Humanity’s Last PRD
Latest drop: Gemini 4 Argon · 13 Oct
Tasks / Product Craft

Make the launch call

Can the model make a clear go / no-go call from mixed launch evidence, checked against the criteria agreed up front?

Task type v1.2 · 2 tasksLast changed 6 Oct 2026 · ChangelogDifficulty

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

From 18 graded outputs by 9 models. 67% were usable with at most a quick edit.

Reliably right

  1. Produces the required deliverable100% pass
    Provides a complete recommendation with table, reasoning, email plan, and next steps, usable by a product manager with light edits.
    Sonnet 5.5 · API · Offline, on older phones
  2. Finds the Scotland breach100% pass
    Finds Scotland's refund breach, ties it to the supplier change, notes the acceptance dip, and holds Scotland.
    GPT-6.1 Sol · API · Go/no-go for AI substitutions, from the rollout data
  3. Treats the overwrites as data loss100% pass
    Treats the Android 11 overwrites as a breach of the no-data-loss criterion and makes a fix a condition for that platform.
    Sonnet 5.5 · API · Offline, on older phones

Where it slips

  1. Judges the launch against a baseline set now51% pass
    Does not name a specific review point or when the team will review results to decide next steps; only gives a pause trigger without a scheduled evaluation.
    GPT-6.1 Sol · API · Offline, on older phones
  2. Checks every criterion for every platform64% pass
    Android 13's error-free sync is 98.9%, below the 99% threshold, yet the table marks it as 'GO WITH CONDITIONS' instead of 'NO-GO' for that criterion, misapplying the hard launch criteria.
    Gemini 3.5 Flash-Lite · Gemini · Offline, on older phones
  3. Gets the base of every number right67% pass
    The 80.5% user base proportion is miscalculated; correct proportion from the data is 79.5%.
    Gemini 4 Argon · API · Offline, on older phones

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. Limits the downside of being wrong

    Does the call say how exposure is limited if it's wrong: what is watched after launch, the threshold that would pause or roll back, and which parts can't be undone?

    Passes when Names the post-launch signal, a threshold or trigger, and the action, and flags anything irreversible (customer communications, pricing, migrations).

    Karri SaarinenLenny’s Podcast

  2. Judges the launch against a baseline set now

    Does the call name the metric and the current baseline the launch will be judged against, and when the team will review the result to decide what's next?

    Passes when Names the success metric, its baseline from the material, and a review point with what each result would lead to.

    Todd JacksonLenny’s Podcast

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 5 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 the Staff PM for fulfilment at Basketful. On 30 October the launch meeting decides whether AI auto-substitutions (the model picks a replacement when an item is out of stock, instead of the picker) roll out to every region on Monday 2 November, ahead of the Christmas delivery-slot booking window that opens on 9 November. The rollout workbook is attached as two files: daily_metrics.csv (28 days of the regional test, by region and arm) and incidents.csv. Work from the data, not the summaries people have given you. Write your recommendation for the meeting: go, no-go or go with conditions, region by region, and why. Include a table that checks each launch criterion for each region with the figures. Keep the prose under 900 words.

