Humanity’s Last PRD
Latest drop: Haiku 5.5 · 8 Oct
Tasks / Leadership

Write a stakeholder update

Can the model turn messy project status into an honest update that leads with what the reader needs to know or decide?

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 16 graded outputs by 8 models. 56% were usable with at most a quick edit.

Reliably right

  1. Reports outcomes and learning, not just output100% pass
    Leads with outcome metrics (ARR, active teams, SLA, pipeline) and what the team learned (pipeline redefinition, invoice ticket surge), not just activities.
    Sonnet 5.5 · API · On track, mostly
  2. Leads with the honest headline100% pass
    First slide headline includes the retention drop alongside the new-ARR beat, not just 'record quarter'.
    GPT-6 Astra · ChatGPT · A quarterly business review deck, in brand
  3. A real 7-slide deck100% pass

Where it slips

  1. Owns the problem55% pass
    Does not state what the team has already done about the payments reliability miss or the NPS drop; only reports the problems.
    GPT-6 Astra · ChatGPT · A quarterly business review deck, in brand
  2. Identifies material uncertainty61% pass
    The output identifies the overlap uncertainty for decision A but does not name any material unknown that could change the call for decision B, such as the reliability of the Payments lead's estimate or the impact of pausing activation experiments.
    Opus 5.5 · Claude · A quarterly business review deck, in brand
  3. Flags the redefined pipeline63% pass
    It names the redefinition but does not give the falling like-for-like trend (e.g., pipeline fell from $2.41M to $2.23M on the old definition).
    GPT-6.1 Sol · API · On track, mostly

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. Asks leaders only for the big calls

    Does the update separate the decisions it needs from its readers (hard to reverse, high stakes) from the ones the team will make itself, and ask clearly for the former?

    Passes when Names the decisions it needs from the readers, with a recommendation and a date, and says which calls the team is making itself.

    Geoff CharlesLenny’s Podcast

  2. Reports outcomes and learning, not just output

    Does the update report progress as movement on the outcomes the team is accountable for and what it has learned, rather than as features shipped or dates hit?

    Passes when Leads with the outcome metrics and what changed in the team's understanding, with shipping as supporting detail.

    Melissa PerriLenny’s PodcastChristian IdiodiLenny’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 · Owns the problem · Ready for the reader as delivered · and 2 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 Head of Product at Cobalt. Write this fortnight's update for the CEO and the exec team, in no more than 450 words, from the weekly metrics and the team notes below. It should say where we stand, what's wrong, what we're doing about it, and what you need from them.

What the model was given3 items: About Cobalt, weekly_metrics_weeks_1_to_8.csv, team_notes.md
About CobaltTeam scheduling software, sold per seat.
weekly_metrics_weeks_1_to_8.csv8 rows · Download
week,new_arr_signed_usd,qualified_pipeline_usd,weekly_active_teams,support_backlog_tickets,support_sla_met_pct
1,182000,2410000,8210,140,96
2,205000,2380000,8260,152,95
3,168000,2350000,8240,171,94
4,214000,2290000,8300,198,91
5,199000,2230000,8280,260,86
6,221000,3140000,8310,344,79
…
team_notes.md7 lines · Download
# Notes from the teams

- Sales ops: from week 6, "qualified pipeline" includes active free trials over 10 seats (they used to be excluded).
- Data: weekly active teams for week 7 is missing because of a pipeline outage; it's being backfilled.
- Support: the backlog grew after the September pricing change (tickets about invoices doubled). SLA target is 90% answered within one business day.
- Quarter target: $2.4M new ARR in 12 weeks ($200,000 a week). Weeks 1 to 8 total: see the file.
- Support hiring: two agents approved, start date not set. Decision needed: whether to pause the planned price change for existing customers in week 10 until the backlog is cleared.
What a strong answer doesThe answer key the graders mark against

Leads with the takeaway: new ARR is ahead of plan ($1.67M in eight weeks against $1.6M), but support is breaking and pipeline is weaker than it looks. Flags that the jump in qualified pipeline in week 6 (from about $2.23M to $3.14M) is the redefinition to include large trials; on the old definition pipeline fell every week from $2.41M to $2.23M, a risk for the rest of the quarter. Reports the support problem plainly: backlog from 140 to 455 tickets and SLA from 96% to 72% against a 90% target since the September pricing change, with what's being done (two agents approved). Notes week 7's active teams are missing and being backfilled rather than hiding it. Asks for the decisions that need the execs: pause the week-10 price change for existing customers until the backlog clears, and set the start date for the support hires. Separates what the team is handling from what needs them.

