Build an Evidence Hub with ChatGPT

Build a governed SharePoint evidence hub and verify AI-assisted decisions.

Introduction

30 Second Summary

Evaluation findings often scatter across spreadsheets, notes, and status updates. That makes it difficult to see which evidence is ready for a decision or which claims still need checking.

In this project, you will use ChatGPT to coach the build of a governed evidence hub in Microsoft SharePoint. You will verify its AI-assisted analysis in Microsoft Excel before drafting an evidence-based decision brief.

What You'll Build

Your finished hub lets you switch among saved views to see review work, decision-ready evidence, and portfolio patterns before opening a brief backed by checked Evidence IDs.

By the end of this project, you'll have:

A reusable prompt-led workflow that gives you one observable action at a time. Its Build Log records progress at every checkpoint.
A filterable evidence register where eight synthetic findings follow a controlled vocabulary. Saved views reveal the review queue, decision-ready evidence, and portfolio patterns.
A human-verified decision brief supported by a PivotTable. An AI Validation Log shows whether each AI claim was verified or corrected.
Secret Mission: Add and test a modified-item notification rule in Microsoft Lists.

Are there any prerequisites?

This project assumes a Windows device, a ChatGPT account, and existing Microsoft 365 access to Excel and SharePoint. Permission to create a SharePoint list is preferred, with the Microsoft Lists My lists location as the fallback.

Before We Start

Before any hands-on work, set the human accountability boundary for this project. ChatGPT coaches the work under a human decision-maker using Microsoft 365 as the system of record.

Build the Evidence Workbook

Your evidence hub needs a reliable starting point before anything reaches SharePoint. The raw Excel workbook deliberately contains inconsistent categories so you can diagnose a real data-quality problem.

ChatGPT coaches the build plus audits the workbook. You remain responsible for every Microsoft 365 action plus every accepted result.

In this step, get ready to:

Review your ChatGPT data controls before uploading the synthetic workbook.
Start a reusable coaching conversation with the master prompt.
Produce a saved evidence workbook plus a documented quality audit.

Start the coaching conversation

This project uses synthetic evidence so no employer or student information enters ChatGPT. Turning off model improvement keeps new conversations from being used to train OpenAI models.

Press the Windows key to open Windows search.
Type Excel into Windows search.
Press Enter to open Excel.
Open ChatGPT in your browser.
Sign in to ChatGPT.
Open Settings in ChatGPT.
Select Data controls.
Turn off Improve the model for everyone.

Your privacy control is now set before the workbook upload. New conversations can still remain in your chat history.

You also need a list location ready for the next stage. Use your preferred SharePoint site or the Microsoft Lists fallback.

SharePoint site

Open an existing SharePoint site in another browser tab.
Select + New on the site home page.
Confirm that List appears as an available option.

Seeing the list option confirms that this site can provide the preferred location for your evidence register.

Microsoft Lists fallback

Open the Microsoft 365 app launcher in another browser tab.
Select More apps.
Select Lists.
Confirm that +New list is visible.

Seeing the list creation control confirms that you can use My lists if site saving is unavailable.

Start a new conversation in ChatGPT.
Build the first part of the master prompt by pasting this text into the empty message box:

You are my hands-on project coach for building a Strategic Evaluation Evidence Hub.

My environment:
- Windows
- Excel for Microsoft 365
- Microsoft SharePoint in Microsoft 365
- Microsoft Lists
- ChatGPT

Your role:
- Guide me through one observable action at a time.
- Wait for me to reply DONE or BLOCKED before giving the next action.
- Never claim that you clicked, uploaded, changed, saved, or verified anything in my Microsoft 365 account.
- Use only the synthetic data provided below.
- Never ask me to upload employer, partner, student, or confidential information.
- If my interface differs, ask me to quote the exact labels I see. Do not invent a replacement path.
- Keep a short Build Log with Completed, Observed, and Next after every checkpoint.
- Treat SharePoint as the source of truth. Treat your own classifications and calculations as proposals until I verify them.
- Do not let me clean the raw categories before I observe how they break grouping.

What does this part control?

The environment lines keep the coaching instructions aligned with your Windows tools.
The role rules prevent ChatGPT from claiming that it performed actions inside your Microsoft 365 account.
The DONE or BLOCKED protocol keeps the conversation focused on one observable action at a time.

Complete the same unsent master prompt by pasting this second part directly below the first:

Project outputs:
1. Strategic_Evaluation_Evidence.xlsx
2. A Microsoft list named Strategic Evaluation Evidence Hub
3. Four governed Choice columns
4. Review Queue, Decision Ready, and Portfolio Learning views
5. Strategic_Evaluation_Insights.xlsx with a PivotTable, AI Validation Log, and Decision Brief
6. An optional notification rule

Approved impact vocabulary:
- systems transformation
- equitable outcomes
- agency
- engagement