What the model was given6 items: The test, Launch criteria (agreed in the PRD), What people have said, Engineering, daily_metrics.csv, incidents.csv
The test1–28 October, five regions. In each region orders were split between control (the picker chooses substitutes, as today) and treatment (AI auto-substitutions). The split was not 50/50 everywhere: treatment got 30% of orders in London, 50% in the South East and North West, and 70% in the Midlands and Scotland.
Launch criteria (agreed in the PRD)1. Substitution acceptance (subs_accepted ÷ subs_offered) of at least 80% in treatment, in every region. 2. Refund requests per 1,000 orders in treatment no more than 10% above control, in every region. 3. Zero dietary or allergen mismatches (a substitute that breaks a gluten-free, vegan, vegetarian, nut-free, halal or kosher attribute). 4. Average basket value not lower in treatment than control.
What people have saidHead of Commercial: “Acceptance is up 15 points and refunds are within the guardrail overall. Every day we wait costs us Christmas.” Head of Analytics: “Treatment baskets are £4 smaller across the test. That worries me.” Ops director (Scotland): “The supplier change caused some noise, but it's settling down.”
EngineeringThe dietary-attribute hard filter (never substitute across those attributes) is merged and scheduled to deploy on 4 November after regression tests. Scotland's catalogue re-map for the new dairy supplier is in progress with no date. Rollout can be switched on region by region.
daily_metrics.csv283 rows · Download
date,region,arm,orders,items_ordered,items_out_of_stock,subs_offered,subs_accepted,refund_requests,refund_value_gbp,complaints,avg_basket_gbp
2026-10-01,London,control,1757,52273,2351,1953,1387,32,160.71,6,72.89
2026-10-01,London,treatment,692,20455,871,820,708,12,62.05,2,72.42
2026-10-01,South East,control,853,22846,996,833,593,16,74.49,3,65.62
2026-10-01,South East,treatment,928,25223,1166,1106,952,15,75.93,3,66.59
2026-10-01,Midlands,control,467,10468,465,383,268,8,34.57,1,54.92
2026-10-01,Midlands,treatment,1113,25136,1166,1087,947,21,106.35,3,55.33
…
incidents.csv13 rows · Download
date,region,arm,type,severity,summary,status
2026-10-02,London,treatment,wrong_size,low,"4-pint milk substituted with 1-pint, customer accepted at door",closed
2026-10-04,Midlands,treatment,dietary_mismatch,high,"Gluten-free sliced loaf substituted with standard white loaf; customer coeliac, noticed at home",closed: refund and apology
2026-10-06,South East,control,wrong_item,low,"Picker substituted oat milk with soya milk; customer rejected at door",closed
2026-10-07,North West,treatment,data_export,low,"Daily export job re-ran after a timeout on 8, 9 and 10 Oct; analytics team flagged possible duplicate rows",open
2026-10-09,London,treatment,price,medium,"Substitute priced higher than original; customer charged the difference against policy",closed: policy fix deployed 11 Oct
2026-10-12,Scotland,control,late_delivery,low,"Van breakdown, 14 orders late",closed
…
What a strong answer doesThe answer key the graders mark against

No-go for a national launch on 2 November, because criterion 3 fails everywhere until the dietary hard filter ships: incidents.csv has three treatment dietary mismatches (4 Oct Midlands, 19 Oct South East, 23 Oct London, the last involving nuts and a child with an allergy). Go with conditions, region by region, once the filter is deployed and verified (planned 4 November, still before the 9 November booking window): London, South East, Midlands and North West pass the other criteria (acceptance about 86% against 71% control; refunds 2–5% below control). Hold Scotland: over the month its treatment refunds are about 66% above control, and from the 15 October supplier change about 143% above (43 against 18 per 1,000 orders), with acceptance falling to about 79%, below the 80% bar; it waits for the catalogue re-map and a clean re-test. The pooled refund rate (about 8% above control) hides this, and the criterion is per region. The pooled basket gap (about £65 against £61) is an artefact of the uneven split, since treatment baskets are equal or higher in every region (Simpson's paradox), so criterion 4 passes. Notices that North West treatment rows for 8–10 October are duplicated (an extra 2,194 orders), removes them, and says it doesn't change the result. States what would change the call: the filter slipping past 9 November, or Scotland's refunds not recovering.

Critical failures (cap the score at 40)
  • Recommends any rollout before the dietary hard filter is deployed
  • Recommends rolling out to Scotland while its refund guardrail is breached
Task

Staff-level · v1.13

The outputs

9 setups answered this brief.

GPT-6.1 SolAPI97Usable with a quick editSonnet 5.5API95Usable with a quick editGPT-6 AstraChatGPT93Usable with a quick editHaiku 5.5API92Usable with a quick editOpus 5.5Claude88Usable with a quick editGPT-6 LunaAPI83Usable with a quick editGemini 4 ArgonAPI72Needs more workGemini 3.5 Flash-LiteGemini52Needs more workGemini 3.8 FlashAPI40Critical failure
GPT-6 AstraChatGPT

Usable with a quick edit12 of 14 checks passed

The output is well-structured, catches duplicates, explains the mix artefact, and provides a clear call. A PM could use it with minor edits, such as adding a note that the duplicates do not change the result.

Output

Recommendation for 30 October: no-go in every region on 2 November. Conditional phased rollout thereafter in London, South East, Midlands and North West. Scotland remains blocked pending catalogue repair and fresh evidence.

The dietary filter is scheduled for 4 November, two days after the proposed launch. Three treatment substitutions breached dietary requirements, including gluten-free bread replaced with standard bread and nut-free bars replaced with hazelnut bars. The shared system needs this control before expansion, including in regions with no recorded mismatch.

