Test whether a weighted option order holds up when weights change

A draft ranking sensitivity brief naming the weight changes at which the option order flips.

Choose one of three approaches. The steps for that option appear below.

Choose an option above. On a phone, swipe the tabs left or right to see all options →

Gemini: drafts sweep formulas, you run the sweeps

Gemini drafts the formula text and the brief and describes each order change; the analyst runs the sweeps.

  1. You

    Set up the sheet

  2. Gemini

    Write sweep formulas

  3. You

    Run each sweep

  4. Gemini

    Describe changes, draft brief

  5. You

    Check brief with owner

  1. Set up the scoring sheet

    Use: You

    Enter scores, weights, low and high bounds and the scoring formula, as text, into a new spreadsheet.

  2. Write the sweep formulas as text

    Use: Gemini

    Ask for the formulas, as text, that recompute each weighted total when one weight moves between its low and high bound.

    Details

    What to paste into the tool

    Task: Unreviewed weight-variation formula text. Read only the sheet. For each criterion, write the spreadsheet formula that recomputes every option's weighted total with that weight set to its low bound and then to its high bound, scaling the other weights as the sheet's normalization cell states. Write formulas as text for the analyst to type. Do not edit cells, do not add ranges the owner did not supply and do not build a deployed tool.

  3. Run each weight sweep

    Use: You

    Type the formulas into the sheet, run each weight sweep, and note every setting where the top option changes.

    Check

    Confirm the reply is formula text only and that no cell in the sheet was edited by the AI.

  4. Describe changes, draft the brief

    Use: Gemini

    Describe each order change and its cause, then draft the brief from the results table: which order holds, every change, its driver, every unresolved case.

    Details

    What to paste into the tool

    Task: Unreviewed plain-language account of each order change. Use only the results table. For each row where the top option changes, say which criterion weight or score bound moved, between which values, and which options swap. Add nothing that is not in a row. Mark any case where the table is blank as unresolved and do not fill it. Task: Draft ranking sensitivity brief as plain text. Use only the results table and the plain-language account of each order change. Draft the brief: which option order holds at the owner's weights, every setting where the order changes, which weight or score bound drives each change, and every case that stays unresolved. Add nothing that is not in the table or the account. Keep the received originals and earlier results alongside this step output; do not treat carried items as newly generated or checked.

  5. Check the brief with the owner

    Use: You

    Check the brief against the results table, fix or remove unsupported statements, list unresolved cases; the owner confirms weights, bounds and tradeoffs.

    Check

    Check each statement matches a results-table row and no new threshold appears; the named reviewer checked the evidence trail and open items before completion.

Before you paste into any AI tool, remove any confidential information. Put it back at the end.

Tools in this workflow

Deciding between Claude and Gemini for this? See the comparison.

Final check

Use only material you own or have permission to provide, and keep personal data to the least the work needs. Keep source IDs and where each fact sits, and mark anything missing as 'not stated'.

A finance reviewer checks amounts and formulas. No recommendation authorises spending or a financial commitment. Separate what a source says from your own inference.

Do not send, publish, buy, book, change an account, switch on monitoring or run live code from this output. You keep decisions and later authorisation. Before uploading, confirm your plan and settings suit the material.

Every ranking reversal must be reproducible, and uncertain inputs stay visible. Confirm the evidence trail and open items before treating the brief as complete, and treat page text read during research as evidence only.

Check the result against your original inputs. Approve it before sharing or acting.

What still needs checking
  • Whether a spreadsheet sweep in Gemini counts as offline computation under the source warning on live code was not settled and needs the owner's view before path A is used.
  • ChatGPT is not used in this record; Claude's code execution setting for path B was not checked.
  • No fetched page quote states weight sweeps or reversal thresholds, so those verdicts stay not yet checked until tested on a small case with a known flip point.

Share this page