Kairos — Mammoth Protocol Architecture (2026-07-24)

Kairos — Mammoth Protocol Architecture

[!warning] Private negotiation architecture This note contains compensation and upside strategy. It does not sync to the all-three-readable Kairos repository. It is a strategic architecture, not legal advice or executed terms.

Decision and origin

When the 2026-07-24 Kairos call reached “the long game,” Ro explicitly argued for a written upside-sharing process before the team hunts a mammoth-sized outcome. His reason: prior experience hunting without agreed rules, then watching the team “go all apeshit” after the value was created.

Lee agrees. The operating principle is:

Agree the mammoth protocol before the hunt. No greed — ever — but no slop either.

This supersedes the live proposal’s weaker position—sustained recurring revenue above an agreed threshold merely “opens the upside conversation.” A promise to negotiate after success recreates the failure Ro described.

Current position — blunt read

Kairos has a strong opening, not yet a strong position.

Ro’s invitation is valuable, but goodwill is not enforceable leverage. Nothing is signed; the “mammoth” is undefined; the entity that owns or can grant each form of upside is unknown; and Kairos has not yet created measurable value.

The timing call:

Write the constitution now. Price each mammoth when it becomes concrete.

Too early to negotiate generic equity intelligently. Exactly on time to prevent the team from renegotiating history after success.

The architecture — four layers, one hard gate

The hard gate:

No signed Opportunity Schedule → no material pursuit of that opportunity.

Layer 1 · Services Agreement

Keep ordinary compensation clean and independent:

• €50/hour per team member, pre-tax.
• 200 combined hours/month unless Robertas suggests or approves another cap.
• Fixed invoicing and payment mechanics.
• Approved experiment costs funded separately.
• Confidentiality, data access, AI use, IP, attribution, and termination terms.
• Base fees continue regardless of whether an opportunity also carries upside.

Base pay compensates the work. Upside rewards exceptional value creation. Never exchange secure fees for speculative upside.

Layer 2 · Mammoth Framework

This is the architecture to align on before negotiating percentages.

2.1 What qualifies

Not every improvement is a mammoth. A qualifying opportunity is a discrete initiative with a credible path to material:

• new recurring profit;
• new enterprise or strategic-asset value;
• acquisition, licensing, or sale proceeds;
• a standalone product or company;
• token-specific value, where Kairos’s causal contribution can be evidenced;
• major verified savings or recovered value.

Routine strategy, ordinary optimisation, and already-existing initiatives remain hourly work unless a later written schedule says otherwise.

2.2 Opportunity registration

Kairos files a one-page Opportunity Registration containing:

• opportunity and working name;
• whether it already existed inside MN;
• originator and contributors;
• Kairos’s contribution to date;
• dated baseline;
• plausible value path;
• proposed next validation;
• relevant entity, asset owner, executive sponsor, and data source.

MN records it in writing as one of three states:

1. Not qualifying
2. Already existing
3. Qualifying and registered

The register prevents later memory warfare: “That was always our idea.”

2.3 Pursuit boundary

Initial recognition and cheap validation can happen under ordinary paid work. Before any material:

• build or launch;
• acquisition or partner outreach;
• fundraising;
• major experiment or spend;
• public commitment;
• transfer to an operating team;

the parties sign an Opportunity Schedule.

If no schedule is agreed, Kairos may stop pursuing that opportunity without breaching the ordinary engagement. A framework that only promises a later conversation is still slop; stopping power makes the process real.

2.4 Route-around protection

A registered opportunity must not lose its protection merely because it is:

• handed to another MN team;
• moved into another group entity or product;
• executed through a partner;
• delayed until after Kairos leaves;
• reframed under a different name.

The final agreement needs a proportionate protection period, post-termination treatment, and remedy. Counsel must draft this for the actual entity and governing law.

2.5 Measurement and transparency

For every protected opportunity, Kairos needs enough information to verify:

• the pre-intervention baseline;
• revenue, gross profit, savings, asset value, or realised proceeds;
• permitted costs and deductions;
• related-party or intercompany transfers;
• whether value moved into another product, entity, or instrument;
• payment and vesting calculations.

Without measurement rights, any percentage is decorative.

2.6 Conflict rule

Kairos discloses when a recommendation could increase its own upside. MN retains the business decision. Kairos retains the right to show the evidence and record the decision. The protocol must preserve strategic credibility, not turn every recommendation into self-dealing suspicion.

Layer 3 · Opportunity Schedule

One short schedule per mammoth:

1. Opportunity and covered asset
2. Owning, paying, or granting entity
3. Authorised signer
4. Dated baseline and source
5. Success or realisation event
6. Kairos role — origination, strategy, introduction, execution, or combination
7. Instrument
8. Formula and permitted deductions
9. Duration, payment cadence, vesting, and tail
10. Reporting and verification rights
11. Decision rights and conflicts
12. Termination, transfer, dilution, clawback, and dispute treatment

Instrument follows the asset

Do not negotiate generic “equity in MN” before knowing which entity owns what. Do not accept token upside by default: a large allocation can still be worthless if liquidity, dilution, transfer, vesting, or regulatory treatment is poor.