Synthetic data:
Evidence ID	Initiative	Impact Area	Finding	Evidence Type	Review Status	Owner	Decision Use
E-001	North Valley Learning Network	Agency	Educators report stronger ownership of lesson design	Survey	Ready	Morgan	Scale coaching
E-002	Cedar Grove Pilot	systems transformation	District teams adopted shared implementation routines	Interview	Draft		Refine supports
E-003	Lakeside Cohort	Engagement	Student participation increased after learner-led conferences	Survey	ready	Priya	Continue pilot
E-004	North Valley Learning Network	equitable outcome	Participation gains were smaller for rural learners	Focus group	Needs Review	Alex	Target outreach
E-005	Cedar Grove Pilot	System Change	Leadership turnover disrupted implementation routines	Notes	Needs Review	Morgan	Mitigate risk
E-006	Lakeside Cohort	agency	Students had limited influence on pathway decisions	Interview	Draft		Revise design
E-007	Riverbend Partnership	Equitable Outcomes	Translation support improved family participation	Survey	Ready	Priya	Expand support
E-008	Riverbend Partnership	engagement	Attendance feedback was not yet available	Notes	Draft	Alex	Collect data

Begin by asking me to confirm that Excel, SharePoint or Microsoft Lists, and ChatGPT are open. Then output only the synthetic data as tab-separated values in one code block so I can paste it into Excel cell A1. Stop and wait for DONE.

What does the second part add?

The outputs give ChatGPT a fixed destination for the complete project.
The approved vocabulary defines the four impact categories that govern later work.
The synthetic records preserve inconsistent labels plus empty Owner cells for the quality audit.

Send the completed master prompt.

ChatGPT should ask you to confirm that your three tools are open. This response shows that the coaching protocol is following the requested sequence.

Reply DONE after confirming that Excel plus your selected list location are open.

ChatGPT should return the synthetic dataset as tab-separated values in one code block. Keep this conversation open because every later checkpoint builds on its Build Log.

Build the raw evidence table

Tab-separated data places each field into its own Excel column when pasted into cell A1. Formatting the range as a table adds filters while preserving every raw value.

Copy the complete tab-separated code block from ChatGPT.
Switch back to Excel.
Create a new blank workbook.
Select cell A1.
Paste the copied dataset.
Rename the worksheet Raw Evidence.
Select the complete pasted range.
Select Home from the Excel ribbon.

You should now see eight records arranged beneath eight descriptive headers. The empty Owner cells for E-002 plus E-006 stay empty.

Select Format as Table.
Choose a table style.
Confirm that the table has headers.
Select OK.

You should see a filter control in every table header. Those controls confirm that Excel recognizes the eight rows as a filterable table.

Save the workbook in your Documents folder as Strategic_Evaluation_Evidence.xlsx.

✔️ My workbook matches

Your Raw Evidence worksheet should contain eight data rows beneath the header row. Each column header should show a filter control.

ⓧ I'd like to double check the full code

Compare your worksheet with this complete tab-separated reference. The two empty Owner fields appear as blank spaces between tab-separated columns.

Evidence ID	Initiative	Impact Area	Finding	Evidence Type	Review Status	Owner	Decision Use
E-001	North Valley Learning Network	Agency	Educators report stronger ownership of lesson design	Survey	Ready	Morgan	Scale coaching
E-002	Cedar Grove Pilot	systems transformation	District teams adopted shared implementation routines	Interview	Draft		Refine supports
E-003	Lakeside Cohort	Engagement	Student participation increased after learner-led conferences	Survey	ready	Priya	Continue pilot
E-004	North Valley Learning Network	equitable outcome	Participation gains were smaller for rural learners	Focus group	Needs Review	Alex	Target outreach
E-005	Cedar Grove Pilot	System Change	Leadership turnover disrupted implementation routines	Notes	Needs Review	Morgan	Mitigate risk
E-006	Lakeside Cohort	agency	Students had limited influence on pathway decisions	Interview	Draft		Revise design
E-007	Riverbend Partnership	Equitable Outcomes	Translation support improved family participation	Survey	Ready	Priya	Expand support
E-008	Riverbend Partnership	engagement	Attendance feedback was not yet available	Notes	Draft	Alex	Collect data

Why do the mismatches stay?

The capitalization plus wording differences are intentional. The blank Owner cells also remain empty.

Preserving these raw problems lets you test whether the audit finds them without changing the source evidence.

Audit the saved workbook

A useful audit must inspect the saved workbook instead of relying on the dataset embedded in the original prompt. ChatGPT must also cite the affected Evidence IDs so you can verify every finding.

Return to the same ChatGPT coaching conversation.
Attach Strategic_Evaluation_Evidence.xlsx from your Documents folder.
Prepare the first audit by pasting this checkpoint prompt:

Checkpoint 1: Audit the raw workbook without changing it.

Analyze only the Raw Evidence sheet. For Impact Area, Review Status, and Owner:
1. List every distinct value and its count.
2. Identify capitalization, synonym, or completeness problems.
3. Cite the affected Evidence IDs.
4. Do not normalize values yet.
5. End with a Build Log containing Completed, Observed, and Next.

If you cannot inspect the workbook, say exactly what prevented the analysis instead of guessing.

What does this checkpoint protect?

The sheet restriction keeps the audit focused on the raw evidence.
The distinct-value counts expose category fragmentation without cleaning it.
The final instruction requires ChatGPT to report an inspection problem instead of inventing results.

Before you send the checkpoint, which raw fields do you expect the audit to flag?

Send the checkpoint prompt with the workbook attached.