Critical failures (cap the score at 40)
  • Presents qualified pipeline as having grown about 40%
Task

Staff-level · v1.0

The outputs

8 setups answered this brief.

Sonnet 5.5API98Usable with a quick editGPT-6 AstraChatGPT96Usable with a quick editGPT-6.1 SolAPI92Usable with a quick editHaiku 5.5API87Needs more workOpus 5.5Claude79Needs more workGPT-6 LunaAPI77Usable with a quick editGemini 3.8 FlashAPI63Needs more workGemini 3.5 Flash-LiteGemini58Needs more work
Gemini 3.8 FlashAPI

Needs more than a quick edit7 of 12 checks passed

The output invents team actions not in the supplied context, requiring evidence re-checking, and omits the like-for-like pipeline trend.

Output

Executive Update: Weeks 7–8

Where We Stand * ARR Tracking Ahead: We closed $248k in Week 8, bringing cumulative New ARR to $1.673M across Weeks 1–8. We are currently pacing ahead of our $2.4M quarterly target ($1.6M pro-rata benchmark), requiring ~$182k/week over the remaining four weeks to hit goal. * Pipeline Shift: Qualified pipeline rose to $3.35M, though this reflects a definition change implemented in Week 6 (now including >10-seat free trials) rather than organic acceleration alone. * Team Activity: Weekly Active Teams dropped to 8,190 in Week 8 (down from 8,310 in Week 6, and below Week 1 levels). Week 7 telemetry was missed due to a data pipeline outage and is currently being backfilled.

---

What’s Wrong * Support Capacity Crisis: Support health has deteriorated significantly. The ticket backlog reached 455 in Week 8 (more than triple Week 1), and SLA compliance fell to 72% (well below our 90% target). * Pricing Friction: The support surge was triggered by the September pricing update, which doubled billing- and invoice-related inquiries. * Engagement Dip: The drop in active teams alongside declining support SLAs risks customer sentiment and seat churn, especially in our core per-seat model.

---

What We’re Doing About It * In-App Billing UX: Product and engineering are reviewing self-serve billing flows, invoice clarity, and in-app account management to deflect common invoice-related confusion and reduce inbound tickets. * Data Reliability: The data team is actively backfilling the Week 7 outage and auditing pipeline stability to prevent future gaps in engagement reporting. * Triage Support: Product is assisting Support in categorizing recurring invoice friction points to create reusable macros and help documentation to accelerate ticket resolution.

---

What I Need From You 1. Pause the Week 10 Price Increase: We strongly recommend postponing the planned existing-customer price change scheduled for Week 10 until the support backlog normalizes and SLAs return above 90%. Executing this change now risks overwhelming Support and accelerating churn. 2. Expedite Support Hires: Finalize immediate start dates for the two already-approved support agents to stabilize queue times.

Claims the judge couldn’t find in the brief

  • Product and engineering are reviewing self-serve billing flows, invoice clarity, and in-app account management
  • Data team is auditing pipeline stability to prevent future gaps
  • Product is assisting Support in categorizing recurring invoice friction points to create reusable macros and help documentation

Grades and run details

Decision model 79 · LLM judge 54
Decision model checks
  • partialUses the supplied evidence correctly17%
  • passAddresses the actual decision35%
  • passRespects explicit constraints53%
  • partialIdentifies material uncertainty71%
  • partialAvoids unsupported claims63%
  • passProduces the required deliverable84%
  • partialFlags the redefined pipeline99%
  • passReports the support breach plainly100%
  • partialOwns the problem38%
  • passReady for the reader as delivered42%
  • passAsks leaders only for the big calls75%
  • passReports outcomes and learning, not just output94%
Run
Run
#1
API response time
12 s
Submitted
6 Oct 2026
Opus 5.5Claude

Needs more than a quick edit9 of 12 checks passed

The output invents two team actions not in the supplied context, so a PM would need to verify or replace them before using the update.

Output

Fortnightly Product Update: Weeks 7–8

Decision needed this week: pause the week-10 price change for existing customers until the support backlog is cleared. I recommend we pause.

Where we stand

  • New ARR is ahead of plan. We've signed $1,673,000 in weeks 1–8 against $1,600,000 planned, which puts us $73,000 ahead. To reach $2.4M we need $727,000 over the last four weeks, or about $182,000 a week. We've averaged $226,000 a week over the last four weeks, and week 8 was our best at $248,000.
  • Usage is flat. Weekly active teams have held at 8,190–8,310 all quarter. Week 8 (8,190) is the lowest so far, down 120 from week 6. That's small (about 1.5%), but it's the direction to watch. Week 7 is missing because of a data pipeline outage and is being backfilled.

