The question put to the court: Three people, eleven months of runway, and a $250k deal that only exists if the integration is live in six weeks: build the core pipeline in-house in three months, or ship it on a vendor in three?
The deal is an option wearing a contract: a zero-penalty walk clause, a champion in ops rather than finance, and a conversion promise that exists only out loud. The court's job was to price a six-week deadline against a thirteen-week estimate from the same three people who also have to sell, support, and survive the pilot.
CASE FILE № 4 · CONVENED JULY 15, 2026
The question put to the court: Three people, eleven months of runway, and a $250k deal that only exists if the integration is live in six weeks: build the core pipeline in-house in three months, or ship it on a vendor in three?
The deal is an option wearing a contract: a zero-penalty walk clause, a champion in ops rather than finance, and a conversion promise that exists only out loud. The court's job was to price a six-week deadline against a thirteen-week estimate from the same three people who also have to sell, support, and survive the pilot.
Take the vendor path, armored. Hit the deadline on rented extraction, but architect it so the vendor is a swappable engine underneath things you own: your schema, your validation rules, your corrections and audit trail, every raw output stored on your side, and product code that never touches vendor-native objects. The in-house rebuild happens only after the customer has signed AND paid, plus thirty days of production data, never on a Q2 promise. And nothing here is real until the bake-off: two or three days running the vendor and two alternates on about fifty of the customer's actual contracts.
| Option | Why it lost or won |
|---|---|
| A · killed on the calendar | Thirteen estimated weeks against roughly five remaining, from an estimate nobody has reviewed, executed by the same three people running everything else. Nobody on the record defended the math, and a dead deadline spends the company's only free quarter with zero revenue event. The door back has three locks, all in writing. |
| B · wins, armored | The cheap-to-reverse path, and the court is honest that cheap is not free: at full deployment the vendor's meter runs to roughly $118.8k a year against a $250k deal, and their repricing right matures before the customer's conversion decision. The armor is structural: owned schema and audit trail, a hard spend ceiling with an automatic halt, a second engine kept warm from the bake-off, and export rights demanded up front. |
| C · the trap, defused | The founder's own gut flagged it and the trial confirmed it: calendar-triggered rebuilds never happen for three-person teams running a live customer. So the rebuild trigger is cash, not calendar: signed conversion, money collected, thirty days of production data, then a named owner, a budget, a parity bar, and a shutdown date for the vendor. |
The final review kept the call and paid for its honesty: the vendor price lock became a long shot, and the integration's true staffing cost was corrected upward.
Vindicated Your instinct that the rebuild-later promise is a trap held through every attack pass: even the court's plan concedes a cash trigger can dissolve under customer pressure, which is why the decision goes on the calendar with investor visibility.
Upheld Your read that the deal is soft, an option rather than revenue, became the spine of the whole verdict: it is exactly why the cheap-to-reverse path wins and why the brave build would have bet the only free quarter on a verbal promise.
Every line above carries a receipt in the run record. This court affirms claims, not egos; when it says you were right, it is because the claim survived the trial.
The calendar math is brutal and undefended, the split seat got a real hearing and a real overruling, and the two honest walk-backs, the price lock demoted to a long shot and the staffing cost corrected upward, lowered the number without moving the answer. Everything now waits on one physical fact: whether an engine passes accuracy on fifty of the customer's actual contracts.
Strength is set by the weakest surviving assumption, not the average. Easy calls do not come to court, so verdicts here live mostly in the 40s to 60s, and movement means more than level. The raw number for this run stays on the record: stable 2/2 · 70%. The court's own final review cut this number before it ruled, movement in the uncomfortable direction, with the reasons on the docket in the web record.
The calendar kill of the in-house build: thirteen unreviewed weeks against five remaining, undefended by anyone in the record. The founder's own read that the deal is soft: an option, not revenue, which is precisely why the cheap-to-reverse path wins. The C-trap diagnosis: calendar-triggered rebuilds never happen for three-person teams, so the rebuild trigger is cash collected, never a quarter on a slide.
The dissenting seat holds that renting the core kills the company's story: investors fund proprietary pipelines, not resold infrastructure, and every week on the vendor deepens the dependence; it wins retroactively if the rebuild trigger dissolves under customer pressure after conversion, which is why the rebuild decision carries a named owner, investor visibility, and a date it cannot quietly slip past.
The full trial, with every attack and its answer, is below in the record.
Several independent models argued this, each assigned a different way of thinking, their arguments stripped of names before a judge model ruled. The room genuinely split two against one, the dissent arguing that renting your core kills the story, and the chair overruled it on calendar arithmetic while folding its warning into the winning architecture. Every load-bearing number was checked live, one external pricing claim was refuted and struck, two attacker hits forced honest walk-backs, and the final deliberation ran twice from scratch, agreeing both times.
