How to read someone in one line
Everyone reads as one phrase: <tier> <classifier> (<scope>).
Blake Emal is a lead captain (Megaphone). Venus is an independent ambassador. Charlene is a directed ambassador (Jeffrey replies).
That is the whole model. Two words: the classifier says what human function they hold, the tier says how much of it they carry alone. It works like “senior engineer”, one title made of two parts.
Michael sits above the ladder entirely, at owner, in all five buckets at once. He is not
a peer of the leads and is never described as one.
Only the word lead means a function is covered. owner does not, and everything below
lead is capacity.
Canonical taxonomy for what a human is FOR at Rye Valley in an agent-heavy business.
Source: Michael’s framework, 2026-08-22. This is orthogonal to job title. A person’s
role: says what they do; classifiers: says which irreducibly human function they hold.
Every page in people/ carries a classifiers: list in frontmatter. Values must come
from the five below, lowercase, hyphenated.
The five
ambassador
The human face. Customer, prospect, partner, and capital-facing relationship work. Premise: the last place a person wants an AI is when they are frustrated, being sold to, or being asked for money. Covers sales, partnerships, account management, support for paying humans, interviewing, and fundraising.
context-farmer
Builds and maintains the system with product judgment. Feeds the right information, in the right format, into the right system, in the right order, at the right time. Splits into a senior half (system architecture) and a sourcing half (subject-matter research, which does not require company or startup knowledge).
captain
Senior owner of a domain. Knows the quality bar, the edge cases, and the second-order effects. Makes sure the system does what it is supposed to do for the end user: evaluates outputs, tweaks constraints, recognizes when agents go sideways. At Rye Valley this maps to vertical leads.
referee
Authority to stop play. A human QA layer across the whole business, not just code. Taste applied in a polishing and critical context rather than a steering one. Not everything gets reviewed; these do:
- subjective or brand work where “correct” is fuzzy
- work with no clear definition of done
- edge cases the system has not been taught
- high-stakes output where a miss is expensive
- calibrating whether the current quality bar is still right
steward
Capital allocator. Budgets for compute and other irreversible spend decisions that affect the health of the business: burn rate, fundraising, shutting projects down. In most companies this is the founder for a long time, possibly forever.
Rules
- Multiple classifiers is the goal, not an exception. The best team members are context farmer, captain, and referee at once. That combination IS the modern definition of seniority. Never treat overlap as a conflict of interest or propose splitting a person’s buckets on separation-of-duties grounds. Michael has explicitly ruled on this (2026-08-22).
stewardstays separate. It is the one bucket you deliberately do not merge into the others.- Classifiers are per-person, not per-vertical. A captain of one vertical is not automatically a captain elsewhere.
- Do not infer a classifier from a job title. Assign only from observed
responsibility. If unsure, use
classifier_status: proposedand raise it with Michael rather than guessing silently. - Former and inactive people carry
classifiers: []andclassifier_status: n/a-former.
Tiers (the depth ladder)
A classifier says WHICH human function someone holds. A tier says HOW MUCH of it they carry alone. Both are required; neither substitutes for the other. Four people can share one classifier and still leave the bucket one deep, because only one of them is a lead. The tier field exists so a headcount in a bucket never gets misread as coverage.
Tiers apply WITHIN a classifier, not across the company, with one exception: owner is
absolute and applies everywhere. An independent ambassador and an independent captain
are not comparable in scope or pay, only in autonomy.
owner
Michael, and only Michael. Owner applies to all five classifiers, always. He is owner ambassador, owner context-farmer, owner captain, owner referee, and owner steward simultaneously, including in verticals he is nowhere near day to day. This is not a claim about how he spends his time; it means the authority in every bucket terminates with him.
Where a lead’s authority is final against everyone but Michael, the owner’s is final full stop. Leads are not his peers and should never be described as being at his level. He outranks a captain inside that captain’s own vertical and the referee on the referee’s own call, by default, without needing to hold the domain knowledge that lead has.
Because owner is universal, it is the one tier that is not scoped: there is no such thing as owner of one vertical, and no one else ever holds it.
Deliberate exception to how tiers are read: owner does not count as coverage of any
bucket. Not because the authority is lesser, but because coverage asks whether a
function survives without the founder, and answering that with the founder is circular.
Michael holds all five classifiers; all five are still uncovered unless a lead holds them
too.
lead
Owns the function. Sets the quality bar rather than meeting one, absorbs novel and
ambiguous situations, and is accountable for second-order effects. If this person is gone
for a month, the function degrades and nobody else can simply cover it. This is the only
tier that constitutes real coverage of a bucket. Ranks below owner.
independent
Executes the function unsupervised on known work, and can be trusted with a live account, client, or system without review. Handles variation within a defined scope but escalates genuinely new situations rather than ruling on them. Can train and brief others at their tier or below. Real capacity, not real cover: an independent can absorb volume for a lead, never replace one.
directed
Executes a defined queue or a scoped surface. Work is bounded, instructions are supplied, and novel situations escalate by default. May be new, may be part-time, may be deliberately narrow. Adds throughput, adds no resilience.
n/a
Inactive or former. Carries tier: n/a with no scope note.