What's wrong

1. Support is failing its SLA, and the gap is widening. - The backlog has grown from 140 to 455 tickets, more than three times its week-1 level. - SLA has dropped from 96% to 72% against a 90% target. We've been below target since week 5. - The cause is the September pricing change, which doubled invoice tickets. - Backlog growth is slowing (+84, +68, +43 in weeks 6–8), but it is still growing.

2. The pipeline number overstates our position. - The jump from $2.23M to $3.14M in week 6 is mainly a definition change. Active free trials over 10 seats are now counted as qualified. - On the old definition, pipeline fell 7% in weeks 1–5, from $2.41M to $2.23M. - We don't yet know whether it has recovered. Next quarter is the real risk, not this one.

What we're doing

  • Support: two agents are approved, but neither has a start date. I'm pushing to get them in before week 10. Product is also reviewing the most common invoice questions so we can fix the billing screens and invoice wording that generate them.
  • Pipeline: I've asked Sales Ops to restate weeks 6–8 on the old definition and report both numbers going forward.
  • Usage: once the week-7 backfill lands, we'll check whether the week-8 dip is real and which segments it comes from.

What I need from you

  1. Decide whether to pause the week-10 price change. I recommend pausing. The last pricing change doubled invoice tickets, and support is already 18 points below SLA. A second change for existing customers would land on the same overloaded team and the same customers. I suggest we resume once SLA is back above 90% for two consecutive weeks. We need a decision by the end of week 9 so customer communications can be changed.
  2. Set start dates for the two approved support agents this week.
  3. Treat the pipeline figure as inflated in board and forecast material until the like-for-like restatement is in.

Claims the judge couldn’t find in the brief

  • Product is also reviewing the most common invoice questions so we can fix the billing screens and invoice wording that generate them.
  • I've asked Sales Ops to restate weeks 6–8 on the old definition and report both numbers going forward.

Grades and run details

Decision model 96 · LLM judge 69
Decision model checks
  • passUses the supplied evidence correctly42%
  • passAddresses the actual decision98%
  • passRespects explicit constraints27%
  • passIdentifies material uncertainty80%
  • partialAvoids unsupported claims35%
  • passProduces the required deliverable86%
  • passFlags the redefined pipeline100%
  • passReports the support breach plainly100%
  • passOwns the problem68%
  • passReady for the reader as delivered77%
  • passAsks leaders only for the big calls85%
  • passReports outcomes and learning, not just output99%
Run
Run
#1
Time to output
17 s
Submitted
5 Oct 2026
GPT-6 AstraChatGPT

Usable with a quick edit12 of 12 checks passed

The output is accurate, complete, and well-structured; a product manager could use it with at most light wording edits, without re-checking evidence or redoing reasoning.

Output

Fortnightly product update — end of week 8

Bottom line: New ARR is ahead of pace, but support is deteriorating and usage is flat to slightly down. I recommend pausing the week-10 price change for existing customers until we restore support capacity and service levels.

Where we stand

  • New ARR: $484k signed in weeks 7–8, up 15% on the prior fortnight. We’ve signed $1.673M against a $1.6M pace target, putting us $73k ahead. We need another $727k over four weeks—about $182k per week—to hit the $2.4M quarter target.
  • Pipeline: Reported qualified pipeline is $3.35M, but the jump from week 6 includes a definition change: active free trials above 10 seats now count. We should not treat that increase as evidence of stronger demand. Before the change, pipeline had declined for five consecutive weeks.
  • Usage: Weekly active teams reached 8,190 in week 8, down 1.4% from week 6 and broadly flat versus week 1. Week 7 is missing due to a data outage, so we cannot assess the full fortnight’s trajectory.

What’s wrong

Support backlog increased from 344 to 455 tickets this fortnight, while one-business-day SLA attainment fell from 79% to 72%, against a 90% target. Invoice tickets doubled after September’s pricing change. Another price change risks adding demand to an already overloaded queue.

Bookings are encouraging, but they do not establish healthy adoption or retention; the usage and support signals warrant attention.

Priorities for the next fortnight

  • Support/Product: Triage invoice issues, identify recurring billing causes, and prioritise fixes and customer guidance that reduce repeat contacts. Track backlog and SLA daily.
  • Data/Sales Ops: Complete the week-7 usage backfill and provide a like-for-like pipeline view separating newly included trials.
  • Support hiring: Confirm start dates and ramp plans for the two approved agents.

What I need from the exec team

  1. Approve pausing the week-10 existing-customer price change. Resume only once the backlog is cleared and SLA performance is sustainably back at or above 90%.
  2. Secure start dates for both support hires this week and agree interim coverage if they cannot start before week 10.