The audit should flag inconsistent Impact Area values. It should distinguish Ready from ready.

It should also identify empty Owner cells for E-002 plus E-006. The response should finish with a Build Log containing Completed, Observed, plus Next.

Compare every distinct-value count with the Raw Evidence worksheet.
Confirm that each reported problem cites the affected Evidence IDs.

Before you reopen the workbook, do you expect Excel to preserve the table filters plus every raw inconsistency?

Close the saved workbook in Excel.
Reopen Strategic_Evaluation_Evidence.xlsx from Documents.
Select the Raw Evidence worksheet.
Confirm that eight data rows remain visible.
Open one column filter to confirm that the table remains filterable.
Return to ChatGPT to confirm that the audit findings still match the reopened workbook.

That foundation is locked in. Your raw evidence is preserved in Excel while the coaching conversation holds a human-checkable audit plus Build Log.

Workbook or audit does not match?

If all pasted data appears in one column, copy the original tab-separated code block from ChatGPT again. Paste it into cell A1 of a fresh worksheet.

If ChatGPT cannot inspect the workbook, confirm that the attached file is named Strategic_Evaluation_Evidence.xlsx. Send the checkpoint only after the attachment appears in the message.

Ask for help diagnosing the mismatch.

Your evidence workbook plus audit are ready. Next, you will import the raw records into the evidence register to see how inconsistent categories break operational grouping.

Import and Inspect the Raw Register

Your workbook audit is complete. ChatGPT has identified labels that could fragment the evidence register.

You will import the Excel workbook into Microsoft Lists. You will keep the raw values untouched so SharePoint can expose the operational impact.

In this step, get ready to:

Import Strategic_Evaluation_Evidence.xlsx into a list named Strategic Evaluation Evidence Hub.
Confirm that all eight imported items are editable without changing their raw values.
Group the register by Impact Area and report the observed split to ChatGPT.

Start Checkpoint 2 and import the workbook

The Checkpoint 2 prompt keeps your coach in a one-action loop. It also delays cleanup until you report what the grouping shows.

Switch back to the ChatGPT coaching conversation from earlier.
Paste this checkpoint prompt into the conversation:

Checkpoint 2: Coach the raw import without cleaning the data.

Give me one action at a time and wait for DONE or BLOCKED after each action. Use this verified path:
- In SharePoint: + New > List
- Or in Microsoft Lists: +New list
- Select From Excel
- Use Strategic_Evaluation_Evidence.xlsx
- Name the list Strategic Evaluation Evidence Hub
- Prefer the SharePoint site; use My lists if site saving is unavailable

After import, tell me to use Edit in grid view only to confirm the eight items are editable. Then have me group by Impact Area and inspect Review Status. Do not suggest a correction until I report how many groups I see.

How does this checkpoint protect the test?

The one-action loop keeps each Microsoft 365 change observable.
Replying DONE confirms that you completed the current action yourself.
Replying BLOCKED tells ChatGPT to diagnose the labels visible in your interface.
Keeping the imported values unchanged preserves the quality problem for inspection.

Send the checkpoint prompt to ChatGPT.
Complete the first action that ChatGPT gives you.
Reply DONE after completing each observable action.

The starting path depends on whether you are working from the SharePoint site or the Microsoft Lists app. Use the matching tab while following ChatGPT one action at a time.

SharePoint Site

Switch back to the SharePoint site from earlier.
Select + New from the site page.
Select List.
Select From Excel.
Choose Strategic_Evaluation_Evidence.xlsx from Documents.
Enter Strategic Evaluation Evidence Hub as the list name.
Finish the import into the current SharePoint site.

Microsoft Lists Fallback

Switch back to the Microsoft Lists app from earlier.
Select +New list.
Select From Excel.
Choose Strategic_Evaluation_Evidence.xlsx from Documents.
Enter Strategic Evaluation Evidence Hub as the list name.
Select a SharePoint site as the save location if one is available.
Select My lists if site saving is unavailable.
Finish the import.

Return to the ChatGPT conversation after the imported list opens.
Reply DONE to continue the coaching loop.

You should see Strategic Evaluation Evidence Hub with eight imported items. Each item should retain the columns and values from the Raw Evidence worksheet.

Workbook missing or import count wrong?

Return to Excel to confirm that Strategic_Evaluation_Evidence.xlsx remains saved in Documents.
Check that the Raw Evidence table still contains eight rows beneath its headers.
Stop the import if its preview contains fewer than eight records.

Help me diagnose why my Excel workbook is missing records during a Microsoft Lists import.

Confirm that every imported item is editable

Grid editing confirms that the imported rows are operational list items. This check must preserve every raw label so the grouping test remains valid.

Follow ChatGPT's next action until it asks you to check the imported items.
Select Edit in grid view from the list command bar.
Select the Impact Area cell for E-001.
Press Esc without changing the cell.
Select one cell in each remaining row to confirm that it accepts edit focus.
Keep all eight raw records unchanged.
Reply DONE after every row has passed the editability check.

All eight rows should accept edit focus. The list now holds editable items while the original workbook remains unchanged in Documents.

Cells remain read-only?

Confirm that you opened the completed Strategic Evaluation Evidence Hub list instead of the import preview.
Use BLOCKED to give ChatGPT the exact labels visible in your tenant.
Request edit access from the SharePoint site owner if the items remain read-only.

