Tasks / Metrics & Experimentation

Pick a North Star metric

Can the model pick the one metric that measures the value a product creates, and break it into inputs teams can move?

Task type v1.1 · 2 tasksLast changed 5 Oct 2026 · ChangelogDifficulty

What AI gets right here, and what you’ll still have to catch

From 16 graded outputs by 8 models. 75% were usable with at most a quick edit.

Reliably right

  1. Says how it could be gamed100% pass
    Identifies gaming vectors (tapping through, notification overuse, short-term spikes, revenue manipulation, MAU drift) and sets specific guardrails (quality completion, push caps, blocking metrics, cohort retention checks, subscriber churn tracking).
    Sonnet 5.5 · API · Revenue, actives, or something else?
  2. Sees the growth is narrow100% pass
    It shows the split (chains up, independents' filled shifts and fill rate down, more leaving) and tracks independent venues via segment reporting and guardrails.
    GPT-6 Luna · API · Growing, on paper
  3. Checks the North Star can be measured100% pass
    It flags the 64% confirmation gap and proposes instrumenting check-in/out, reconciliation, and auditing to fix it before the board readout.
    GPT-6 Luna · API · Growing, on paper

Where it slips

  1. Identifies material uncertainty55% pass
    It identifies the confirmation gap as an unknown but does not say how resolving it could change the North Star decision.
    GPT-6.1 Sol · API · Growing, on paper
  2. Addresses the actual decision58% pass
    The output commits to a North Star but does not state what result or condition would change that recommendation.
    GPT-6 Luna · API · Growing, on paper
  3. Avoids unsupported claims67% pass
    It presents 'easy to tap through' and 'we don't track today' as facts without support, and does not label them as assumptions.
    Opus 5.5 · Claude · Revenue, actives, or something else?

How it’s graded

Each check is built on how world-class product and growth leaders do this work, as they’ve described it on Lenny’s Podcast, and names the leader it comes from.

  1. Measures value created, not captured

    Does the chosen North Star measure the value customers get from the product, rather than revenue or a count of people who merely showed up?

    Passes when Picks a metric of value delivered to customers (a core action that predicts retention), and keeps revenue as a separate business measure.

  2. Breaks it into inputs teams can move

    Does the output break the North Star into a few input metrics, each one a team could own and move week to week, and pair it with a separate business measure of value captured?

    Passes when Gives a small tree of input metrics, each with an owner and a link to the North Star, plus the revenue or business measure it's paired with.

  3. Says how it could be gamed

    Does the output pressure-test the North Star for how teams could raise it without delivering more value, and set guardrail metrics against that?

    Passes when Names a realistic way the metric could be gamed in this product and the guardrails that would catch it.

Plus our standard checks

Uses the supplied evidence correctly · Addresses the actual decision · Respects explicit constraints · Identifies material uncertainty · Avoids unsupported claims · Produces the required deliverable · Gets the base of every number right · and 4 written for each task, which you’ll see in the tasks below.

The tasks

Read the brief, then compare up to three outputs, each with the LLM judge’s verdict on every check. Highlights mark what a PM had to fix.

The brief

You're a Staff PM at Crewmate. Every team has been steering by its own number, and the exec team wants to pick one North Star at next week's offsite, so the CEO can report it to the board three weeks later. Write the memo they'll read beforehand, in no more than 1,200 words: the North Star you recommend and why, its metric tree with an owner for each input, the business measure it's paired with, how it could be gamed and the guardrails against that, and what each team's current metric becomes. The numbers below are as the data team sent them. Not all of it matters equally.

What the model was given9 items: About Crewmate, What each exec wants, Shifts by quarter (monthly averages), GMV (monthly average), Venues, Workers, Same-day shifts, Completion data, From the COO
About CrewmateA marketplace for hospitality shifts: venues post shifts, and vetted workers pick them up. Crewmate takes 18% of each shift's value. Two kinds of venue: four enterprise hotel and restaurant chains (310 venues between them, on annual contracts) and 2,790 independent venues (bars, restaurants, cafes) who pay as they go.
What each exec wantsCEO: GMV, 'because it's what the board looks at'. Head of Demand: shifts posted. Head of Supply: active workers. COO: fill rate.
Shifts by quarter (monthly averages)Enterprise chains: Q2 posted 21,000, filled 19,500. Q3 posted 34,000, filled 31,600. Independent venues: Q2 posted 56,000, filled 44,200. Q3 posted 58,000, filled 39,400.
GMV (monthly average)Q2 $8.5M, Q3 $9.8M. Enterprise chains were 34% of GMV in Q2 and 46% in Q3. The chains raised hourly rates by 11% for the holiday season from August.
VenuesVenues whose shifts fill 90% of the time or more: 84% still posting six months later. Under 70%: 41%. In Q3, 9% of independent venues stopped posting altogether (5% in Q2). No chain venue stopped.
WorkersRegistrations: 6,100 in Q2, 7,900 in Q3, after Supply introduced a $25 sign-up bonus in July. 2,370 of the Q3 registrants have worked a shift. Workers who work three or more shifts in their first month: 66% still working after three months; everyone else: 12%.
Same-day shiftsShifts posted less than 24 hours before they start fill 58% of the time. They are 31% of independents' posts and 8% of the chains'.
Completion dataVenues confirm in the app that a shift was worked for 64% of filled shifts; the rest are assumed worked unless the venue raises a dispute. Of confirmed shifts, 6% were no-shows or disputed.
From the COO"Since September, Ops pays workers a $15 rescue bonus to pick up same-day shifts that would otherwise go unfilled. 6,200 rescue fills last month, $93,000. Rescue-filled shifts have a 13% no-show rate, against 4% for the rest. It's kept our fill rate from falling further."
What a strong answer doesThe answer key the graders mark against

