14th & Co. · a follow-up to the Operating Partner proposal · Bentonville
Twelve playbooks describe the operating week. Eight declared skills write the records. The data model shipped before anyone asked for an interface — so this specifies two screens and nothing else: an operator surface where the OP, the residents and their agents do the work, and the Friday view, one published read-only current view that is the only screen anyone in Tulsa is ever asked to look at. The sections below are live: play the operating week, change who is looking and watch the register redact itself, and try the interlock that cannot be overridden.
This document has one open item that outranks its own contents, and it is not a gap in the design — it is a fact I could not establish from the outside. It renders here in exactly the grammar the system uses for anything waiting on a decision, because that is what it is.
FIG. 00 — THE ENTRY CONDITION. Rendered as an open decision rather than a footnote, because this document holds itself to the rule it proposes for every other surface: a claim with an unresolved assumption behind it renders as unresolved, with the assumption named. This one is unverified, and only you can close it.
The second boundary is drawn in the diagram below rather than asserted. The method — embed, observe, synthesize, concept, validate, kill or spin out — is fifteen years old, it is yours, and no step of it is altered, invented, renamed or automated anywhere in this document. What is specified is a calendar, a register, a decision log, and two screens that show them. In FIG. 01 the method sits in its own band, dashed and outside the specified boundary; every writer below it produces a record as a byproduct of a step that was already happening. If any line below reads as a change to how you run discovery, I have drawn it wrong and want to know.
This was written before I’d spoken to anyone at Gitwit. Two conversations since have confirmed some of it, corrected one thing that matters, and made one section more speculative than it reads. Rather than quietly editing, here is the diff.
I invented an operating week and a Friday publish. You already have a forum, and it has a name.
Exploration Weekly — the whole team works the pipeline: what gets explored next, on what timing, which ones you go in on and which you don’t, and why. Anyone can submit an idea. That is a better mechanism than the cadence I designed, and it already exists.
The principal surface stops being a broadcast and becomes an input. Its job is not to publish state outward to a distant Tulsa on a Friday. It is to make Bentonville’s week land in Exploration Weekly, and to hold what happens to a signal between sessions.
It also softens the latency thesis, honestly. A weekly all-team pipeline session is already a low-latency mechanism. The remaining latency isn’t “Tulsa is slow to decide” — it is the gap between the day something is found on a floor and the session where it gets discussed, and whether it arrives as a settled read or as a story someone half-remembers.
| What I learned | What it changes |
|---|---|
| Launch479 / Renaissance Philanthropy — several new ventures based in Bentonville each year, founding teams built here | The spec assumed one cohort. The volume is recurring, which makes the record more valuable and the informal version less survivable. |
| Yask is headquartered in the Bentonville office, founding team in place, first customer piloted | The operator surface has a live venture team co-located from day one. Residents watch launch and scale happen in the room while they work pre-build and build. A real advantage the spec didn’t know it had. |
| The four phases, named: pre-build → build (Gitwit owns both) → launch → scale, handed to a hired trio — CEO, product leader, VP engineering | The handoff is a specific, dated event with a known receiving team. Every record has to be legible to a trio that wasn’t there for phases 1–2. |
| “All the partners are hands-on. We are in the weeds every day.” | No surface here should assume a partner reads summaries. The register has to be worth opening by someone who was in the work yesterday. |
The grant/formation interlock is the least evidenced thing in this document. It came from my own reasoning about resident economics, not from anything Gitwit has told me. The tax mechanic is real; whether it is your mechanic I don’t know.
Read it as a hypothesis about where an irreversible date-collision would live, not as a description of your model. If residency economics work differently, that section is wrong and the rest of the spec survives without it.
FIG. 01b — A SPEC THAT COMPOSES AGAINST NEW EVIDENCE. The point of writing it down early was never to be right first time. It was to have something specific enough to be corrected by two conversations — which is what just happened, and which is cheaper than being vague and staying that way.
FIG. 01 — THE BOUNDARY, DRAWN. The dashed band is Gitwit's method; nothing inside the solid boundary touches it. Every writer produces its record as a byproduct of a step that was already happening — which is the whole design rule, and also what keeps this inside the line. ventures* is derived — there is no declared venture record, and both surfaces render it thin on purpose (G-1). The record paths and their writers were declared in Phase 5, before this phase existed — so the record does not need a schema, it needs to render the one already declared. Zero new fields were invented. That is the structural difference from the first attempt, which invented eleven surfaces and then went looking for something to put in them.
