The decision ledger
Decisions with a fork in them, and what the council said
Every entry below is one real session. Five models write independently, review each other's work without knowing whose it is, and a chairman writes the synthesis. Nothing here is edited into agreement.
We keep the dissent because that is the part you cannot get anywhere else. A verdict tells you what to do. The disagreement tells you when to stop trusting it.
Every factual claim is checked against primary sources after the session. What failed the check is printed on the page, not quietly removed.
- Invalidated
Productize or hire your first subcontractor
Productization is a demand problem wearing a delivery problem's clothes — and this pipeline can't feed it.
A solo B2B consultant at full billable capacity should productize into a fixed-scope offer rather than hire their first subcontractor.
- Invalidated
Stay flat-subscription or go usage-based
The council rejected the question — the cost ratio isn't the decisive variable, and the observed cost rise may not be real.
A B2B AI product whose inference cost runs a third of revenue should move to usage-based pricing rather than defend a flat subscription.
- Invalidated
Turn down your biggest client's expansion offer
Take the money — but the decision turns on how confident you are in your own business development, not on the concentration number.
An independent consultant whose largest client is already 45% of revenue should turn down that client's offer to expand the engagement, and spend the capacity on diversifying instead.
- Validated
Decline your first six-figure enterprise deal
Decline — but the real variable is when the cash lands, not whether the deal is good.
A two-person product team should decline its first six-figure enterprise deal when that deal requires SOC 2 and a single-tenant deployment.
- Invalidated
Narrow to one vertical, or stay general
Narrow the offer, not the industry — and the runway can't absorb declining 82% of your referrals on evidence this thin.
A generalist consultant should narrow to a single industry vertical even though doing so shrinks their addressable pipeline in the short term.
- Validated
Commit to one model, or abstract behind a router
Commit — and settle it with your own eval set rather than the argument, at a 2-point accuracy delta.
A small AI product team should commit to a single frontier model and build deeply against it, rather than abstract behind a multi-model router.
What a ledger entry is not
It is not advice about your situation. Each entry runs against a composite case file with stated numbers, and the verdict is only as good as those numbers.
We do not claim five models beat one. We ran the comparison — six pre-registered decisions, each judged blind against a single strong model — and the council did not win. We are publishing that because a claim we only tested when we expected to like the answer would not be worth much.
The council's citations are the weakest part of its output. We check them and publish the result either way — the full fact-check ledger is here. Several numbers in these entries did not survive.