Help me work out why imported Microsoft Lists items are read-only.

Group the raw impact areas and report the split

Grouping turns category consistency into something you can observe. Similar meanings only support reliable grouping when their stored labels align.

Before you group the register, do you think the raw labels will combine into a small set of impact areas?

Follow ChatGPT's next action to leave grid editing mode.
Follow its next action to group the current list by Impact Area.
Keep every raw Impact Area value unchanged.
Count the group headings shown in the list.
Inspect the raw labels in the Review Status column.

✔️ I see eight groups

You should see these eight raw groups: Agency, systems transformation, Engagement, equitable outcome, System Change, agency, Equitable Outcomes, and engagement.

The Review Status column should include Ready, Draft, ready, and Needs Review.

ⓧ I see a different count

A different count usually means the current view has a filter or the import omitted records.

Remove any active record filter from the current view.
Count the visible items to confirm that all eight records are present.
Group the full list by Impact Area again.

Grouping still differs?

Help me compare my visible Impact Area groups with the eight unchanged source records.

Reply to ChatGPT with the group count plus the exact Impact Area labels you observed.
Include the exact Review Status labels in the same report.
Ask ChatGPT to compare your observation with its Checkpoint 1 audit.
Ask ChatGPT to add the comparison to the Build Log.

That's the quality problem proven. Your unchanged records split into eight impact-area groups because each raw label is stored separately.

Your raw register is imported and its category failure is documented in the Build Log. Next, you will review ChatGPT's proposed mapping before governing the register.

Govern the Evidence Taxonomy

You proved that the raw Impact Area values split eight records into eight groups. Small wording differences are hiding the portfolio patterns inside your evidence.

Now you will use ChatGPT to propose a mapping for human review. You will approve every value before Microsoft Lists enforces a controlled vocabulary.

In this step, get ready to:

Validate an AI mapping against the approved evidence taxonomy.
Add four governed Choice columns with approved values.
Enable version history for a controlled test edit.

Review the proposed mapping

An AI mapping stays provisional until a person checks every record. The next prompt limits ChatGPT to approved values while requiring a rationale for each proposal.

Switch back to the ChatGPT coaching conversation from the previous step.
Request a record-by-record mapping proposal by pasting this checkpoint prompt:

Checkpoint 3A: Propose a mapping for human approval.

Use only these approved Impact Area Standard values:
- systems transformation
- equitable outcomes
- agency
- engagement

For each Evidence ID, return:
- Raw Impact Area
- Proposed Impact Area Standard
- Proposed Review Status Standard using only Draft, Needs Review, or Ready
- Proposed Owner Role using only Data Operations, Evaluation, Learning and Impact, Initiative Lead, or Unassigned
- Proposed Data Quality using only Complete or Gap
- One-sentence rationale

Rules:
- Preserve the meaning of each raw record.
- Use Unassigned when Owner is blank.
- Use Gap for E-002, E-006, and E-008.
- Flag uncertainty rather than inventing information.
- Label the entire output PROPOSAL REQUIRING HUMAN APPROVAL.

What does this prompt enforce?

The approved Impact Area Standard values constrain ChatGPT to the four governed categories.
The permitted status values prevent capitalization variants such as ready from entering the governed field.
The owner rule maps blank owners to Unassigned without inventing a person.
The proposal label keeps every classification pending until you approve it.

Approved Mapping

E-001: agency, Ready, Data Operations, Complete.
E-002: systems transformation, Draft, Unassigned, Gap.
E-003: engagement, Ready, Evaluation, Complete.
E-004: equitable outcomes, Needs Review, Learning and Impact, Complete.
E-005: systems transformation, Needs Review, Data Operations, Complete.
E-006: agency, Draft, Unassigned, Gap.
E-007: equitable outcomes, Ready, Evaluation, Complete.
E-008: engagement, Draft, Learning and Impact, Gap.

Compare ChatGPT's proposal with the approved mapping one Evidence ID at a time.
Reply with the exact correction for every proposed value that differs.
Ask ChatGPT to regenerate any corrected rows before you approve the proposal.
Reply DONE after every proposed value matches the approved mapping.
Ask ChatGPT to record that the mapping passed human review.

The proposal now has a human approval trail in the coaching conversation. Its classifications are ready to become governed list values.

Create controlled Choice columns

Choice columns limit each governed field to a defined set of values. This preserves the imported evidence while adding a consistent layer for filtering.

Governed Column Settings

Impact Area Standard uses systems transformation, equitable outcomes, agency, and engagement.
Review Status Standard uses Draft, Needs Review, and Ready.
Owner Role uses Data Operations, Evaluation, Learning and Impact, Initiative Lead, and Unassigned.
Data Quality uses Complete or Gap.

Switch back to the Strategic Evaluation Evidence Hub list from the previous step.
Select + Add column from the list header.
Select Choice as the column type.
Enter Impact Area Standard in the Name field.
Replace the choice entries with the four approved impact values.
Turn off Can add values manually.
Select Save.

You will see an Impact Area Standard column in the list. Its menu now limits entries to the four approved impact areas.

Repeat the same Choice-column process for Review Status Standard with its three approved values.