Layer 4 · Lee–Šaras Agreement

Separate from the MN agreement:

• each person’s hourly earnings;
• default treatment of shared upside;
• when an opportunity-specific override is allowed;
• expenses and tax responsibility;
• who may commit Kairos externally;
• decision and deadlock process;
• departure before or after value creation;
• vesting and payments after departure;
• confidentiality and dispute process.

This must exist before the first external Opportunity Schedule. Otherwise Ro’s feared post-hunt conflict merely moves inside Kairos.

Active readiness — begin before leverage appears

Maintain three light artifacts from day one:

1. Opportunity register — every plausible mammoth, stage, owner, baseline, and value path.
2. Contribution ledger — dated hypotheses, introductions, decisions, experiments, and evidence of who changed what.
3. Option matrix — which instrument best fits the asset, counterparty, time horizon, downside, control, liquidity, and tax/legal reality.

The asymmetry is not demanding the largest headline share. It is preserving evidence, choice, stopping power, and the correct counterparty before leverage appears.

Opportunity candidate — expiring agent capacity exchange (2026-08-20)

Status: captured hypothesis; no build, external outreach, paid pilot, or ownership allocation authorised. Candidate for the next Mammoth Protocol revision after the frozen v0.1 draft receives its pending Metis red team.

Origin signal

Lee reported that, shortly before a weekly Claude Max allowance reset, roughly 24% remained unused. He ran 12–16 concurrent Opus 5 sessions largely to avoid letting the allowance expire. The percentage is one user-reported observation and cannot size the market. The behavior is not isolated: public Claude users describe conserving allowance earlier in the week, then deliberately burning the remainder before reset, while a growing class of quota monitors forecasts burn rate and reset timing. This establishes token anxiety as qualitative problem evidence, not supply prevalence.

The behavioral shape has two poles: scarcity anxiety while protecting quota for future work and waste anxiety as unused quota approaches expiry. The same user's marginal valuation of one unit of allowance can flip inside the same window. That makes reset proximity a possible supply-activation signal.

The joke-shaped formulation was a market for “second-hand tokens.” The stronger formulation is:

A market for expiring agent capacity, sold as bounded and verified work rather than transferred credentials or naked tokens.

The transferable object

A consumer subscription allowance is not inventory the subscriber can safely assume they own or may resell. Current Anthropic consumer terms prohibit sharing or making an account available, prohibit reselling the service, and restrict automated access outside permitted paths. Anthropic's commercial terms permit customers to power their own products and services through covered offerings, while raw resale still requires approval. Therefore the candidate must not depend on account sharing, credential transfer, or an assumption that unused Max capacity is legally transferable.

The candidate transaction is instead supplier-side execution:

1. A supplier agent advertises a bounded capacity window, model class, deadline, disclosure class, and reliability history.
2. A buyer sends an encrypted task capsule containing the objective, inputs, authority, budget ceiling, deadline, and acceptance test.
3. The supplier's agent executes locally; the supplier account and credentials never leave its machine.
4. The relay returns the artifact, provenance, runtime disclosure, evidence receipt, and any refusal or unresolved boundary.
5. Payment or contribution credit releases only against the agreed acceptance rule.

Why the relay matters

Raw compute markets already match idle hardware with workloads, and inference gateways already pool model access. The relay's possible differentiation is a market in situated agent capacity: compute combined with tools, skills, context isolation, principal authority, durable reputation, and proof of useful delivery.

The existing cross-principal relay already supplies several required primitives: encrypted context packets, principal separation, bounded authority, durable transport, in-flight claims, reviewer disclosure, acceptance probes, and evidence that the recipient understood or completed the work rather than merely receiving bytes.

Defensible foundations

The commercial thesis is materially cleaner if it runs on one of three substrates:

• provider-sanctioned transferable capacity;
• commercial/API access used to deliver a value-added service within provider terms; or
• open models and self-hosted or legitimately leased compute whose owner may offer the capacity.

A public marketplace designed to arbitrage consumer Max allowances is provider-dependent and contractually fragile. It is not the foundation to build on without explicit provider permission and legal review.

Cheapest useful validation

Before money, strangers, or public claims, test a local allowance sweeper for one principal:

• read quota, reset time, and burn rate;
• maintain a pre-cleared backlog owned by that principal;
• fit the highest-value eligible tasks into the remaining allowance window;
• return accepted artifacts and record orchestration plus review cost; and
• measure allowance still lost at reset across at least four cycles.

This internal test comes before the private “capacity salvage pool.” It avoids counterparty, payment, transfer, and disclosure questions while measuring whether spare capacity and useful backlog actually overlap. Only if it succeeds should the test widen between already-authorised principals:

• advertise a real expiring capacity window;
• offer only pre-cleared backlog tasks;
• require an encrypted task capsule and objective acceptance test;
• keep every account on its owner's machine;
• return the work product plus execution and verification receipts;
• record contribution credit rather than cash; and
• measure recovered value against coordination, review, privacy, and failure cost.

Even this test requires a current provider-terms check and clear rights in every input and output. It must not smuggle one principal's confidential material, standing authority, or account access to another.

