The question put to the court: Laid off at 38 with two kids and three doors: the safe $140k corporate offer, the $115k startup offer with equity, or $9k and six months of savings to retrain for the AI-product roles that keep rejecting her?
Eleven years at one company ended in a reorg. Two AI-product interviews already said no, and both said why. Severance is ticking, two offers carry real deadlines, and the household runs on one and a fraction incomes. The court's job: pick one door and name honestly which ones close.
CASE FILE № 1 · CONVENED JULY 15, 2026
The question put to the court: Laid off at 38 with two kids and three doors: the safe $140k corporate offer, the $115k startup offer with equity, or $9k and six months of savings to retrain for the AI-product roles that keep rejecting her?
Eleven years at one company ended in a reorg. Two AI-product interviews already said no, and both said why. Severance is ticking, two offers carry real deadlines, and the household runs on one and a fraction incomes. The court's job: pick one door and name honestly which ones close.
Take the $140k corporate offer, and sign it by August 5. Then, on your own time and your own gear, build one small public AI artifact aimed at the exact screens that rejected you, with a shipping check at week six so it cannot silently rot. Before you sign anything, read the invention-assignment and moonlighting clauses: one hour of reading protects the whole plan. The startup and the full-time retrain both die as offered, and each gets a door back with conditions attached.
| Option | Why it lost or won |
|---|---|
| A · wins, sharpened | The only door that pays her to fix the thing actually blocking her. It never touches the family's savings, it ends the résumé-gap clock, and it funds the AI artifact the rejections asked for. The trial sharpened it rather than flattering it: the side-build got smaller, gained a week-six shipping check, and the offer's IP clauses became a pre-signature reading assignment. |
| B · killed, door priced | Dies on arithmetic, not vibes: a $25k pay cut into a company with founder-claimed runway shorter than typical seed-to-A timelines, equity that likely dilutes before it means anything, and a failure mode where her income and the household's plan die together. The least-contested kill in the record. The door back has four locks and a July 29 deadline. |
| C · killed on evidence | Buys the wrong thing. Both rejections named missing shipped AI work, not a missing certificate, and the market data points the same direction. It burns the buffer to purchase a credential nobody asked for, during the exact window when a résumé gap costs the most. Four separate lines of reasoning arrived at the same kill, including the court's own second attacker. |
The final review cut the court's confidence where the attack landed: the hours were optimistic, the runway was overstated, and a five-year earnings comparison got demoted to illustration.
Vindicated Your diagnosis was exactly right: the AI-product screens rejected you for having no shipped AI work, not for lacking a certificate. Both rejections said so, the market data agrees, and every plan in this verdict is built on your own read.
Recalibrated Your feeling that you get one more big swing was met with better math: after honest health-coverage accounting you have more swings than you feared, and fewer than you hoped. The winning door spends none of them.
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 kills are unanimous and the signature is safe: even if everything ambitious in this plan fails, she is a salaried product manager with intact savings. The one unproven link is whether a shipped artifact converts rejections into interviews, and the plan tests that for free this week before a single line of code gets written.
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 · 65%. 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 diagnosis: she is being screened out for missing shipped AI work, not a missing certificate. Both rejections said it, the data agrees, and it survived every pass untouched. The startup kill: pay cut, short claimed runway, diluting equity, and a household whose income and plan would fail together. The least-contested finding in the record. The floor logic: even total failure of the ambitious half leaves a salaried product manager with intact savings, which is why the signature is safe regardless.
No seat dissented on the signature; the surviving argument is about odds, not doors: the attacker holds that one more solo artifact converts nothing in a market flooded with them; the plan concedes the point in advance by testing conversion this week, free, in three hiring-manager conversations before any code gets written.
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. All arrived at the same signature by different roads, so the trial attacked the roads: every load-bearing claim was checked live, an adversary model went after the winning plan's assumptions, a second one broke the boldest alternative, and the final deliberation ran twice from scratch, agreeing both times while cutting its own confidence where the attacks landed.