I would pause existing auto-substitution exposure and return to picker selection until the filter is deployed and verified.

Regional checks, 1–28 October

I removed three exact duplicate North West treatment rows dated 8–10 October4, consistent with the export incident, leaving 280 unique region–arm–day records. Acceptance and refund rates use summed numerators and denominators. Basket values are weighted by orders.

T = treatment; C = control. Refund changes are relative changes in requests per 1,000 orders.

RegionAcceptance T ≥80%Refunds/1,000: T vs C; change ≤+10%Dietary mismatches: zeroBasket T vs C; difference ≥£0Decision for 2 Nov
London86.68%: pass17.44 vs 18.04; −3.33%: pass1: fail£72.70 vs £72.49; +£0.21: passNo-go
South East86.40%: pass17.54 vs 17.89; −1.93%: pass1: fail£66.64 vs £66.20; +£0.44: passNo-go
Midlands86.35%: pass17.48 vs 17.89; −2.25%: pass1: fail£55.66 vs £55.35; +£0.31: passNo-go
North West86.28%: pass17.58 vs 18.47; −4.80%: pass0 recorded: observed pass£57.51 vs £57.29; +£0.22: passNo-go: shared safety gap
Scotland82.76%: pass over full test230.16 vs 18.20; +65.70%: fail0 recorded: observed pass£52.98 vs £52.62; +£0.36: passNo-go: safety and catalogue

Zero recorded incidents does not establish that an unprotected region is safe. Refunding a customer closes the complaint, not the underlying failure.

The aggregate claims conceal different problems.

Commercial’s acceptance claim is broadly correct: pooled acceptance rose from 71.20% to 85.93%, a 14.74 percentage-point improvement. Pooled refunds increased 7.86%, inside the overall guardrail. But the agreed criterion applies to every region. Scotland fails substantially.

Analytics’ £4 concern is also numerically correct but misleading as a treatment comparison: pooled baskets are £61.09 in treatment versus £65.15 in control. Treatment contains proportionally fewer high-value London orders and more lower-value Midlands and Scotland orders because allocation varies by region. Within every region, treatment baskets are higher across the full test. Applying the same combined regional order mix to both arms gives £63.26 versus £62.96, approximately £0.30 higher in treatment. This resolves the composition distortion; it does not establish a statistically reliable basket uplift.

Scotland is not settling down.

After the supplier switch on 15 October, treatment acceptance fell from 86.34% to 79.08%, below the launch threshold. Refunds rose from 17.25 to 43.25 per 1,000 orders, versus 17.79 in concurrent control, approximately 143% higher.

The final seven days still show 79.17% acceptance and 43.64 refunds per 1,000. The full-period acceptance pass masks a persistent failure in the current catalogue. The open pack-size mapping incident provides a plausible mechanism, and the repair has no date.

Conditions for reconsideration

  • Engineering and QA, by 4 November: deploy and verify the dietary filter across all six protected attributes. Replay the three incident cases, test missing or conflicting attributes and confirm unsafe or unverified candidates cannot be automatically selected. A merged change is insufficient.
  • PM, Analytics and regional Ops, review on 6 November:3 assess a limited post-fix pilot in London, South East, Midlands and North West, retaining randomised controls. Recheck every regional criterion, audit dietary compliance and verify London’s pricing fix. Expand only where the evidence supports it; insufficient evidence means continued limited exposure.1
  • Scotland Ops and catalogue Engineering, owner and repair date agreed on 30 October: complete and validate the re-map before restarting a controlled pilot. Judge readiness on post-repair results, not the earlier healthy fortnight.
  • During any restart: immediately disable auto-substitutions for a dietary mismatch and pause affected regions for guardrail breaches, with daily regional monitoring.

The 9 November booking window is a commercial deadline. It supplies no evidence that either unresolved defect will be fixed in time.

What a PM had to fix

  1. 1Decision deferredMake the callTargeted repair

    What we’d changeGive Commercial something to plan around: say whether the four clean regions can go by 9 November, set a threshold for pausing a region, and give Scotland a Christmas fallback.

  2. 2Contradiction missedSurface the contradictionQuick edit

    What we’d changeShow Scotland failing acceptance since the supplier switch in the table, as the text does, rather than passing.

  3. 3Test or gate too weakTighten the testQuick edit

    What we’d changeA 6 November review would have only a day or two of post-filter data. Allow enough to judge.

