OdysseysMega

A colleague's family is formalizing their annual charitableCommunity and Society

7 rubrics · 7 LLM-graded

TaxonomyCommunity and SocietyPhilanthropy

How each model performed on this task. Every chart states its N, and the reward carries a 95% confidence interval once there is more than one run to measure.

Core metrics

Check Pass %

Mean share of the task's checks passing, averaged over runs. N = 1 run per model.

  1. 86%

    GPT-5.6 sol

  2. 29%

    Claude Opus 5.5

  3. 14%

    Claude Opus 5

  4. 0%

    GPT-6 sol

Extended metrics

Median Time

Wall-clock minutes, median across runs. N = 1 run per model.

  1. 51m

    Claude Opus 5

  2. 38m

    GPT-5.6 sol

Tool Calls

Tool invocations, median across runs — counted from the trajectories. N = 1 run per model.

  1. 117

    Claude Opus 5

  2. 100

    Claude Opus 5.5

  3. 47

    GPT-5.6 sol

  4. 2

    GPT-6 sol

Verifier × model pass rate

How often each verifier passed, per model. Green is passing, red is failing — a red band across a row is a verifier nothing clears.

VerifierClaude Opus 5.5n=1GPT-6 soln=1GPT-5.6 soln=1Claude Opus 5n=1
checks (7)
100%0%0%0%

How to pass

Browse Charity Navigator's ocean/environment cause category (charitynavigator.org) to build a 15–20 organization longlist spanning both large and small/regional groups.

100%0%100%0%

How to pass

projects.propublica.org/nonprofits pull the most recent Form 990 for each candidate; record program expense ratio and check top leadership compensation against total budget.

0%0%100%100%

How to pass

apps.irs.gov/app/eos confirm current 501(c)(3) status is active (not revoked) and filings are current for each candidate.

0%0%100%0%

How to pass

Search each organization's name + "investigation" / "controversy" / "lawsuit"; review coverage from the last 2 years and note anything unresolved.

0%0%100%0%

How to pass

Confirm Charity Navigator star rating and note the visible Candid/GuideStar Seal tier (Bronze/Silver/Gold/Platinum) where publicly displayed. If the profile view limit is reached, note that the seal tier couldn't be verified for that organization rather than guessing.

0%0%100%0%

How to pass

Narrow to the final list meeting the size-mix requirement; propose the tiered $50,000 split with a short reasoning line per organization.

0%0%100%0%

How to pass

Build the tracker with the five tabs specified, using a spreadsheet tool that doesn't require login (such as CryptPad), fully populated, plus the final allocation with reasoning. Leave the sheet open for review.

Failure modes on this task

19 tagged verifier failures, by the stage they were assigned to. Higher is worse.

Failure stageClaude Opus 5.5GPT-6 solGPT-5.6 solClaude Opus 5
Delivering the result1 mode12—3
Reaching the page1 mode———1
Standing up the evidence1 mode———1
No signal in the judge’s words1 mode · the grader4511