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 against live sources before any attack: sixteen claims confirmed, four in part, eleven held as unverified, and zero fabrications found. | |||
| W1·1 | The record leaned on salary bands and market claims that could have been folklore. | The checker confirmed the load-bearing bands against live sources: the target roles really do pay what the plan assumes. Her metro-specific number stays flagged as her own belief. | Confirmed · held |
| W1·2 | One seat's supporting statistics, wage premiums, posting percentages, and startup mortality rates, could not be matched to sources. | Downgraded to direction-only and barred from carrying weight. Zero fabrications and zero refutations in this record; the weakest citations were caught and labeled before the ruling. | Downgraded · quarantined |
| Wave 2 · the attack on the verdict An adversary model attacked the winning plan's soft tissue: the hours, the runway, the conversion odds, the contract fine print. Its best hits were conceded and built into the plan. | |||
| W2·1 | Six hours a week of side-building during new-job onboarding, with a four- and seven-year-old, is a fantasy for the first three months. | Partially conceded and built into the plan: the artifact shrank, a week-six public-shipping check was added so the build pauses honestly instead of rotting silently, and the November 30 checkpoint stands. | Conceded · re-scoped |
| W2·2 | Solo demo projects are commodity now; every pivoting product manager has one, so the artifact may convert nothing. | Landed as the verdict's honest weak link, stated in the confidence itself. The counter is cheap and scheduled first: three hiring-manager conversations this week asking whether the artifact would clear their screen, before any code exists. | Landed · tested first |
| W2·3 | The eighteen-month runway ignores what family health coverage costs when nobody is employed. | Conceded: effective runway restated to thirteen-to-fifteen months. The correction strengthened the winning door, which is the only one that spends none of it. | Conceded · restated |
| W2·4 | If the employer's contract claims her side work, the entire pivot plan belongs to someone else. | Adopted as a pre-signature step: read the invention-assignment and moonlighting clauses before signing, and the build runs on her own time and gear regardless. | Adopted · gated |
| Wave 3 · the counter-attack on the alternative A second adversarial pass attacked the boldest alternative, the full-time retrain, and broke it on callback decay, buffer consumption, and asset mismatch. | |||
| W3·1 | The braver alternative was refusing both offers and retraining full-time; the court killed it too quickly. | The second attacker tested it and broke it three ways: callback rates decay fastest exactly in the gap it creates, the household's buffer becomes mandatory consumption on unproven odds, and it purchases a credential the screens never asked for. | Landed · retrain fell |
| W3·2 | Split effort fails both halves; a job plus a side-build is two half-commitments. | Answered with the plan's structure: one half is a paycheck that cannot fail her, and the other now has two expiry dates instead of none. Split effort fails when both halves are open-ended; neither is. | Rejected · answered |
All seats arrived at the same signature, each by different roads, so the court attacked the road instead: the hours, the runway, the conversion odds. What survived is a plan with a floor.
| When | Gate |
|---|---|
| This week | Before writing any code: three informational chats with AI-product hiring managers, asking straight whether the planned artifact would clear their screen. It tests the plan's weakest link for free. And read the offer's invention-assignment and moonlighting clauses before signing. |
| By July 29 | Let the startup offer expire unless all four locks open: bank-verified runway past 20 months, a named lead investor, three customer references, and contractual AI work shipping externally within about 120 days. Anything missing, no counter-offer conversation. |
| By August 5 | Sign the corporate offer. |
| Week 6, then November 30 | Something public shipped by week six of employment or the build pauses honestly. By November 30: a working artifact and at least five AI-product interviews in motion, or the pivot re-plans. Failing that test leaves a well-paid product manager with intact savings, which was the designed floor all along. |
the salary that never touches the family buffer, ends the résumé-gap clock, and funds the reinvention while someone else pays for the runway
what eighteen months becomes once family health coverage with nobody employed enters the math; several swings, not one, and fewer than it felt like
shipped AI artifacts on her résumé: the thing both rejections actually named, which no certificate fixes and a paycheck can fund
The court does not pretend the losing options were foolish. The startup's door reopens July 29 behind four locks that convert founder charisma into bank statements, named investors, paying customers, and a dated ship clock. The retrain's door reopens only as proof-first: final-round interviews earned while employed, never a resignation on faith. And the winning door carries its own honesty: if the new job turns out to have zero AI surface, it demotes to a bridge year, signed and useful, while the search continues from strength.
Pre-registered: something public shipped by week six of employment, or the side-build pauses. By November 30, a working artifact and at least five AI-product interviews in motion, or the pivot plan re-plans from scratch. Failing the test costs nothing that matters: the floor was the point, and the floor is a $140k job with the family's savings intact.
Every council verdict makes dated, checkable predictions. This one grades itself November 30: either the artifact shipped and interviews are in motion, or the pivot re-plans and says so. The scorecard fills in either way.