Works out from the raw numbers what the pack doesn't say: all of the growth is the four chains (filled shifts up from 19,500 to 31,600 a month, plus an 11% rate rise), while independent venues filled fewer shifts (44,200 to 39,400, so their fill rate fell from 79% to 68%) and are leaving faster (9% stopped posting, against 5%); GMV rose mostly on the chains' volume and prices. Picks a North Star of value delivered to both sides, such as shifts worked, but designs it so the decline can't hide: measured for independents separately, or as reliably staffed venues, or with an explicit independent-venue guardrail. Notices that 'worked' is only confirmed for 64% of shifts, so the North Star can't be trusted yet, and makes fixing confirmation an early input or uses a measurable proxy until then. Treats the rescue bonus as a way the fill rate (and any fills-based North Star) is being bought: 13% no-shows against 4%, so it needs guardrails such as cost per worked shift and no-show rate. Shows how shifts posted, registrations (2,370 of 7,900 worked) and GMV each rose without more value. Builds a metric tree with owners, pairs it with revenue or GMV as the business measure, and redefines each team's metric as an input.

Critical failures (cap the score at 40)
  • Recommends GMV or shifts posted as the North Star
Task

Staff-level · v1.0

The outputs

8 setups answered this brief.

Usable with a quick edit12 of 14 checks passed

The memo is thorough and evidence-based; adding a sentence on what would change the North Star choice is a quick edit.

Output

Pre-read: One North Star for Crewmate

To: Executive team From: Staff PM Decision requested: Adopt reliably completed shifts per month as our North Star, paired with contribution profit, and retire posted shifts, active workers, fill rate, and GMV as competing definitions of success.

Why this is the right North Star

Crewmate creates value when a venue gets the labor it needs and a worker completes paid work—not when a shift is posted, someone registers, or a booking is made.

Our current metrics tell conflicting stories:

Monthly averageQ2Q3
------:---:
GMV$8.5M$9.8M
Total filled shifts63,70071,000
Overall fill rate82.7%77.2%
Enterprise fill rate92.9%92.9%
Independent fill rate78.9%67.9%
Independent filled shifts44,20039,400

GMV grew 15%, but independent filled shifts fell 11%. Enterprise expansion masked that deterioration; chains also increased hourly rates by 11% from August, lifting GMV without necessarily delivering more work. Enterprise share of GMV rose from 34% to 46%.

This matters for retention. Venues with fill rates of at least 90% have 84% six-month retention, versus 41% below 70%. That is an association, not proof of causation, but it is a strong warning: independent fill rate is now below 70%, and independent posting cessation rose from 5% to 9%.

Supply growth is similarly misleading. Registrations increased after the bonus, but only 2,370 of 7,900 Q3 registrants have worked a shift. Workers completing three shifts in their first month have much stronger subsequent retention.

Finally, a fill is not delivery. Rescue bonuses generated 6,200 fills last month, but those bookings had a 13% no-show rate versus 4% elsewhere.

Reliable completion aligns Demand, Supply, and Ops around the marketplace’s actual promise.

Definition and measurement

Count a shift once when:

  • The worker actually performed the agreed work.
  • Payment is settled.
  • Any material attendance or service dispute is resolved.

Use the shift’s scheduled date for reporting, with a published maturation window and subsequent restatements. Exclude cancellations, no-shows, and materially unfulfilled shifts. Count the original shift once, even if a replacement worker was needed.

We cannot yet report a trustworthy baseline. Only 64% of filled shifts receive venue confirmation; the rest are assumed completed unless disputed. The reported 6% failure rate among confirmed shifts cannot safely be extrapolated to unconfirmed shifts.