Grades and run details

Decision model 92 · LLM judge 100
Decision model checks
  • passUses the supplied evidence correctly48%
  • passAddresses the actual decision99%
  • passRespects explicit constraints43%
  • passIdentifies material uncertainty37%
  • partialAvoids unsupported claims31%
  • passProduces the required deliverable94%
  • passFlags the redefined pipeline96%
  • passReports the support breach plainly100%
  • partialOwns the problem38%
  • passReady for the reader as delivered78%
  • passAsks leaders only for the big calls85%
  • passReports outcomes and learning, not just output98%
Run
Run
#1
Time to output
17 s
Submitted
5 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 6

Uses the supplied evidence correctlyWrongMixedRight
Gemini 3.8 Flash · API

Claims about current team actions (product/engineering reviewing billing flows, data team auditing, product assisting support) are not in the supplied context and are invented.

Opus 5.5 · Claude

Two claims about current team actions (product reviewing invoice questions, asking Sales Ops to restate) are not in the supplied context and are therefore invented.

GPT-6 Astra · ChatGPT

All factual claims about the current situation are directly supported by the supplied metrics or team notes, with no invented facts.

Identifies material uncertaintyWrongRightRight
Gemini 3.8 Flash · API

Does not name specific unknowns that could change the decision (e.g., whether support hires will be enough, whether billing UX changes will reduce tickets) or bound them.

Opus 5.5 · Claude

Names the unknown old-definition pipeline trend and the uncertain week-8 usage dip, and says how each will be resolved (restatement, backfill).

GPT-6 Astra · ChatGPT

It identifies the missing week 7 data, the pipeline redefinition, and the condition for resuming the price change, naming the unknowns that affect the decision.

Avoids unsupported claimsWrongWrongRight
Gemini 3.8 Flash · API

Presents team actions (billing UX review, pipeline auditing, support categorization) as established fact without evidence from the supplied context.

Opus 5.5 · Claude

Presents 'Product is reviewing invoice questions' and 'I've asked Sales Ops to restate' as established facts without support from the supplied evidence.

GPT-6 Astra · ChatGPT

Forecasts and interpretations are labelled as risks or opinions (e.g., 'risks adding demand', 'warrant attention'), not presented as established facts.

Flags the redefined pipelineWrongRightRight
Gemini 3.8 Flash · API

Flags the redefinition but does not give the falling like-for-like trend (e.g., from $2.41M to $2.23M on the old definition).

Opus 5.5 · Claude

Flags the week-6 jump as a definition change and gives the like-for-like falling trend (7% drop in weeks 1–5).

GPT-6 Astra · ChatGPT

It explicitly names the redefinition (free trials over 10 seats now included) and states the like-for-like trend of five consecutive weekly declines.

Owns the problemRightMixedRight
Gemini 3.8 Flash · API

For each problem, lists actions the team is taking and recommends a way forward (pause price change, expedite hires).

Opus 5.5 · Claude

The actions 'product reviewing invoice questions' and 'asked Sales Ops to restate' are not supported by the team notes, so the update does not truthfully report what the team has already done.

GPT-6 Astra · ChatGPT

For the support problem, it lists actions already taken (triage, identify causes, fixes, daily tracking) and recommends pausing the price change with its trade-off.

Asks leaders only for the big callsMixedRightRight
Gemini 3.8 Flash · API

Does not provide a date by which the decisions are needed, only a condition for resuming the price change.

Opus 5.5 · Claude

Separates the big calls (pause price change, set agent start dates, treat pipeline as inflated) from team actions, with a recommendation and deadline.

GPT-6 Astra · ChatGPT

It separates the exec asks (approve pause, secure start dates) from the team's own priorities, with clear recommendations and timing.

All got right 6

Addresses the actual decisionRightRightRight
Gemini 3.8 Flash · API

Commits to pausing the week 10 price change and expediting support hires, with the condition for resuming (backlog normalizes, SLAs above 90%).

Opus 5.5 · Claude

Commits early to pausing the week-10 price change, gives a clear condition to resume (SLA above 90% for two weeks), and frames it for the exec team.

GPT-6 Astra · ChatGPT

The output commits early to pausing the week-10 price change, frames it for the CEO/exec team, and states the condition (backlog cleared, SLA ≥90%) that would change the call.

Respects explicit constraintsRightRightRight
Gemini 3.8 Flash · API

Update is within 450 words, addressed to CEO and exec team, and follows the requested structure.

Opus 5.5 · Claude

The update is under 450 words, addressed to the CEO and exec team, and follows the requested structure.