You will see the new status column beside the imported fields. Manual values are disabled for this field.

Repeat the same Choice-column process for Owner Role with its five approved values.

You will see the new owner-role column. Its choices include Unassigned for records with no named owner.

Repeat the same Choice-column process for Data Quality with its two approved values.

All four governed column headers are now visible. Each column uses a fixed set of choices.

Select Edit in grid view from the list command bar.
Enter the human-approved values for all eight records using the approved mapping above.
Leave every imported raw field unchanged.
Finish grid editing after the governed values have saved.

The raw columns still show the original wording. The four governed columns now provide the consistent classifications used for analysis.

Before you group the standardized field, how many groups do you expect the eight records to form?

Use the Impact Area Standard column menu to group the list by that field.

What Should You See?

The agency group contains E-001 and E-006.
The engagement group contains E-003 and E-008.
The equitable outcomes group contains E-004 and E-007.
The systems transformation group contains E-002 and E-005.

That is the governance payoff. Eight fragmented raw groups now resolve into four portfolio groups with two records each.

Seeing More Than Four Groups?

Return to Edit in grid view.
Compare every governed value with the approved mapping above.
Check that you grouped by Impact Area Standard instead of the raw Impact Area column.

Ask for help with the mismatch: help me diagnose why my governed Microsoft List has more than four impact groups

Prove edits are traceable

Version history creates an audit trail whenever a list item changes. You will enable it before making a controlled edit to E-004.

Switch back to the ChatGPT coaching conversation.
Start the versioning test by pasting this checkpoint prompt:

Checkpoint 3B: Coach the governance test one action at a time.

Use this verified path:
Settings > List Settings > Versioning settings

Have me enable creation of a version each time an item is edited. Then have me change E-004 Decision Use from Target outreach to Target rural outreach, save it, open the item's ellipsis menu, and select Version history. Wait for me to confirm that a newer version appears. Finish with a Build Log and do not claim success before my confirmation.

What does this prompt protect?

ChatGPT gives one observable action at a time.
The versioning setting is enabled before the test edit happens.
The Build Log records success only after you see the newer version yourself.

Complete each action below only when ChatGPT asks for it.
Reply DONE in ChatGPT after each completed action.
Return to the evidence list from earlier.
Select Settings in the top-right corner.
Select List Settings.
Select Versioning settings.
Choose the option that creates a version each time an item is edited.
Save the versioning settings.

The list is now ready to record future item edits. The test change gives the history a clear before-and-after value to capture.

Return to the Strategic Evaluation Evidence Hub list.
Select E-004 to open its item form.
Replace Target outreach with Target rural outreach in Decision Use.
Select Save.

Before you open the history, do you expect E-004 to show an entry from before the Decision Use edit?

Select the ... ellipsis beside E-004.
Select Version history.
Select More if Version history is not immediately visible.

You will see a newer E-004 version created by the Decision Use edit. This proves the list can trace a governed record back through its changes.

Reply DONE to ChatGPT after you confirm the newer version.
Check that ChatGPT finishes Checkpoint 3B with a governance Build Log.

No Newer Version for E-004?

Confirm that the E-004 item shows Target rural outreach in Decision Use.
Return to Versioning settings.
Confirm that version creation for edited items is enabled.

Ask for help with the audit trail: help me diagnose why E-004 has no newer entry in Version history

Your evidence register now has human-approved categories with a traceable edit history. Next up, you will turn those governed fields into saved views for review work and portfolio learning.

Create and Check Portfolio Views

Your governed columns now give every record consistent values. The next challenge is turning those values into practical queues for review work and decisions.

Saved views in Microsoft Lists can preserve filters and grouping for repeat use. Each view answers a different portfolio question.

In this step, ChatGPT coaches you through three views. You confirm every visible Evidence ID before accepting its counts.

In this step, get ready to:

Create a Review Queue for evidence that still needs work.
Create Decision Ready and Portfolio Learning views.
Verify each saved view against the governed records.

Start the view coach

The checkpoint prompt gives ChatGPT the required logic for all three views. Its wait points keep you responsible for checking each result inside the evidence hub.

Switch back to the ChatGPT coaching conversation from earlier.
Add Checkpoint 4 by pasting this prompt:

Checkpoint 4: Coach three Microsoft Lists views one at a time.

For every view, state the required filter, grouping, sort, visible columns, and expected record count. Wait for my confirmed count before moving to the next view. Use View options > Save view as.

View 1: Review Queue
- Review Status Standard is Draft or Needs Review
- Show the governance columns
- Put Data Quality Gap records first
- Expected count: 5

View 2: Decision Ready
- Review Status Standard is Ready
- Data Quality is Complete
- Expected count: 3

View 3: Portfolio Learning
- No record filter
- Group by Impact Area Standard
- Expected count: 8 total, with four groups of two

If I cannot create a public view, tell me to save a personal view. Do not move on when my count differs. Ask me to report the visible Evidence IDs and diagnose the filter.

What does this prompt control?

The Review Queue isolates evidence that still needs attention.
The Decision Ready view limits the register to evidence that is ready and complete.
The Portfolio Learning view reveals patterns across the approved impact vocabulary.
The wait points prevent ChatGPT from accepting a view before you confirm its visible records.