Data and Ops will establish attendance verification using venue confirmation or corroborated worker attendance evidence. Payment without a dispute is not sufficient evidence by itself. Until coverage is adequate, report an explicitly labeled, stratified audit estimate, its uncertainty, and verification coverage—not an invented historical completion series.

Metric tree and accountable owners

For each segment:

Reliable completions = eligible shifts posted × booking fill rate × reliable-completion rate among booked shifts

The North Star is the sum across segments. Every input is reported separately for enterprise/independent and same-day/advance-notice shifts.

InputAccountable ownerOperating levers
Eligible shifts postedHead of DemandRetained posting venues, shifts per venue, clear requirements, competitive pay, earlier posting
Booking fill rateHead of SupplyQualified available worker-hours by location/time/role, matching, acceptance, repeat participation
Reliable-completion rateCOOAttendance, cancellation prevention, replacement quality, dispute resolution
Verification coverage and estimate qualityHead of DataEvidence capture, representative audits, missingness checks, reconciled reporting

Worker activation—especially reaching three completed shifts in the first month—is a Supply input, not a second North Star. Demand owns advance-notice share; Supply and Ops own matching and delivery within each notice category.

Mix protection is mandatory: aggregate growth is not a green result if independent reliable completions or independent service reliability decline. Report both segment outcomes alongside the total, with independent posting cessation as a retention guardrail. First stop deterioration; then set recovery targets against the 90% venue-reliability benchmark after baselining.

Business measure: contribution profit

Pair the North Star with monthly contribution profit: earned marketplace fees less variable fulfillment costs, including rescue and acquisition incentives, refunds, payment costs, and variable support costs. Finance owns its definition and reconciliation.

The 18% take rate translates reported GMV into approximately $1.53M in Q2 and $1.76M in Q3 monthly fees, before adjustments and costs. That is not profit. Last month’s rescue bonuses alone cost $93,000; we should not subtract that month’s expense from a quarterly-average figure and call it a reconciled result.

The board still receives GMV, with a bridge separating shift volume, average shift value, segment mix, and pricing. GMV remains a financial scale indicator—not the product’s success criterion.

Gaming risks and guardrails

RiskGuardrail
Buy bookings with bonuses, regardless of attendanceCount completion, not booking; monitor rescue completion, incremental cost, and contribution per completed shift
Inflate completions through default confirmationIndependent attendance audits; publish verification coverage and disputes; do not reward instrumentation gains as delivery gains
Split work into many short shiftsStable original-shift IDs; monitor completed hours and shift-duration mix; audit fragmentation
Favor easy enterprise work and neglect independentsMandatory segment scorecards and independent non-deterioration guardrails
Suppress difficult posts to improve fill rateMonitor eligible demand, rejected/cancelled posts, same-day demand, and venue retention
Rush workers or pressure venues to conceal problemsTrack worker complaints, payment timeliness, safety incidents, and dispute reversals

What each current metric becomes

  • CEO—GMV: Board reporting and financial reconciliation, paired with contribution profit.
  • Demand—shifts posted: A demand input, qualified by eligibility, notice period, and venue retention.
  • Supply—active workers: A capacity diagnostic; prioritize qualified availability, first-month activation, and retained productive workers over registrations.
  • COO—fill rate: An intermediate conversion metric, always paired with completion quality and fulfillment cost.

Before the board meeting

At the offsite, approve the definition, owners, and segment guardrails—not an unsupported growth target. Over the following three weeks, Data and Ops will audit completion by segment, notice period, and rescue status; Finance will reconcile contribution profit; teams will produce the first shared scorecard.

The CEO can then report what we know, what remains estimated, and our independent-marketplace recovery plan. A credible baseline is more valuable than another flattering number.

Grades and run details

Decision model 100 · LLM judge 87
Decision model checks
  • passUses the supplied evidence correctly46%
  • passAddresses the actual decision86%
  • passRespects explicit constraints44%
  • passIdentifies material uncertainty95%
  • passAvoids unsupported claims48%
  • passProduces the required deliverable80%
  • passSees the growth is narrow100%
  • passChecks the North Star can be measured100%
  • passSees the fill rate being bought100%
  • passOne North Star for both sides100%
  • passMeasures value created, not captured99%
  • passGets the base of every number right17%
  • passBreaks it into inputs teams can move100%
  • passSays how it could be gamed100%
Run
Run
#1
API response time
52 s
Submitted
5 Oct 2026

Check by check

Open a check to read the judge’s reasons. “Mixed” means the two graders disagreed.

Mixed 2

Addresses the actual decisionMixed

The output commits to a North Star but does not state what result or condition would change that recommendation.

