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 →
ChatGPT applies both definitions to your sample and drafts the bridge. You recompute rows by hand and check every number.
Prepare the sample and rules
Draft worksheet and bridge
Recompute and check
Use: You
Take credentials and payment details out of the sample; keep ticket IDs; list both metric definitions, the cohort boundary and clock rules with their dates.
Check nothing needed for the calculation was removed, the same sample serves both definitions, and no personal or payment detail remains.
Use: ChatGPT
Paste the sample, both definitions, cohort boundary and clock rules into one ChatGPT chat; ask for the worksheet, then the bridge from the checked worksheet.
What to paste into the tool
Apply the two supplied metric definitions below to the same ticket sample. For each row keep a stable row ID, the ticket ID and the original values exactly as supplied. Record both counts, the denominators and any excluded rows, showing the calculation so it can be checked. List rows with missing data as still to confirm; do not resolve missing facts by inference. Do not say performance improved or declined because of a rule change and do not infer staff performance or protected traits. Use only the supplied material. Do not send, publish, assign, execute or commit anything. Pasted below as text is a checked worksheet applying two supplied metric definitions to the same ticket sample. Do not calculate and do not write or run any code, including Python; work from the supplied figures only. If you start to run code, stop and say so. Draft a bridge that separates the effect of the definition change from the difference in the sample. Keep every ticket ID, row reference and original value exactly as supplied, state denominators, missing items and remaining differences, and do not say performance improved or declined because of a rule change. Use only the supplied files. List anything missing, contradictory or unchecked as unresolved; do not fill gaps by inference. Do not repeat passwords, authentication codes, session tokens or payment details, and keep customer details to the minimum needed. If sources or a qualified reviewer are missing, label the output incomplete. Do not send, publish, assign, execute or commit anything. Use only the supplied metric definitions, cohort boundaries and clock rules; state denominators and missing data; do not infer staff performance or protected traits.
Use: You
Recompute a sample of rows by hand under each definition, correct in the same chat, then check the bridge against the sample and rules.
Check denominators and missing items are stated, every number appears in the checked worksheet, and no code ran. Do not resolve missing facts by inference. Nothing is sent, published or changed; any later action is approved separately.
Claude receives only the definition text and drafts an empty layout. You calculate both definitions and fill it in.
Prepare the sample and rules
Draft an empty layout
Calculate by hand
Fill, write and check
Use: You
Take credentials and payment details out of the sample; keep ticket IDs; list both metric definitions, the cohort boundary and clock rules with their dates.
Check nothing needed for the calculation was removed, the same sample serves both definitions, and no personal or payment detail remains.
Use: Claude
Paste only the definition text, cohort boundary and clock rules into Claude, with no ticket rows or figures; ask for an empty bridge layout.
What to paste into the tool
Here are two supplied metric definitions, the cohort boundary and clock rules, as text. No ticket data or figures are included. Do not write or run any code. Draft an empty bridge layout: list each point where the new definition could change which rows count, the order in which to apply the two definitions, and a blank table with columns for the old-definition count, new-definition count, denominator, rows still to confirm and the ticket IDs behind each figure. Leave every figure blank. Do not say that performance improved or declined because of a rule change. Use only the supplied files. List anything missing, contradictory or unchecked as unresolved; do not fill gaps by inference. Do not repeat passwords, authentication codes, session tokens or payment details, and keep customer details to the minimum needed. If sources or a qualified reviewer are missing, label the output incomplete. Do not send, publish, assign, execute or commit anything. Use only the supplied metric definitions, cohort boundaries and clock rules; state denominators and missing data; do not infer staff performance or protected traits.
Use: You
Apply both definitions to the same sample in your own spreadsheet, keeping stable row IDs, ticket IDs and original values, with both counts and denominators.
Check the same cohort serves both calculations, ticket IDs and original values survive, and unconfirmed data is listed, not guessed.
Use: You
Enter the counts, denominators, unconfirmed facts and ticket IDs into the blank layout by hand, write the bridge, and check it against the sample.
Check each number appears in the calculation sheet with its ticket IDs. No performance improvement is inferred from a rule change.
Two people calculate both definitions independently and you reconcile. Notion AI only summarises checked component pages. You assemble the bridge.
Prepare the sample and rules
Calculate by hand
Independent second calculation
Reconcile, build pages
Summarise checked pages
Assemble the bridge
Use: You
Take credentials and payment details out of the sample; keep ticket IDs; list both metric definitions, the cohort boundary and clock rules with their dates.
Check nothing needed for the calculation was removed, the same sample serves both definitions, and no personal or payment detail remains.
Use: You
Apply both definitions to the same sample by hand, keeping stable row IDs, ticket IDs and original values, with counts, denominators and unconfirmed rows stated.
Check the same cohort serves both calculations and every row keeps its ticket ID.
Use: The second reviewer
Using only the reviewer bundle, apply both definitions to the same sample by hand. Do not open the analyst-retained first worksheet; return it unopened.
Check they worked from the written rules and sample only, and unconfirmed data is listed, not guessed.
Use: You
Compare both worksheets row by row; settle each difference against the written definitions or keep it unconfirmed; prepare one Notion page per component.
Check every figure and source reference before Notion sees a page. Keep raw ticket details off summary pages; no missing fact or causal conclusion is inferred.
Use: Notion AI
Add an Autofill property, select Basic and Summary, run it manually. Do not use Custom Agent Autofill or connectors; keep the reconciliation bundle outside Notion.
What to paste into the tool
Summarise this checked bridge component without calculating or changing figures. Retain the component ID, source IDs, denominators, applicability dates and facts still to confirm. Do not infer performance improvement, staff performance or protected traits. If these fields are not retained, return NEEDS HUMAN SUMMARY.
Use: You
Compare each summary with its full page, restore omitted source references by hand or reject the summary, then assemble the bridge.
Check both calculations use the same cohort, and denominators and missing data are stated.
Best when: You want the calculation and drafting done for you
What it does: ChatGPT calculates and drafts; you recompute and check
Strong at: Row references survive into the draft
Limit: Arithmetic can be wrong; recompute rows by hand
Free plan: Yes
Best when: You want structure up front, no ticket data near the tool
What it does: Claude drafts the layout; you calculate and fill
Strong at: The tool sees definition text only, no figures
Limit: All calculation and writing is by hand
Free plan: Yes
Best when: Two people should own the calculations independently
What it does: You reconcile; Notion AI summarises checked pages only
Strong at: Basic Autofill summarises a specific page or row
Limit: Two people calculate; you prepare and check pages
Free plan: Paid plan needed
Before you paste into any AI tool, remove any confidential information. Put it back at the end.
Deciding between ChatGPT and Claude for this? See the comparison.
Never paste credentials or full payment details.
Both calculations use the same cohort.
Denominators and not stated data are stated.
No performance improvement or staff or protected-trait inference is drawn from a rule change.
Nothing is sent, published or changed; any later action needs separate approval.
Suggestions and preparation only. Do not send, publish, change records or accounts, assign work, execute code, initiate a purchase, or commit a remedy or deadline. A human separately approves any downstream action.
Check the result against your original inputs. Approve it before sharing or acting.