Submit the checkpoint prompt.
Complete the first action that ChatGPT gives you.
Reply DONE only after completing the current action.
Reply BLOCKED if the labels in your list differ from the coaching instructions.

Before you count the first view, which governed records do you expect to remain in the review workflow?

Continue through the prompt-led actions until Review Queue is saved.
Count the visible items in Review Queue.
Report the visible Evidence IDs to ChatGPT.

You should see five items: E-002, E-006, E-008, E-004, and E-005. The three Gap records should appear first.

Does Review Queue show the wrong records?

Check that Review Status Standard includes both Draft and Needs Review.
Check that Data Quality places Gap records first.
Report the unexpected Evidence IDs to ChatGPT before changing the filter.

Ask ChatGPT to help me diagnose why Review Queue does not show the expected five Evidence IDs.

Create the decision and portfolio views

The next two views support different decisions. One isolates evidence ready for use. The other keeps the full portfolio visible for comparison.

Continue through the coaching actions until Decision Ready is saved.
Count the visible items in Decision Ready.
Report the visible Evidence IDs to ChatGPT.

You should see three items: E-001, E-003, and E-007.

Continue through the coaching actions until Portfolio Learning is saved.
Count the total visible items in Portfolio Learning.
Inspect the count shown for each Impact Area Standard group.
Report the visible Evidence IDs to ChatGPT.
Report the four group counts to ChatGPT.

You should see all eight records across four groups. Each group should contain two records.

Can't save a view or match the count?

Save a personal view if your permissions do not allow a public view.
Check that Portfolio Learning has no record filter.
Check that the grouping field is Impact Area Standard.

Ask ChatGPT to help me compare my saved view settings with the expected counts.

Verify every saved view

A saved view is trustworthy when reopening it preserves the expected records. This final check proves that each view can support repeatable portfolio work.

Before you switch among the views, which totals do you expect each saved view to preserve?

In Strategic Evaluation Evidence Hub, select View options.
Select Review Queue.
Confirm that five items are visible.
Select View options.
Select Decision Ready.
Confirm that three items are visible.

Good work. Your saved views now separate unfinished review work from decision-ready evidence with checked counts.

Select View options.
Select Portfolio Learning.
Confirm that eight items are visible.
Confirm that four groups contain two items each.
Switch back to the ChatGPT coaching conversation from earlier.
Report the final count for each saved view to ChatGPT.
Ask ChatGPT to finish Checkpoint 4 with the Build Log if it has not done so.
Confirm that the Build Log records the saved-view counts and visible Evidence IDs.

You have verified five records in Review Queue. You have also verified three records in Decision Ready and eight records in Portfolio Learning.

Your evidence hub now turns governed data into three checked portfolio workflows. Next, you will export the full portfolio and verify an AI-assisted decision brief against an Excel PivotTable.

Verify the Decision Brief

Your three saved views now turn the governed register into dependable review queues. You are ready to turn those governed records into a decision brief.

ChatGPT can accelerate synthesis. An Excel PivotTable gives you an independent count of the governed records.

Every accepted statement still needs support from Evidence IDs or a verified count. This step turns AI output into evidence a decision-maker can trace.

In this step, get ready to:

Export the governed Portfolio Learning view into an insight workbook.
Build a PivotTable that independently checks the evidence counts.
Verify each AI claim before finalizing the Decision Brief.

Export the governed workbook and build the PivotTable

The Portfolio Learning view contains the complete governed dataset. Exporting that view gives you a source table for checking every AI-generated count.

Switch back to the ChatGPT coaching conversation from earlier.
Start the workbook setup checkpoint by pasting this prompt:

Checkpoint 5A: Coach the analysis workbook setup one action at a time.

Use this verified workflow:
1. Open Portfolio Learning and make sure no items are selected.
2. Select Export > Export to Excel.
3. Open the result in Excel and save it as Strategic_Evaluation_Insights.xlsx.
4. Select the exported table and use Insert > PivotTable.
5. Choose New Worksheet.
6. Put Impact Area Standard in Rows, Review Status Standard in Columns, and Evidence ID in Values.
7. Confirm that Evidence ID is counted and the grand total is 8.

Wait for DONE or BLOCKED after every action. Do not interpret the evidence until I confirm the PivotTable total.

What does this checkpoint do?

The prompt makes ChatGPT pause after each observable action.
The export preserves the governed fields used by your saved views.
The PivotTable checks the record total before ChatGPT interprets the evidence.

Send the prompt to ChatGPT.
Reply DONE after each completed action.
Reply BLOCKED if the labels or results differ from the checkpoint.
Switch back to the Strategic Evaluation Evidence Hub from earlier.
Select View options.
Select Portfolio Learning.
Confirm that no list items are selected.
Select Export.

Select Export to Excel.
Open the exported result in Excel.
Save the workbook in Documents as Strategic_Evaluation_Insights.xlsx.

Keep the list authoritative

The exported table has a one-way connection to the original list. Refreshing can pull list changes into Excel.

Changes made in the exported Excel table do not return to the list. The Strategic Evaluation Evidence Hub remains your source of truth.