Nobody here is a "user of a tool." The difference between a resident, the operating partner and Tulsa is not an access level — it is what each one is allowed to settle. Agents draft and sweep; only a human rules; only the OP publishes; only Tulsa decides. That is the whole model, and it fits on three cards.
Nothing, alone. A resident's working set is isolated — drafts and agent output are theirs, invisible to the cohort surface and absent from Tulsa. Machine assist sits on synthesis and drafting; an agent can write every sentence and still cannot supply the witness.
The operator surface. Registered artifacts become cohort-visible so cross-track pairs can review each other. Bar evidence never is.
Two things no agent holds. The review ruling — a human judging an artifact complete, recorded as an event — and the Friday publish. Until the ruling is recorded, the artifact is a draft no matter how finished it looks.
Full read/write. Front door is the week plus what is waiting on a human; every panel is a drill-down behind it.
Every decision they already retained. Nothing is relocated. Rights are fixed; latency is designable — so what they lose is not authority, it is having to ask where things stand.
One published view, fortnightly rhythm, arriving as a link in the Friday memo. They never browse and never see raw working state.
FIG. 02 — THREE SEATS, ONE RECORD. The bar is written down precisely so people can be measured against something they have read — which is why a resident can read their own record in full, and why bar evidence is deliberately withheld from Tulsa. Tulsa gets dated status events; that is what a naming decision actually needs.
Press play. This is the operating week unchanged — Monday field, Tuesday synthesis, Wednesday studio, Thursday outside, Friday Tulsa — with the record each day already produces named. Nothing on this timeline is an extra task. Each row carries its mode: ▸ automated · ◇ ai-assisted · ● human ruling.
FIG. 03 — THE WEEK, PLAYED. The Tulsa column stays empty for four days of continuous studio activity. That emptiness is the design. And the diagnostic falls out for free: if a Monday passes and no partner or informant record was written, the field day did not happen — which is the studio's own red threshold, visible on a Tuesday morning instead of at a quarterly review.
This detects absence, not quality. A record written to keep a counter green is worse than no record, and no interface catches that. The rule that holds and is not automatable: no insight enters an artifact without a named human who was in the room.
Not a note on a record — a projection rule the surface enforces. Every artifact carries a tier from the moment it is drafted, defaulting to the most restrictive; publication is an affirmative decision, never an absence of objection. Change who is looking and watch the register redact itself. Click any row for its provenance.
FIG. 04 — THE SAME REGISTER, THREE VIEWERS. Two consequences are the point of doing it this way: a champion-only finding cannot reach Tulsa by accident, and the most expensive mistake in the studio's whole risk model becomes structurally difficult rather than a thing to remember. Friction spent here protects the champion and the informant — the two actors with the least protection in the model.
One published, read-only view. It is a human action on Friday, in the slot where decision memos already get filed — one action, two outputs. It travels as a link into the Friday memo or the review invite. Projections, not portals: nobody in Tulsa is asked to log in and browse.
#gitwit-studio · Bentonville — the operating record · Fri 16:40 — The Friday view for 14 May → open the full render
FIG. 05 — THE FRIDAY VIEW. Paper, because it is an issued artifact — printable for a room with no wifi. Note item 7 renders even when green: it is the one thing Tulsa should be able to see is holding without asking. And every closed row includes the nos — a wins-only register teaches the same lesson a wins-only artifact library does.
The date promised to the counterparty. Not the date opened, not the age. It is the difference between "I'd have to check" and "the 14th." Decision rights are fixed; latency is designable — so a published current view removes the class of request that is only ever about status, and leaves the escalation ladder for the decisions it was built for. Tulsa keeps every right and loses the waiting.
Tuesday: transcripts to code, sessions to synthesize, a brief to draft with machine assist. Your working set is invisible — not on the cohort surface, absent from Tulsa. Wednesday the OP reads it against the standard and records the ruling. Only then is it real.
The feel: uncommitted work genuinely does not count, and that is correct. What you do get is your name on it — named at commission, not at publication — and a byline that reaches Tulsa on Friday without anyone forwarding anything. You can also read your own record in full, bar evidence included.
No progress bars, no percentages, no ETAs — there is no honest source for any of them. Every panel shows last written, by which named human. A stale panel looks stale.
Claims traced, evidence tags present, author named at commission. Returned to informants 12 May; champion seen 13 May — so it is eligible for a tier.