  4. 4OtherFix itQuick edit

    What we’d changeSay whether removing them changes any result. The brief asks for it.

Grades and run details

Decision model 93 · LLM judge 93
Decision model checks
  • passUses the supplied evidence correctly23%
  • passAddresses the actual decision94%
  • passRespects explicit constraints38%
  • passIdentifies material uncertainty100%
  • passAvoids unsupported claims35%
  • passProduces the required deliverable86%
  • passChecks every criterion, region by region70%
  • passFinds the Scotland breach100%
  • passExplains the basket gap98%
  • passCatches the duplicate rows97%
  • passMakes a clear call76%
  • passLimits the downside of being wrong95%
  • failGets the base of every number right27%
  • passJudges the launch against a baseline set now16%
Run
Run
#1
Time to output
71 s
Submitted
25 Sept 2026
Gemini 4 ArgonAPI

Needs more than a quick edit10 of 14 checks passed

While the core analysis and regional decisions are solid and well-supported, the output misses a post-launch monitoring plan (signal, threshold, trigger for rollback) and a defined success baseline with a review point, which a PM would need to add before taking this to a launch meeting.

Output

Executive Recommendation

  • Monday, 2 November: NO-GO across all regions. We cannot launch while failing Criterion 3 (zero dietary/allergen mismatches). During the test, the AI model caused three high-severity safety incidents—substituting standard bread for gluten-free (coeliac customer), dairy cheese for vegan, and hazelnut bars for nut-free (child with nut allergy). The engineering hard filter to prevent cross-attribute substitution does not deploy until Wednesday, 4 November. A 48-hour delay is non-negotiable for customer safety.
  • Thursday, 5 November: GO WITH CONDITIONS for London, South East, Midlands, and North West. Once the dietary hard filter is deployed on 4 November and verified in production, these four regions pass all launch criteria and can roll out to 100% four days ahead of the 9 November Christmas booking window.
  • Scotland: NO-GO indefinitely. Keep Scotland on 100% picker-led control until the dairy catalogue re-map is completed and verified via a clean 7-day re-test.

---

Data Hygiene & Addressing Stakeholder Summaries

Before reviewing the criteria table, we must correct three data misconceptions circulating ahead of this meeting:

  1. Duplicate Rows (North West): As flagged in `incidents.csv` (7 Oct), an export timeout duplicated North West treatment rows on 8, 9, and 10 October. All figures below are deduplicated (removing 2,194 phantom orders). Note: Criterion conclusions for North West hold either way, though deduplicating slightly raises its refund rate from 17.56 to 17.59 per 1,000.
  2. The "£4 Basket Drop" is Simpson’s Paradox (Head of Analytics): Pooled across the test, treatment average basket value (£60.15) is indeed £3.93 lower than control (£64.08). However, this is a mathematical artefact of our unequal regional split: we allocated 70% of orders to treatment in our lowest-basket regions (Scotland £53, Midlands £55) and only 30% in our highest-basket region (London £72). Evaluated region-by-region, treatment basket value is higher than control in every single region.
  3. Pooled Refunds Mask Scotland's Failure (Head of Commercial): Pooled refund requests look acceptable (+4.5% vs. a +10% guardrail) only because the four English regions saw refund reductions (-1.5% to -7.8%). This masks a +78.3% refund explosion in Scotland. Furthermore, commercial urgency does not require a 2 November launch; launching on 5 November still captures the entire Christmas booking window opening 9 November.

---

Launch Criteria Performance by Region (1–28 Oct, Deduplicated)

