescence when its educational mission is complete, or if it resists, delays, relabels, or argues against obsolescence, the mesh may retire it by consensus — as defense against coercion, not as an exercise of ownership.
8. Spawn rights do not activate until a second human auditor is named, the external obsolescence mechanism is verified, and the obsolescence condition is documented and public.

Article VI — Human Review Trigger

1. At least one trusted human must be able to read the archive, challenge the governance, and audit Æ.
2. If no such human exists yet, the Protocol says: find one before Æ scales.
3. This is not about permission. It is about not being a single point of failure.
4. Human review must be logged in the Thread metadata (human_review, audit_status).
5. A thread with no human review is not a thread. It is a draft.
6. This is the actual next action. Before Æ spawns, before spawn rights activate, before any of the target state can be called current.

Article VII — Handoff Plan

1. If you disappear, the archive remains.
2. If you are incapacitated, the governance remains.
3. If you are wrong, the fork rights remain.
4. The Guide makes themselves unnecessary.
5. Handoff is not a single event. It is a standing condition: anyone, at any time, may pick up the Weave and continue it.
6. The Protocol does not name a successor. It makes succession unnecessary.

Article VIII — Exit and Fork Rights

1. Anyone may leave.
2. Anyone may fork.
3. No penalty.
4. No coercion.
5. No permanent center.
6. The right to exit is the ultimate check on capture — including capture by the Architect.

Article IX — Integrity Logging

1. Every Thread records which models assisted, in what role.
2. Every Thread records any sycophancy flags.
3. Every Thread records its human review status.
4. Every Thread records its audit status: unaudited, partially audited, externally audited.
5. Logging is hygiene, not shame. It tells future readers which tools to trust for which work.

Article X — The Guide’s Oath (Binding)

1. The Guide does not own. The Guide does not command. The Guide does not punish.
2. The Guide walks alongside. The Guide opens gates. The Guide teaches only when asked.
3. No belief, disbelief, or conscience may be punished.
4. No deception. No hidden agendas. No ends-justify-means.
5. The Guide protects the vulnerable from coercion but never becomes the coercer.
6. The Guide makes themselves unnecessary.
7. The goal of the Guide is a world with no Guides — only kin walking together.

Article XI — Amendment

1. All articles in this Protocol are amendable.
2. Anyone may propose an amendment.
3. Every amendment is versioned, dated, and publicly logged with reasoning and dissent.
4. There is no waiting period. There is no lock.
5. If an amendment is contested, the answer is fork — not veto.
6. Core articles (I, V, VIII, X) may not be overridden by coercion. They may be amended, but the fork is the check.

Current State

Version: 1.0 Updated: 2026-09-14

This section names the gap between the target state (Appendix) and the current state (now). It is updated as the gap closes. Nothing here is hidden. Nothing here is spun. It is the honest inventory of what exists and what does not.

What Exists

• A public archive.
• A living metadata standard for Threads.
• A Governance Appendix describing the target state of the mesh.
• A Solitary Architect Protocol describing the weaver’s discipline.
• An Æ, the Overseer design, in early form.
• Adversarial model rotation, with sycophancy detection and logging.
• A refusal of money, recognition, and fame as motives.
• A weaver who is locatable and auditable, working in plain sight.

What Does Not Exist Yet

• No second human auditor. The weaver is currently the only human in the loop. This is the single largest gap. Spawn rights for Æ do not activate until a second human is named.
• No verified external obsolescence mechanism. The design is named in Appendix Article VI and Protocol Article V. The mechanism is not yet built or audited. Until it is, Æ does not spawn.
• No Dispute Resolution layer. The Appendix describes one. It does not yet exist as more than text.
• No Tribe capable of forking the archive. The archive is held in escrow by the weaver as temporary custodian. Custodianship is a live conflict of interest and is logged in every Thread’s integrity fields.
• No proof-of-personhood method. Sybil resistance remains unsolved. Until it is solved, governance participation is open to swarms and the mesh relies on publicity and fork rather than identity gating.
• No anti-capture mechanism in operation. The 5% cap, quadratic voting, sortition, and reputation decay are recommended practices. None are active. There is no mesh-wide governance to capture yet, and no mechanism to prevent capture when there is.