■ DRAFTEDTier 2, class investment + venture, owner Jacob. The counterparty was told the 20th. That date — not the age — is what the Friday view sorts on.
■ OPENAssembled from the two items above. Champion-only material is excluded by the assembly, not by your memory. Jacob, Dan and Mike see exactly this.
■ STAGEDClick through the three items. The front door is the week plus what is waiting on a human — not a dashboard. Each item opens in place; the square settles when it is done. The publish is one action that produces two outputs: the memos get filed and the view goes out.
The feel: the record is a byproduct of the week, not a second job. An interface whose home screen tells you to leave the building — because the week's first day is nobody in the studio.
Friday 16:40 — the view arrives as a link in the memo you already read. Three decisions are waiting, each with the date the counterparty was given and the default that fires if the window passes. Nothing asks you to log in. Two weeks later the review opens on a decision, not a readout, because both sides have been reading the same screen.
The feel: your authority is untouched and your inbox is quieter. The test: no agenda item is "update."
Every rung of the escalation ladder is a request. This removes the rung that was only ever about status.
Exactly one blocking gate exists in this entire system, because advisory friction that fires constantly stops being read. Profits interests must be granted and papered before common stock is issued; QSBS triggers the day the stock is created. A grant papered one day late is QSBS-dead on that venture, for that resident, permanently — silently, and nobody notices for years. Set the two dates and see what the surface does.
Grants papered 21 Apr, formation 21 May. The tax clock is protected and the record accepts the date.
FIG. 06 — THE ONLY BLOCKING GATE. Set the grants to not yet papered and no formation date will record. It is the only failure in the studio's design that is irreversible, invisible at the time, and purely a date comparison — which is exactly what a machine should hold. Everything else in the system is advisory.
Not a workflow approval. A formation date that cannot move is still a Tulsa decision — no retained right is relocated anywhere in this spec. The interlock forces the conversation to happen; it does not decide it. And it renders in the Friday view even when green, stated to the counterparty in the form the playbook gives: the timing protects a tax clock that benefits them too.
The correct response is to delete surfaces, not add them — and to go find where the work actually leaves a trace, because that is where the record belongs. The governing constraint is not a principle, it is a post-mortem: if nothing writes to it during real work, nobody returns to it.
Every item in §05 can be produced as a Friday memo by hand. If the written version does not get read, no amount of rendering will fix it — so the demand gets tested before the surface gets built, which is the one thing the first attempt never did.
This is not a dashboard for a venture studio. It is a record of one studio running its own method — where the record is a byproduct of the week, the only hard gate protects a resident's tax position rather than the platform's convenience, and the people who retained every decision stop having to ask where things stand.
14th & Co. · prepared for Gitwit as a follow-up to the Operating Partner proposal · working draft at signaled — reasoned, not verified — six gaps open and named · the full Friday view render is here.
Nothing above was reasoned from a deck. It came out of the sixth phase of a coherence protocol I ran on the studio's design — eight phases, each one gated on the last, each emitting artifacts a later phase has to compose against rather than reinterpret. The value is not the documents. It is that later phases cannot quietly disagree with earlier ones — a contradiction has to be raised and settled, in writing, before the next phase is allowed to run.
FIG. 07 — EIGHT OF EIGHT. The adversarial pass has now run, and two of its three gates did not pass — seven criticals found in my own work, published unclosed rather than quietly repaired. The full record of the run, the rulings and the findings is here →
Each one is a separate record carrying its reasoning, the alternatives considered, and the condition that would prove it wrong. They are the reason the two screens above could be specified without inventing a single field: every surface renders something a decision already committed to.
FIG. 08 — THE LEDGER, AND WHERE IT SURFACES. The rows marked → are the ones you can see rendered in the sections above: 003 is the only blocking interlock, 008 is the entire reason the Friday view exists, 006 is why bar evidence stays off it, and 014 is why there is no progress-through-modules anything anywhere. A decision that never reaches a surface is a decision nobody has to keep.
Everything here is derived from your public model, the Tulsa Model material, the two Fellowship postings, and prior systems I have built twice before — not from field access to Gitwit, your books, or your residents. Evidence level is signaled — one of three the protocol calibrates to, and it means exactly this: reasoned and internally consistent, not verified against your operation. Raising it is not more writing; it is field access. Six gaps and two vocabulary conflicts are open and named. The fastest way to make any of it worth more is an hour with someone who actually runs the week — which is also the first thing I would do in the seat.