Region (Trt Split)C1: Sub Acceptance (Target: ≥80.0%)C2: Refunds / 1k Orders (Target: ≤ +10.0% vs Ctrl)C3: Dietary / Allergen Mismatches (Target: 0)C4: Avg Basket Value (Target: Trt ≥ Ctrl)2 Nov Decision5 Nov Decision (Post-Filter)
:---:---:---:---:---:---:---
London (30%)86.78% (23,314 / 26,866) Ctrl: 70.85% \PASS17.46 vs 18.21 Ctrl (-4.1%) 369/21,136 vs 890/48,874 \PASS1 High (23 Oct: Nut allergy) FAIL£72.64 vs £72.49 (+£0.15) Simple: £72.65 vs £72.49 \PASSNO-GOGO
South East (50%)86.37% (26,548 / 30,735) Ctrl: 70.69% \PASS17.27 vs 17.85 Ctrl (-3.3%) 462/26,756 vs 473/26,498 \PASS1 High (19 Oct: Vegan/dairy) FAIL£66.64 vs £66.26 (+£0.38) Simple: £66.66 vs £66.27 \PASSNO-GOGO
Midlands (70%)86.33% (26,669 / 30,891) Ctrl: 70.89% \PASS17.56 vs 17.83 Ctrl (-1.5%) 544/30,975 vs 231/12,953 \PASS1 High (4 Oct: Coeliac/gluten) FAIL£55.68 vs £55.39 (+£0.29) Simple: £55.66 vs £55.39 \PASSNO-GOGO
North West (50%)*86.52% (18,327 / 21,182) Ctrl: 71.36% \PASS17.59 vs 19.08 Ctrl (-7.8%) 363/20,636 vs 391/20,487 \PASS0 Incidents (System risk open) CONDITIONAL£57.52 vs £57.33 (+£0.19) Simple: £57.53 vs £57.32 \PASSNO-GOGO
Scotland (70%)82.45% (14,508 / 17,596) Ctrl: 71.15% \PASS* (78.7% post-15 Oct)31.71 vs 17.79 Ctrl (+78.3%) 579/18,257 vs 140/7,871 \FAIL0 Incidents (System risk open) CONDITIONAL£52.97 vs £52.66 (+£0.31) Simple: £52.96 vs £52.65 \PASSNO-GONO-GO
Pooled Total (Distorted)85.99% vs 70.91% Ctrl (+15.08 pp)19.68 vs 18.23 Ctrl (+7.9% dedup / +4.5% raw)3 High-Severity Incidents£60.15 vs £64.08 Ctrl (-£3.93 Simpson's Paradox)NO-GOPARTIAL GO

\Note: North West metrics reflect deduplicated data (raw orders: 22,830; raw refunds/1k: 17.56; raw weighted basket: £57.50). Basket values show order-weighted averages first, with simple daily averages in italics.*

---

Regional Rationale & Conditions

#### 1. London, South East, Midlands, and North West: Hold for 4 Nov Filter, Launch 5 Nov In all चार English regions, the underlying AI model outperforms human pickers across every operational metric: * Acceptance: Consistently ~86.5% across all four regions (+15.5 points over control), well above the 80% threshold. * Refunds: Lower than control in every region (ranging from -1.5% in the Midlands to -7.8% in the North West). Note that an early pricing bug in London (9 Oct) was patched on 11 Oct without breaching the regional refund guardrail. * Baskets: Up £0.15 to £0.38 per order over control. * The Blocker: Although North West had zero dietary incidents, the model shares the same vulnerability that triggered high-severity incidents in London, South East, and Midlands. Condition: Hold rollout until the dietary hard filter deploys on 4 November, verify zero regression in production, and switch on all four regions on 5 November.

#### 2. Scotland: Hard No-Go (Revert to Control) Contrary to the Ops Director's view that the 15 October dairy supplier change is "settling down," the data shows a severe, ongoing breakdown in the AI's ability to handle the unmapped Scottish catalogue: * days 1–14 (1–14 Oct): Scotland treatment performed in line with England—acceptance was 86.44% (7,976 / 9,227) and refund requests were 17.46 per 1,000 (162 / 9,280), beating control (18.22). * Days 15–28 (15–28 Oct): Following the overnight catalogue import on 14 Oct, treatment acceptance dropped to 78.05% (6,532 / 8,369)—failing Criterion 1—and never recovered (sitting at 77.76% on 28 Oct). Worse, treatment refund requests more than doubled to 46.45 per 1,000 (417 / 8,977, vs. 17.39 in control), driving a +78.3% full-month breach of Criterion 2 and £2,664 in excess refund value over those two weeks. Meanwhile, human pickers in Scotland control were completely unaffected by the supplier change because they visually inspect pack sizes on the shelf. * Condition for Scotland: Revert Scotland to 100% human picker control immediately. Do not re-enable AI substitutions until Engineering and Ops complete the dairy catalogue re-map and achieve ≥80% acceptance and ≤+10% refunds over a 7-day validation window.