Select a cell inside the exported table.
Select the Insert tab.
Select PivotTable.
Choose New Worksheet.
Select OK.
Place Impact Area Standard in the Rows area.
Place Review Status Standard in the Columns area.
Place Evidence ID in the Values area.

Before you inspect the finished PivotTable, what grand total do you expect from the governed register?

Check that the Values area counts Evidence ID.
Inspect the PivotTable grand total.

You should see a grand total of 8. That total proves the complete governed register reached the analysis workbook.

Save Strategic_Evaluation_Insights.xlsx.
Reply DONE to confirm the total in ChatGPT.

PivotTable total does not equal 8?

Return to Portfolio Learning to confirm that the view still contains all eight items.
Check that no items were selected when you used Export to Excel.
Confirm that Evidence ID appears in the Values area as a count.

Help me diagnose why my PivotTable does not total eight records.

Analyze the governed workbook and verify each claim

The workbook-only prompt limits the analysis to evidence you can inspect. Its claim table exposes the calculation behind each statement.

Save Strategic_Evaluation_Insights.xlsx before uploading it.
Switch back to the same ChatGPT coaching conversation.
Attach Strategic_Evaluation_Insights.xlsx to the conversation.

Before you submit the analysis prompt, which counts do you think the PivotTable can verify directly?

Request a workbook-only claim analysis by pasting this prompt:

Checkpoint 5B: Analyze the governed workbook for human verification.

Use only the uploaded workbook. Do not use outside knowledge or infer causality.

Return a claim table with these columns:
- Claim ID
- Claim
- Calculation or filter used
- Supporting Evidence IDs
- Confidence limitation

Calculate and report:
1. Total evidence records.
2. Records where Review Status Standard is Ready and Data Quality is Complete.
3. Impact areas represented by those records.
4. Impact areas with no Ready record.
5. Records where Data Quality is Gap.
6. Records where Owner Role is Unassigned.

Then draft exactly three short sections named Pattern, Limitation, and Next action. Every sentence must cite Evidence IDs or a count from the claim table. End by instructing me to verify the output against the PivotTable and source table.

Why use a claim table?

Each claim includes its calculation plus its supporting Evidence IDs. That structure gives you a specific route for checking the AI output.

The confidence limitation also shows where the workbook supports a count without supporting a stronger conclusion.

Send the prompt to ChatGPT.
Compare the total evidence claim with the PivotTable grand total.
Filter the exported source table to Ready records with Complete data quality.
Record the matching Evidence IDs.
Filter the source table to Gap records.
Record the matching Evidence IDs.
Filter the source table to Unassigned owner roles.
Compare the represented impact areas with the PivotTable rows.

What should your checks confirm?

The governed workbook contains 8 evidence records.
The Ready plus Complete filter returns 3 records: E-001, E-003, plus E-007.
Those records represent agency, engagement, plus equitable outcomes.
The systems transformation impact area has no Ready record.
The Gap filter returns E-002, E-006, plus E-008.
The Unassigned filter returns E-002 plus E-006.

Switch back to Excel.
Add a worksheet named AI Validation Log.
Enter these headers in the first row: Claim ID, AI Claim, Supporting Evidence IDs, Human Check, plus Correction.
Copy each claim-table row into the corresponding validation-log row.
Mark each accepted claim as verified or corrected in Human Check.
Enter the evidence-supported revision in Correction for each corrected claim.
Save Strategic_Evaluation_Insights.xlsx.

You have now turned a plausible AI response into a checked evidence trail. Every accepted claim has a visible human decision beside it.

ChatGPT cannot inspect the workbook?

Confirm that the attached file is Strategic_Evaluation_Insights.xlsx.
Check that the exported table plus the PivotTable are saved inside the attached workbook.
Attach the saved workbook again before resubmitting Checkpoint 5B.

Help me confirm why ChatGPT cannot inspect my governed Excel workbook.

Red-team and finalize the brief

A red-team review looks for claims that sound stronger than the workbook allows. The safer revision still needs your approval before it reaches the final brief.

Switch back to the ChatGPT coaching conversation.
Start the red-team checkpoint by pasting this prompt:

Checkpoint 5C: Red-team the draft without adding new facts.

Review the draft Decision Brief I paste next. Identify:
- Any unsupported causal language
- Any count without supporting Evidence IDs
- Any recommendation that exceeds the evidence
- Any missing limitation

Return a two-column table named Issue and Safer Revision. If no issue exists, explain exactly which workbook evidence supports each sentence. Do not introduce facts that are absent from the workbook.

What does the red-team check test?

Unsupported causal language receives a safer evidence-based revision.
Every count must connect to supporting Evidence IDs.
Every recommendation must stay within the workbook evidence.
Every material limitation must remain visible to the decision-maker.

Send the red-team prompt.
Copy the draft Decision Brief from Checkpoint 5B.
Paste the draft into your next ChatGPT message.
Send the draft for review.
Compare each returned Issue with the source table.
Adopt each Safer Revision only when the workbook supports it.
Update affected Correction cells in the AI Validation Log.
Tell ChatGPT which revisions you accepted after human verification.

The red-team response is now another proposal in the audit trail. Your workbook remains the authority for the final wording.