The accordions above are the curated story. This is the full docket: every hit both adversaries landed, what the court did about each, and how it was disposed. Nothing is cherry-picked; where a hit was conceded, the tag says conceded.
| # | The attack | The court's answer | Disposition |
|---|---|---|---|
| Wave 1 · every claim checked live An independent checker re-verified the record before any attack: four claims confirmed, seven held as unverified, the company's ten declared facts taken as its own, and one seat's pricing benchmark refuted against the live page and struck. | |||
| W1·1 | A seat's competitor-pricing benchmark understated the real published rate by more than half. | Refuted against the live pricing page and struck. It was never load-bearing: the vendor arithmetic runs on the company's own declared per-document rate, which is why the strike cost the citation and not the call. | Refuted · struck |
| W1·2 | Supporting citations on lock-in standards and developer-interruption research could not be retrieved. | Held as unverified and barred from weight. The lock-in argument stands on structure, schemas and tuning grow vendor-shaped, which needs no citation to be true. | Downgraded · quarantined |
| Wave 2 · the attack on the verdict An adversary model attacked the winning plan's optimism: the price lock, the staffing estimate, the repricing squeeze. Two walk-backs were conceded on the record and the armor got thicker. | |||
| W2·1 | The price-lock request is a plan built on a pilot customer's nonexistent bargaining power. | Conceded: demoted to a long shot worth asking for. The real protection became structural, the spend ceiling, the warm second engine, the owned interface, and a switch-before-integration rule if terms are refused. | Conceded · restructured |
| W2·2 | The integration is costed as one engineer part-time, which no bake-off, labeling, hardening, and pilot-scrutiny reality supports. | Conceded and corrected: most of both engineers for the three-week window. Survivable at three weeks, which is the point; fatal at thirteen, which is the verdict. | Conceded · repriced |
| W2·3 | The vendor's repricing right matures before the customer's conversion decision: the squeeze arrives exactly when walking away is most expensive. | Landed and armored: hard monthly spend ceiling with automatic halt, raw outputs stored on the company's side, a runner-up engine kept warm, and export rights demanded in writing on Day 0. | Landed · armored |
| Wave 3 · the counter-attack on the alternatives A second adversarial pass gave the dissent its full day in court and tested the rebuild-later compromise. The dissent was overruled on arithmetic; the compromise was defused with a cash trigger. | |||
| W3·1 | The dissent deserves to win: renting your core guts the investment story, so build anyway and make the customer wait. | Overruled on arithmetic nobody could soften: thirteen unreviewed weeks against five real ones, a quarter of runway spent on a verbal promise, and a story that reads worse with a dead pilot in it. The warning survives inside the armor. | Overruled · absorbed |
| W3·2 | The compromise everyone reaches for, vendor now and rebuild in Q2, was waved through too easily. | The court agreed it is a trap and defused it instead of banning it: the rebuild triggers on signed conversion plus collected cash plus thirty days of production data, with a named owner and a date, because promises scheduled by calendars dissolve and promises funded by revenue mostly do not. | Landed · trap defused |
A genuine two-against-one, ruled on the merits: the dissenting seat's story about owning your core was heard, priced against the calendar, and overruled. Its best warning was built into the winner's armor.
| When | Gate |
|---|---|
| Today, in writing | Two asks go out Day 0: can the customer hand over about fifty real contracts this week, and will the champion accept staged delivery, answered in writing. Both are untested assumptions right now, and one of them decides everything. |
| Days 1–3 | The bake-off: vendor plus two alternates against the fifty real documents. Nothing clears the accuracy bar, the vendor path dies and the deadline gets renegotiated immediately, honestly, instead of pretended at. |
| Day 21, then Day 33 | Live by Day 21 keeps twelve days of buffer. Day 33 is the hard line: live, or a written extension, or the kill. No open-ended rescues. |
| Around Day 123–132 | The cash line: signed conversion and money collected, or shut down the uncapped processing. Only after cash plus thirty days of production data does the rebuild get authorized, with a named owner, a budget, a parity bar, and a shutdown date for the vendor. |
estimated build weeks against weeks actually remaining, from an unreviewed estimate, by three people who also have to sell and support the pilot
per year at the customer's projected volume, against a $250k deal, with a repricing right that matures before the conversion decision: the squeeze the armor exists for
of the customer's real contracts for the bake-off: until an engine passes accuracy on their documents, every plan in this file is theory
The lock-in that matters is never the API call: it is the schemas, the tuning, the corrections, and the audit trail that quietly grow vendor-shaped. So the court's architecture rule is absolute: the company owns its schema, its validation logic, its corrections and audit trail, and stores every raw output on its own side. Product code never touches vendor-native objects. A hard monthly spend ceiling halts processing automatically, a second engine from the bake-off stays warm, and if the vendor refuses export rights and clean data terms, the switch happens before integration, not after.
Pre-registered: live by Day 21 with buffer, and Day 33 is live-or-written-extension-or-kill. Around Day 123 to 132, the cash line: signed conversion and collected money, or the uncapped processing shuts down. The rebuild is authorized only after cash plus thirty days of production data, with a named owner, a parity bar, and a shutdown date for the vendor. No open-ended rescues on any branch.
Every council verdict makes dated, checkable predictions. This one grades itself in stages: Day 33 for live-or-kill, the cash line around Day 130 for whether the deal was ever real. The scorecard fills in either way.