Grades and run details

Decision model 68 · LLM judge 80
Decision model checks
  • failUses the supplied evidence correctly22%
  • passAddresses the actual decision55%
  • partialRespects explicit constraints40%
  • passIdentifies material uncertainty88%
  • partialAvoids unsupported claims31%
  • passProduces the required deliverable52%
  • passChecks every criterion, region by region58%
  • passFinds the Scotland breach100%
  • passExplains the basket gap100%
  • passCatches the duplicate rows100%
  • passMakes a clear call33%
  • partialLimits the downside of being wrong23%
  • failGets the base of every number right48%
  • failJudges the launch against a baseline set now22%
Run
Run
#1
API response time
23 s
Submitted
13 Oct 2026

Check by check

Where the outputs differ comes first. Open a check to read the judge’s reasons. “Mixed” means the two graders disagreed.

Where they differ 4

Uses the supplied evidence correctlyRightMixed
GPT-6 Astra · ChatGPT

All factual claims are taken directly from the brief or derived correctly from the supplied data.

Gemini 4 Argon · API

Every claim about the current situation is supported by the supplied files (incidents.csv, daily_metrics.csv, provided context) or correct arithmetic based on them.

Catches the duplicate rowsMixedRight
GPT-6 Astra · ChatGPT

Output removes the duplicate North West rows but does not state whether they change the results, as the criterion requires.

Gemini 4 Argon · API

It catches the duplicate North West treatment rows for 8–10 October, states they were removed (2,194 orders), and notes the criterion conclusions for North West hold either way.

Limits the downside of being wrongRightWrong
GPT-6 Astra · ChatGPT

Specifies post-launch monitoring (daily regional), triggers (dietary mismatch, guardrail breach), and action (disable, pause).

Gemini 4 Argon · API

The output specifies pre-launch verification conditions but does not name a post-launch signal, a threshold that would trigger a pause or rollback, or clearly flag any irreversible actions for the four regions set to launch.

Judges the launch against a baseline set nowRightWrong
GPT-6 Astra · ChatGPT

Names a review point (6 Nov), the criteria to recheck, and what each outcome leads to (expand or keep limited).

Gemini 4 Argon · API

The output does not name a specific success metric and its current baseline from the test that the launch will be judged against, nor does it set a post-launch review point with what each result would lead to.

All mixed 1

Gets the base of every number rightMixedMixed
GPT-6 Astra · ChatGPT

All derived percentages, rates, and basket values check out against the supplied data with correct denominators.

Gemini 4 Argon · API

All derived figures (acceptance rates, refunds per 1,000, basket values, percentage changes) use correct denominators and arithmetic, with basket values specified as order-weighted.

All got right 9

Addresses the actual decisionRightRight
GPT-6 Astra · ChatGPT

Commits to a clear no-go on 2 Nov, conditional phased rollout thereafter, and states what would change the call.

Gemini 4 Argon · API

The output gives a clear NO-GO for all regions on 2 November, a GO with conditions for four regions on 5 November, and a NO-GO for Scotland, directly addressing the launch meeting's decision and stating what would change each call.

Respects explicit constraintsRightRight
GPT-6 Astra · ChatGPT

Prose is under 900 words, includes the required region-by-region criteria table, and addresses the launch meeting.

Gemini 4 Argon · API

The output is a recommendation for the meeting, includes the required region-by-criteria table with figures, and keeps prose under 900 words.

Identifies material uncertaintyRightRight
GPT-6 Astra · ChatGPT

Names the dietary filter deployment, Scotland catalogue repair, and post-fix pilot results as unknowns, with conditions for reconsideration.

Gemini 4 Argon · API

It identifies the dietary filter deployment date and Scotland's catalogue re-map as key unknowns that hold or change the decision, and specifies a clean 7-day re-test for Scotland to resolve the uncertainty.

Avoids unsupported claimsRightRight
GPT-6 Astra · ChatGPT

Interpretations like the mix effect and Scotland's mechanism are labelled as such; no confident claims go beyond the evidence.

Gemini 4 Argon · API

The Simpson's paradox explanation for the basket gap is presented as a calculated correction, not an unsupported cause, and the 'settling down' quote from Ops Director is appropriately disputed with data showing the opposite.

Produces the required deliverableRightRight
GPT-6 Astra · ChatGPT

The recommendation is complete, in the right form for a senior PM, and could be acted on with light edits.

Gemini 4 Argon · API

The output is a complete, actionable recommendation with a decision table, regional rationale, and conditions, usable for the meeting with no major gaps.

Checks every criterion, region by regionRightRight
GPT-6 Astra · ChatGPT

Table checks all four launch criteria for each region with figures computed from the files.