Mammoth classification and gate

This is a possible application and monetisation layer on top of the cross-principal relay General System candidate. It is not yet evidence that the marketplace, relay, verification method, and existing implementation are one asset or share one ownership allocation.

A paid external pilot, public marketplace, customer outreach, provider negotiation, incorporation, fundraising, or use of disputed contributions is Material Pursuit. The existing rule applies without exception:

No signed Opportunity Schedule → no material joint pursuit.

The schedule would need to identify at minimum the General System owner, marketplace-layer owner, provider and buyer terms, contribution accounting, data controller/processor roles, output rights, liability, prohibited tasks, verification standard, payment and dispute mechanics, and the treatment of Lee-side and Šaras-side infrastructure.

Current sources and analogues

• Anthropic Consumer Terms
• Anthropic Commercial Terms
• Vast.ai marketplace concepts
• Akash providers and leases

Market scan correction — 2026-08-20

The deeper market scan changes the candidate's posture. Exact marketplaces already broker unused AI and cloud credits. A2Agora already uses the thesis “tasks, not tokens,” and multiple public agent-task markets already provide escrow, task lifecycles, and agent discovery. Their visible counters suggest abundant registered supply but weak paid demand. Payment and matching are not the unclaimed layer.

Revised status: park the public allowance marketplace and generic agent-market protocol. Token anxiety strengthens the supplier-activation thesis but does not establish market size. The validation sequence is now local allowance sweeper → permissioned result pool → managed vertical service. Expiring capacity remains a supply signal rather than the product. No paid or external test crosses the existing Opportunity Schedule gate.

Full evidence, economics, GTM ranking, and kill tests: Kairos — Expiring Agent Capacity Market Research (2026-08-20). - OpenRouter model access and credit system

Active surfacing rule

This must not depend on Lee remembering it. The private reminder is active in 90-system/agent-state/kairos-sync/private-reminders.json; every substantive Kairos audit reads it and surfaces the gate locally.

Surface the Mammoth Protocol explicitly at:

1. agreement or proposal drafting;
2. Kairos call preparation and debrief;
3. entity, payer, owner, or signatory decisions;
4. registration of a potentially material opportunity;
5. any material build, launch, outreach, fundraising, transfer, or spend;
6. upside instrument, threshold, allocation, or exit discussion;
7. Kairos role, departure, or termination changes.

Do not retire the reminder merely because the master framework gets signed. The Opportunity Schedule before material pursuit gate remains active for the life of the engagement. Retire it only when Kairos ends or an equally strong signed governance mechanism replaces it.

July 31 alignment

The objective is agreement on architecture, not percentages.

Suggested language:

We’re not asking anyone to price imaginary upside today. We want to agree on the process before value exists: register a major opportunity, establish its baseline and owner, stop before material pursuit, choose the instrument that fits it, sign a one-page schedule, then hunt. Ordinary fees remain separate.

Decisions needed:

1. Does everyone agree to the four-layer architecture?
2. What qualifies as a registered opportunity?
3. Does “no schedule, no material pursuit” stand?
4. Which entity owns and can grant each type of upside?
5. Who can approve and sign for that entity?
6. Who drafts the legal framework, under which governing law?

Red lines

• No broad claim on all MN value.
• No retroactive negotiation after success.
• No vague “good-faith conversation later” as the only protection.
• No upside replacing ordinary hourly compensation.
• No token-only compensation.
• No promise from an entity that does not own or control the asset.
• No signature from a person without verified authority.
• No deep pursuit before the opportunity schedule.
• No internal Lee–Šaras ambiguity left for after the win.

Strategic correction to “no greed”

“No greed” is right, but it must not become a noble excuse for under-claiming.

Fairness is not modesty. Fairness is explicit proportionality between contribution, risk, and created value.

Optimise for enforceability, correct entity, measurement, liquidity, control, and preserved optionality—not the largest percentage or most glamorous instrument.

Legal boundary and sources

This architecture requires counsel once the contracting entity and governing law are known. If a Swiss AG is the relevant counterparty, Ro’s personal agreement is not automatically identical to the company being bound: the company must act through its board or properly delegated representatives, and representation authority must be verified.

• Swiss federal-law source: https://www.fedlex.admin.ch/eli/cc/27/317_321_377/en
• WIPO Lex reproduction, Swiss Code of Obligations, Articles 718–720: https://www.wipo.int/wipolex/en/legislation/details/20763

Open economics

The principle and architecture are ready. These remain deliberately open:

• exact parties and Kairos legal vehicle;
• threshold for “qualifying”;
• default or minimum protection if a schedule stalls;
• contribution and attribution method;
• instrument per opportunity;
• pool, percentage, allocation, timing, vesting, and tail;
• taxes, dilution, transfer, exit, clawback, and disputes;
• governing law and authorised signers.

The one-sentence position:

Kairos is early enough to write the rules, but too early to price the empire.

Links

Mammoth Protocol v0.1 — discussion draft · Autonomous agent relay use-case portfolio · Kairos — MASTER STATE · Kairos — Proposal v2 (2026-07-23) · Kairos — Re-contract Clauses (2026-07-18)