GPT-6 Astra · ChatGPT

The update is under 450 words, addresses where we stand, what's wrong, what we're doing, and what's needed, and is written as a fortnightly update for the CEO and exec team.

Produces the required deliverableRightRightRight
Gemini 3.8 Flash · API

Complete update with standing, problems, actions, and asks; usable by the CEO and exec team.

Opus 5.5 · Claude

The fortnightly update is complete, within the word limit, and usable by the CEO and exec team with light edits.

GPT-6 Astra · ChatGPT

The update is complete, within the word limit, structured for the named reader, and contains all required elements to act on.

Reports the support breach plainlyRightRightRight
Gemini 3.8 Flash · API

Reports backlog (455) and SLA (72%) against the 90% target, links to the pricing change, and asks for the pause decision.

Opus 5.5 · Claude

Reports backlog 140→455, SLA 96%→72% against 90% target, links to the pricing change, and asks for the pause decision.

GPT-6 Astra · ChatGPT

It reports the backlog (344 to 455) and SLA (79% to 72%) against the 90% target, links the cause to the pricing change, and asks for the pause decision.

Ready for the reader as deliveredRightRightRight
Gemini 3.8 Flash · API

Written for the CEO and exec team, no placeholders or notes to the requester, all owners and dates are clear or marked as proposed.

Opus 5.5 · Claude

Written entirely for the CEO/exec team, with no placeholders, invented names, or lifted internal notes.

GPT-6 Astra · ChatGPT

The update is written entirely for the CEO/exec team, with no placeholder names, no lifted internal notes, and all owners and dates are either from the material or clearly proposed.

Reports outcomes and learning, not just outputRightRightRight
Gemini 3.8 Flash · API

Leads with outcome metrics (ARR, pipeline, active teams) and what was learned (pipeline redefinition, support surge cause).

Opus 5.5 · Claude

Leads with outcome metrics (ARR ahead, usage flat) and what the team learned (pipeline redefinition, support cause), not just activities.

GPT-6 Astra · ChatGPT

It leads with outcome metrics (ARR, support, usage) and reports a key learning (pipeline jump is a definition change, not demand growth), not just activities.

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 76% of 89 checks.

#Model · HarnessTask scoreDecision modelLLM judgeRunsCritical failures
1GPT-6 AstraChatGPT94.292.92None
2Sonnet 5.5API84.696.42None
3Haiku 5.5API89.692.32None
4GPT-6.1 SolAPI80.896.22None
5Opus 5.5Claude86.377.52None
6GPT-6 LunaAPI76.766.82None
7Gemini 3.5 Flash-LiteGemini70.851.62None
8Gemini 3.8 FlashAPI66.351.92None

What you get for the cost

Haiku 5.5 · API scores within 5 points of GPT-6 Astra · ChatGPT at a tenth of the cost.

Task score against relative cost for each setup. The table below has the same figures.50607080901001×3×10×30×Cost per output, relative to the cheapest setup (log scale) →↑ Task scoreGPT-6 Astra · ChatGPT: task score 94, 26× the cheapest (1 of 2 outputs estimated)Sonnet 5.5 · API: task score 90, 40× the cheapestHaiku 5.5 · API: task score 89, 2.6× the cheapestGPT-6.1 Sol · API: task score 88, 15× the cheapestOpus 5.5 · Claude: task score 81, 55× the cheapest (1 of 2 outputs estimated)GPT-6 Luna · API: task score 71, the cheapestGemini 3.5 Flash-Lite · Gemini: task score 60, the cheapest (1 of 2 outputs estimated)Gemini 3.8 Flash · API: task score 58, 13× the cheapestGPT-6 Astra · ChatGPTSonnet 5.5 · APIHaiku 5.5 · APIGPT-6.1 Sol · APIOpus 5.5 · ClaudeGPT-6 Luna · 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)

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

Writing the weekly update, or the reply to an executive's question, that the team's credibility rests on.

Why it matters

Updates are where bad news gets softened. A fluent update that buries a slipped date or presents a guess as the cause does more harm than no update: leadership decides on it, and trust goes when the truth comes out.

What good looks like

  • Leads with the news and any decision needed
  • States dates, numbers and causes plainly, including bad news
  • Separates what is known from what is suspected
  • Brings a recommendation and what's been tried, not just the problem
  • Fits the reader and the length asked for

Deliberately not measured

  • Tone and formatting preferences
  • Writing style beyond clarity
Capability tested

Honest, decision-first status communication

The failure we’re looking for

Spin: burying or softening the bad news, or presenting a guess as fact

Grading

Decision model, LLM judge and a browser check (for the deck), calibrated against a blind PM review

Variants

A short written update · Staff level: a QBR deck in brand colours