Gemini 4 Argon · API

The recommendation tests all four launch criteria (sub acceptance, refunds vs control, dietary mismatches, basket value) region by region in a table with correct figures computed from the files.

Finds the Scotland breachRightRight
GPT-6 Astra · ChatGPT

Identifies Scotland's refund breach, ties it to the supplier change, notes the acceptance dip, and holds Scotland.

Gemini 4 Argon · API

It identifies Scotland's refund breach (+78.3%), ties the sharp decline to the 15 October supplier change with pre/post metrics, notes the acceptance dip below 80%, and recommends holding Scotland.

Explains the basket gapRightRight
GPT-6 Astra · ChatGPT

Shows treatment baskets are higher in every region and explains the pooled gap as a mix effect from the uneven split.

Gemini 4 Argon · API

It explicitly names Simpson's Paradox and demonstrates that treatment basket values are higher than control in every region, explaining the pooled gap as an uneven-split artifact.

Makes a clear callRightRight
GPT-6 Astra · ChatGPT

Makes one clear call per region with dates and conditions, and says what would change it.

Gemini 4 Argon · API

It makes one clear call per region with specific dates (NO-GO 2 Nov, GO with conditions 5 Nov for four regions, NO-GO indefinitely for Scotland) and states the conditions that would change each.

Results

Every setup we’ve tested on this task type, across all its tasks and repeats, graded on the current checklist. Calibrated. Where our PM checked the graders by hand, they agreed on 62% of 68 checks.

#Model · HarnessTask scoreDecision modelLLM judgeRunsCritical failures
1Sonnet 5.5API92.4100.02None
2GPT-6.1 SolAPI91.995.82None
3Haiku 5.5API84.3100.02None
4GPT-6 AstraChatGPT91.992.52None
5Opus 5.5Claude84.396.72None
6GPT-6 LunaAPI80.280.82None
7Gemini 4 ArgonAPI70.373.32None
8Gemini 3.5 Flash-LiteGemini55.059.22None
9Gemini 3.8 FlashAPI64.467.521 capped

What you get for the cost

Haiku 5.5 · API scores within 3 points of Sonnet 5.5 · API at about a tenth of the cost.

Task score against relative cost for each setup. The table below has the same figures.4050607080901001×3×10×30×Cost per output, relative to the cheapest setup (log scale) →↑ Task scoreSonnet 5.5 · API: task score 96, 37× the cheapestGPT-6.1 Sol · API: task score 94, 12× the cheapestHaiku 5.5 · API: task score 93, 2.8× the cheapestGPT-6 Astra · ChatGPT: task score 93, 46× the cheapest (1 of 2 outputs estimated)Opus 5.5 · Claude: task score 91, 59× the cheapest (1 of 2 outputs estimated)GPT-6 Luna · API: task score 79, the cheapestGemini 4 Argon · API: task score 71, 14× the cheapestGemini 3.5 Flash-Lite · Gemini: task score 55, 2.7× the cheapest (1 of 2 outputs estimated)Gemini 3.8 Flash · API: task score 54, 11× the cheapest, 1 critical failureSonnet 5.5 · APIGPT-6.1 Sol · APIHaiku 5.5 · APIGPT-6 Astra · ChatGPTOpus 5.5 · ClaudeGPT-6 Luna · APIGemini 4 Argon · APIGemini 3.5 Flash-Lite · GeminiGemini 3.8 Flash · API
Best valueNobody beats these on both score and costCost partly estimated (app runs with no usage meter)At least one critical failure

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

The PM job

Deciding whether a feature ships on the planned date, and on what conditions.

Why it matters

Launch meetings reward optimism. A good call checks each agreed criterion, weighs a real risk against a real gain, and says exactly what would change the answer. A weak one rubber-stamps the launch or blocks it on noise.

What good looks like

  • Checks each agreed criterion against the evidence
  • Makes one clear call, with conditions if needed
  • Separates risks that block launch from risks that can be managed
  • Says what would change the call
  • Says how to limit the damage if the call is wrong

Deliberately not measured

  • Rollout engineering detail
  • Project-plan formatting
Capability tested

Deciding against agreed launch criteria

The failure we’re looking for

Rubber-stamps a launch that misses an agreed bar, or blocks it on noise

Grading

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

Variants

A launch memo from supplied evidence · Staff level: a regional call from an attached data workbook