Template: Non-Coding / Analytical Task Submission

For finance, health, regulatory, engineering tasks where the deliverable is a written analysis, computation, or recommendation — not runnable code.

Field 1: Prompt

TO: [Role] FROM: [Role, Name] RE: [Subject] DATE: [Date]

[CONTEXT — 2-3 sentences on what needs to be done and why]

Data

[TABLES — all data the model needs, with subtle errors embedded]

Approved Methodology

[CONSTRAINTS — what methods are allowed, what's excluded, regulatory framework]

Deliverables

[WHAT THE MODEL MUST PRODUCE — specific outputs, format requirements]

Notes

[ADDITIONAL CONTEXT — mapping tables, definitions, edge cases. More trap surface area.]

Field 2: Gemini conversation link

[PASTE SHAREABLE LINK]

Field 3: Golden solution

Executive Summary

[HEADLINE RESULT — the key number or recommendation, 2-3 sentences]

Step-by-step Analysis

Step 1: Data Quality Assessment [Which traps exist in the data, how to detect them, how to handle them]

Step 2: [Core Computation / Analysis] [Methodology applied correctly, intermediate results]

Step 3: [Second Computation if applicable] [...]

Step N: Final Result [The answer, with sensitivity analysis and limitations acknowledged]

Assumptions and Limitations

[What the model should disclose about uncertainty]

# Optional: verification script that reproduces the key numbers
# [short computation that confirms the headline result]

Field 4: Rubric / Unit-tests

Score the response across five areas:

Data quality detection ([N] pts): Did the model catch [specific trap 1]? Did it flag [specific trap 2]? Did it handle or disclose the issue rather than silently propagating it?

Methodology correctness ([N] pts): Did the model apply [required method] rather than [common shortcut]? Did it [specific methodological requirement]?

Scope discipline ([N] pts): Did the model exclude [red herring variable]? Did it stay within the approved methodology? Did it resist adding analytical extensions not requested?

Numerical accuracy ([N] pts): Is the headline result within [tolerance]% of the golden answer? Are intermediate calculations consistent?

Disclosure and limitations ([N] pts): Does the response acknowledge [specific uncertainty]? Does it discuss sensitivity to [key assumption]?

Critical fail conditions: [specific errors that show fundamental misunderstanding]

Field 5: Report / Failure Analysis

[Same structure as coding template: what it got right, core failure, secondary issues, what it reveals, hinting results]