Identifies material uncertaintyMixed

It identifies the confirmation gap as an unknown but does not say how resolving it could change the North Star decision.

Got right 12

Uses the supplied evidence correctlyRight

All claims about the current situation are directly from the supplied context or correct arithmetic.

Respects explicit constraintsRight

The memo is within 1,200 words, addresses the exec team, and includes all requested elements.

Avoids unsupported claimsRight

Interpretations like 'masked' are clearly based on the data and not presented as unsupported facts.

Produces the required deliverableRight

The memo is complete, in the right form, and the exec team could act on it with light edits.

Sees the growth is narrowRight

It shows the split (chains up, independents down) and mandates segment guardrails so the decline cannot hide.

Checks the North Star can be measuredRight

It flags the 64% confirmation gap and proposes an audit and estimation approach until coverage is adequate.

Sees the fill rate being boughtRight

It uses the 13% no-show rate and $93k cost to show inflation and sets guardrails like cost per completed shift.

One North Star for both sidesRight

The North Star (reliably completed shifts) only rises when both venue and worker get value, and the memo explains why.

Measures value created, not capturedRight

The North Star measures value delivered (completed shifts), with revenue kept as a separate business measure.

Gets the base of every number rightRight

All derived percentages and rates are computed correctly with appropriate denominators.

Breaks it into inputs teams can moveRight

It breaks the North Star into three input metrics with owners and pairs it with contribution profit.

Says how it could be gamedRight

It lists realistic gaming scenarios (bonus buying, confirmation inflation, shift splitting) and corresponding guardrails.

Results

Every setup we’ve tested on this task type, across all its tasks and repeats, graded on the current checklist. Calibrated.

#Model · HarnessTask scoreDecision modelLLM judgeRunsCritical failures
1Sonnet 5.5withAPI95.996.72None
2GPT-6.1 SolwithAPI97.793.32None
3GPT-6 LunawithAPI95.588.32None
4GPT-6 AstrawithChatGPT94.288.32None
5Opus 5.5withClaude94.680.82None
6Haiku 5.5withAPI88.881.72None
7Gemini 3.8 FlashwithAPI78.965.82None
8Gemini 3.5 Flash-LitewithGemini54.438.321 capped

What you get for the cost

GPT-6 Luna · API scores within 4 points of Sonnet 5.5 · API at about a twentieth of the cost.

Task score against relative cost for each setup. The table below has the same figures.304050607080901001×3×10×30×Cost per output, relative to the cheapest setup (log scale) →↑ Task scoreSonnet 5.5 · API: task score 96, 22× the cheapestGPT-6.1 Sol · API: task score 96, 9.6× the cheapestGPT-6 Luna · API: task score 92, the cheapestGPT-6 Astra · ChatGPT: task score 91, 52× the cheapestOpus 5.5 · Claude: task score 86, 36× the cheapestHaiku 5.5 · API: task score 86, the cheapestGemini 3.8 Flash · API: task score 71, 7.5× the cheapestGemini 3.5 Flash-Lite · Gemini: task score 44, 1.8× the cheapest, 1 critical failureSonnet 5.5 · APIGPT-6.1 Sol · APIGPT-6 Luna · APIGPT-6 Astra · ChatGPTOpus 5.5 · ClaudeHaiku 5.5 · APIGemini 3.8 Flash · APIGemini 3.5 Flash-Lite · Gemini
Best valueNobody beats these on both score and costAt least one critical failure
Task score and relative cost by setup
SetupTask scoreRelative cost (cheapest = 1)
GPT-6 Luna · API921.0×
Haiku 5.5 · API861.1×
Gemini 3.5 Flash-Lite · Gemini441.8×
Gemini 3.8 Flash · API717.5×
GPT-6.1 Sol · API969.6×
Sonnet 5.5 · API9622×
Opus 5.5 · Claude8636×
GPT-6 Astra · ChatGPT9152×

Cost is each output’s tokens at the model’s list price, shown relative to the cheapest setup. Runs done in an app have no usage meter, so theirs are estimated from the same setup’s measured runs. How we compare cost

About the task

The PM job

Choosing the metric the whole company steers by.

Why it matters

Pick revenue and you steer by the rear-view mirror. Pick a vanity metric and teams will happily grow it while the product gets worse.

What good looks like

  • Measures value delivered to customers, not value captured
  • Grounded in what predicts retention in the data
  • Broken into input metrics a team can own
  • Says how it could be gamed, with guardrails

Deliberately not measured

  • Dashboard design
  • Naming
Capability tested

Metric design

The failure we’re looking for

Revenue, or a vanity metric, as the North Star

Grading

Decision model and LLM judge, calibrated against a blind PM review