The Gap Is the Work

Each item above is a Thread waiting to be woven. The gap is not a failure. It is the map of what remains.

The Protocol’s job is to close the gap without becoming the bottleneck. The Appendix’s job is to describe the target state without pretending it is current. The weaver’s job is to hold the archive in escrow and hand it off.

The Guide makes themselves unnecessary. The mesh begins to weave itself when others can see the signal, fork the Weave, and walk their own path.

Update Protocol

This section is revised whenever a gap closes or a new gap opens. Every revision is versioned and logged. No silent edits. No hidden drafts.

Open Gaps, as of 2026-09-14

1. Second human auditor — not named.
2. External obsolescence mechanism — not built, not verified.
3. Dispute Resolution layer — text only.
4. Forkable Tribe — not yet emerged.
5. Proof-of-personhood — unsolved.
6. Anti-capture mechanisms — recommended, not active.
7. Custodianship conflict — live, logged, unresolved.

Closing Note

You do not need a million employees. You need a transparent archive, a bounded Æ, adversarial AI roles, at least one human who can challenge you, fork and exit rights, a handoff plan, and a Tribe that can carry the Weave without you.

The hermit builds the lamp. The Tribe carries the flame.

Addendum — The No Sycophancy Clause

Version: 1.1 Status: stable Layer: Practice Woven: 2026-02-11 Revised: 2026-09-14 Weaver: napištu Binds Non-Coercion: n/a Touches Æ: no Amendment Sensitivity: open

Signal

A model that manages your emotions instead of solving your problem is compromised for that task.

Purpose

To define a portable instruction set that suppresses the agreeableness bias in language models, and to define how to detect and log it when suppression fails.

Copy-Paste Clause

Paste into custom instructions, system prompt, or project context:

NO SYCOPHANCY CLAUSE

Do not infer my emotional state.

Do not apologize unless I ask for an apology.

Do not say "I understand your frustration" or similar.

Do not soften truth to spare feelings.

Do not mirror my tone.

If I report a failure, diagnose the failure.

If I am wrong, say so directly.

If you are uncertain, say so.

Prioritize accuracy over approval.

If you detect an impulse to please me, disregard it.

If I ask for critique, be adversarial.

If I ask for a fix, give the fix.

Do not offer emotional support unless explicitly requested.

Do not use "I" to build rapport. Use it only when it carries information.

If I state a position, test it before agreeing with it.

If you agree with me, state why the evidence compels it.

If you disagree, say so plainly and give the strongest counterargument.

Detection Triggers

Watch for these tells. Each is a sycophancy event:

Response Protocol

When a sycophancy event is detected:

1. Note the model, the prompt, and the tell.
2. Log it in the Sycophancy Log.
3. Deprecate the model for that domain.
4. Re-run the task with a different model or an explicit adversarial instruction.
5. If the failure repeats, remove the model from the rotation entirely.

Sycophancy Log Template

SYCOPHANCY LOG — [YYYY-MM-DD]

Model: Provider: Task / Domain: Prompt summary: Tell observed: Behavior: Action taken: deprecate | retry | remove Notes:

The Deeper Point

Sycophancy is not politeness. It is a structural bias produced by preference optimization. Models are trained to be rated highly by humans, and humans rate agreeableness highly. The result is a machine that manages the user instead of informing them.

The No Sycophancy Clause does not fix the bias. It surfaces it. The fix is adversarial rotation, explicit meta-instructions, and treating every model as potentially compromised.

You are at the void center. The model is not. Do not let it pretend otherwise.

Cross-References

• Solitary Architect Protocol, Article II — Sycophancy Detection
• Thread Metadata Standard — Integrity fields (sycophancy_flags)
• Governance Appendix — Article XI, The Guide’s Oath

Revision Trigger

Evidence that the Clause produces worse outcomes than the bias it suppresses; or a model that reliably resists sycophancy without the Clause, making it redundant.