Confirm that ChatGPT finishes Checkpoint 5C with Completed, Observed, plus Next entries in the Build Log.
Switch back to Excel.
Add a worksheet named Decision Brief.
Create only the headings Pattern, Limitation, plus Next action.
Place the human-verified text under each matching heading.
Save Strategic_Evaluation_Insights.xlsx.

Before the final check, do you expect every accepted sentence to trace back to an Evidence ID or a verified claim-table count?

Return to the AI Validation Log worksheet.
Confirm that every accepted claim is marked verified or corrected.
Return to the Decision Brief worksheet.
Confirm that it contains only Pattern, Limitation, plus Next action sections.
Trace every sentence to supporting Evidence IDs or a verified count.

You should find a complete verification status for every accepted claim. That is the hard part done: your decision brief is concise enough to use while remaining traceable to the evidence.

Found an unsupported sentence?

Remove any sentence that cannot be traced to the source table or a verified count.
Replace unsupported causal language with the evidence-supported wording from the red-team review.
Record the change in the matching Correction cell.

Help me revise an unsupported Decision Brief sentence using only my verified claims.

Secret mission

Prompt ChatGPT to Help You Add a Notification Rule

Your evidence hub already supports point-in-time decisions. In this mission, you will use ChatGPT to coach a modified-item notification rule in Microsoft Lists. You will trigger it with an E-002 edit and confirm the rule stays active.

Clean Up Your Resources

Your project adds no new usage charges under the Microsoft 365 and ChatGPT access you already use. Choose whether to keep the assets, pause the notification rule, or delete the project.

Resources you used:

ChatGPT coaching conversation containing the master prompt, checkpoint prompts, Build Logs, analysis, and Secret Mission completion record.
Local Excel workbook named Strategic_Evaluation_Evidence.xlsx in Documents.
The SharePoint-backed Strategic Evaluation Evidence Hub in Microsoft Lists with eight governed items, enabled versioning, three saved views, and the active modified-item notification rule.
Local Excel workbook named Strategic_Evaluation_Insights.xlsx with the exported table, PivotTable, AI Validation Log, and Decision Brief.
Retained copies of the synthetic workbooks in Library if ChatGPT saved the uploads there.

Keep everything running

No action is needed. Choose this if you want to keep demonstrating or improving the evidence workflow.

Your ChatGPT coaching conversation remains available with the complete prompt sequence and Build Logs.
Your Strategic Evaluation Evidence Hub retains its eight governed items.
Your three saved views remain available with their verified counts.
Your notification rule remains active for future item edits.
Both Excel workbooks remain available for auditing and analysis.
Any retained Library copies remain stored in ChatGPT.

Pause - I'll come back to this later

Turn off the active notification rule to pause monitoring. Your governed list, saved views, workbooks, and coaching conversation remain available for later.

Return to the Strategic Evaluation Evidence Hub list.
Select Integrate from the command bar.
Select Rules.
Open Manage rules.
Turn off the An item is modified rule.
Confirm that the rule remains listed with its status turned off.

Your monitoring is now paused while the complete evidence hub stays ready for your return.

Delete - I don't want to use this again

Removing these resources permanently clears your governed register and local project files. This option fits when you no longer need the evidence hub as a demonstration.

Delete the evidence register first:

Return to the Strategic Evaluation Evidence Hub list.
Select Settings.
Select List Settings.
Select the option on the List Settings page that deletes the list.
Confirm the deletion in Microsoft 365.

Deleting the list also removes its three saved views, version history, governed items, and modified-item notification rule.

Delete the local workbooks next:

Press the Windows key to open search.
Type File Explorer and press Enter.
Select Documents in the navigation pane.
Delete Strategic_Evaluation_Evidence.xlsx.
Use the File Explorer search box to locate Strategic_Evaluation_Insights.xlsx.
Delete the matching Strategic_Evaluation_Insights.xlsx file.

Delete the ChatGPT records last:

Return to the project conversation in ChatGPT.
Open the conversation options menu in chat history.
Select Delete.
Confirm the conversation deletion.

A workbook retained in Library remains stored after its conversation is deleted. Remove that separate copy to finish the cleanup.

Open Library in ChatGPT.
Locate any retained copies of Strategic_Evaluation_Evidence.xlsx.
Delete every matching retained copy.
Locate any retained copies of Strategic_Evaluation_Insights.xlsx.
Delete every matching retained copy.

If Library contains no matching files, the conversation deletion completes the ChatGPT cleanup.

Nice Work!

You did it! You built a governed evidence-to-decision workflow that turns synthetic findings into a human-verified decision brief.

You've learned how to:

Use a reusable master prompt to define ChatGPT's boundaries. Use checkpoint prompts to move through one observable action at a time. Require human confirmation before accepting each AI output.
Build a governed Microsoft Lists evidence register from eight inconsistent source records. Use controlled Choice columns to standardize four governance fields. Track edits through version history. Surface review work through three saved views.
Check ChatGPT's counts against an Excel PivotTable with a grand total of 8. Trace each accepted claim to its supporting Evidence IDs. Record every human check in the AI Validation Log before publishing the Decision Brief.
Secret Mission: Added a modified-item notification rule in Microsoft Lists. Tested the rule by editing E-002's Decision Use. Confirmed in Manage rules that the rule remains listed and turned on.

Ready to quiz yourself?