Authority (the third axis)
Classifier says WHICH function. Tier says HOW MUCH of it someone carries alone. Neither says WHO DECIDES when two people disagree. That is authority, and it does not track the other two.
Michael holds final say in every vertical and every type of work. He outranks every
captain inside their own domain, every referee on their own call, and every lead on their
own system. This is recorded as the owner tier, which he holds across all five
classifiers at once, and it is true whether or not he is close to the work. Nobody else
carries an authority marker: everyone else’s is implied by their tier.
Two consequences follow, and the second one is the one that gets forgotten.
Everyone else’s authority is bounded by their classifier and tier
A captain’s ruling is final within their vertical, against everyone except Michael. A referee can stop play on anything in scope, against everyone except Michael. An independent decides inside their scope and escalates outside it. Nobody has authority that crosses into another person’s classifier without going through Michael.
Frequent override is a signal about the person, not about the system
The reason to have captains is that they know more about their vertical than Michael does and can be trusted with second-order effects. The reason to have a referee is that somebody can stop play.
When Michael repeatedly reverses a captain or referee inside their own domain, the correct reading is NOT that the system is degrading or that he should defer more to protect the role. It is that this person’s judgment is failing and they should be demoted out of the classifier or let go. He appointed them; a persistent pattern of bad calls is information about the appointment, and the fix is to change who holds the seat, not to preserve the seat by deferring to calls he thinks are wrong.
A captain who has been reversed enough to stop making calls and start asking has already ceased to be a lead. Record that on the roster rather than pretending the bucket is still covered.
Every override is considered
There are two reasons Michael overrides, and both are exercises of judgment, not slips:
- He has context they lack from another part of the business (a client conversation, a financial constraint, a strategic direction not yet shared). Supply the context where it helps, since the gap may be worth closing.
- He has the same information and disagrees. This is equally legitimate. He weighed their judgment against his own and did not defer.
In both cases he considered deferring to their judgment and chose not to. Do not treat an override as evidence of a process problem, do not advise him to defer on taste grounds, and do not describe his overrides as a cost to be minimized or spent. If a pattern of overrides is worth surfacing at all, surface it as a question about whether the right person holds the classifier.
Reading tiers and classifiers together
The honest coverage question is never “how many people hold this classifier,” it is
“how many leads hold it.” A bucket with one lead and three directed people is one
deep. Report it that way. Michael at owner tier never fills the lead slot in this
count.
When a bucket has exactly one lead, name the risk explicitly rather than reporting the headcount. When a bucket has zero leads, that is an unowned function regardless of how many people touch it.
Escalation between tiers is expected and good; it is how a directed person becomes independent. Do not treat a scope limit as a criticism of the person.
Frontmatter schema
classifiers:
- captain
- referee
classifier_status: confirmed # confirmed | proposed | n/a-former
classifier_updated: 2026-08-22
tier: lead # owner | lead | independent | directed | n/a
tier_scope: "What they carry alone, and the boundary."
tier_updated: 2026-08-22tier_scope is optional but strongly preferred for anyone below lead, and required
whenever a lead’s authority is bounded: a captain of one vertical only, or a lead who
works through a vendor relationship rather than equity.
Current assignment (2026-08-22)
| Person | Designation | Status |
|---|---|---|
| Michael Houck | owner, all five buckets | confirmed |
| Blake Stewart | lead ambassador | confirmed |
| Venus | independent ambassador | confirmed |
| Reina Beltran | independent ambassador | confirmed |
| Charlene | directed ambassador | confirmed |
| Ben Ornstein | lead referee | confirmed |
| Blake Emal | lead captain (Megaphone) | confirmed |
| Tatem Lindner | lead captain (JOLT!) | confirmed |
| Emma Sivess | lead captain (Lore, white-label only) | confirmed |
| Rafael Bruno | lead context-farmer | confirmed |
Inactive and former people carry classifiers: [] / classifier_status: n/a-former and
are omitted here (incl. Mehul Jariwala, inactive).
Coverage by leads (the number that matters)
Michael holds all five classifiers at owner tier and is excluded from every lead count by
design (see the owner tier above): the question is what survives without the founder.
| Classifier | Leads | Depth |
|---|---|---|
| ambassador | 1 (Blake Stewart) | One deep. Venus and Reina independent, Charlene directed: capacity, not cover. Michael handles capital and brand himself, at owner tier. |
| captain | 3 (Blake Emal, Tatem, Emma) | The only genuinely covered bucket. Emma’s lead status runs through a white-label vendor relationship, not equity. |
| context-farmer | 1 (Rafael) | One deep, and Rafael is fully loaded on engineering, so Michael does it himself in practice. |
| referee | 1 (Ben) | One deep. |
| steward | 0 | Owner only, by design and permanently. |