A nonprofit report cannot explain where its impact numbers came from

The problem

Reporting staff need a clear evidence trail from programme totals to public statements, while activity counts can be mistaken for outcomes.

Agent job

Reconcile denominators and periods; map each draft claim to a source cell; distinguish activity, outcome and unsupported causal language; flag gaps without inventing results.

Expected output

Claim-to-evidence workbook, caveat list and editable short impact-report draft.

Why an agent / simpler alternative

Assessment

The agent can trace narrative claims to inconsistent tables and flag shifts in definitions. Spreadsheet formulas should handle stable arithmetic. A programme lead must judge valid outcome definitions; a qualified evaluator is needed for causal impact analysis, which is excluded.

Setup requirements

Items

1. Aggregate, privacy-reviewed programme tables with reporting periods

2. Written definitions of participants, activities and outcomes

3. Draft claims and a named programme reviewer

4. Agreed suppression rules for small groups and a source-cell naming convention

Inputs and access

1. Name

   Deidentified aggregate programme tables

   Source or owner

   Prospective consenting customer or specified official publisher

   Sensitivity

   Use synthetic, public or appropriately deidentified material; confirm actual sensitivity before work

   Permission needed

   Owner authorizes scoped access and processing; verify rights and retention before receiving files

2. Name

   Organisation-approved outcome definitions

   Source or owner

   Prospective consenting customer or specified official publisher

   Sensitivity

   Use synthetic, public or appropriately deidentified material; confirm actual sensitivity before work

   Permission needed

   Owner authorizes scoped access and processing; verify rights and retention before receiving files

3. Name

   Draft narrative and reporting period

   Source or owner

   Prospective consenting customer or specified official publisher

   Sensitivity

   Use synthetic, public or appropriately deidentified material; confirm actual sensitivity before work

   Permission needed

   Owner authorizes scoped access and processing; verify rights and retention before receiving files

Proposed steps

1. 1

   Confirm reporting period, approved definitions and privacy limits; reject participant-level records.

   Expected output: Data-definition and privacy manifest

   Approval gate: Programme owner approves aggregate scope and small-cell handling.

2. 2

   Reconcile table totals, duplicate cohort labels and inconsistent denominators.

   Expected output: Reconciliation workbook

   Approval gate: Unexplained differences stay visible; do not average them away.

3. 3

   Inventory every numerical or outcome claim in the draft and map it to specific source cells.

   Expected output: Claim-to-cell matrix

   Approval gate: Claims without evidence remain unsupported.

4. 4

   Classify each claim as activity, measured outcome or unsupported causal interpretation.

   Expected output: Editorial review queue

   Approval gate: Programme expert decides contested definitions; no causal inference invented.

5. 5

   Draft a short report using approved figures with limitations and reporting-period labels.

   Expected output: Editable report draft

   Approval gate: Do not add donor promises or grant eligibility advice.

6. 6

   Have the programme lead verify all claims and inspect the report for small-group disclosure.

   Expected output: Approved evidence pack and unresolved gaps

   Approval gate: Human approves recipients and publication; no donor outreach or submission.

Synthetic output illustration — not an executed result

Synthetic

Yes

Executed

No

Text

SYNTHETIC: Draft claim: “120 people improved their skills.” Available evidence: 120 session attendances, 78 unique participants, no outcome assessment. Status: unsupported outcome claim. Proposed factual wording: “120 attendances recorded.” All figures are invented.

Human approval boundaries

1. Programme lead confirms definitions and numbers; privacy owner approves data minimization; human approves publication. No grant submission or donor outreach.

2. Human service provider reviews quality, scope and quoted fee; customer approves contract, delivery recipients and payment terms. Separately approve purchases, credentials, publication and external sends. No commitments or payments are made in this research.

Failure modes

1. Duplicate participants inflate counts

2. Different definitions make periods incomparable

3. Aggregate tables may still expose small groups

4. Correlation may be presented as causation

Recovery and troubleshooting

Items

1. If totals differ, return to the named data owner with the exact rows and definitions; do not choose the most favorable figure.

2. If small cells risk disclosure, remove or aggregate them according to the owner’s approved rule.

3. If outcome evidence does not exist, deliver an activity report and a data-gap list instead of an impact claim.

Unexecuted validation plan

Sample setup

Use synthetic programme tables with a duplicated cohort, changed denominator and unsupported causal claim.

Acceptance checks

1. All numerical claims reconcile to named cells

2. No participant-level personal data

3. Incomparable periods and unsupported causation explicitly flagged

Stop conditions

1. Stop if data authority or rights are unclear.

2. Stop if critical seeded failures are missed; do not claim readiness.

3. Stop if manual correction time exceeds the pilot effort budget without customer approval.

Executed

No

Costs and effort — unknown or hypothetical

Status

unknown_or_estimate

Currency

CAD

Items

1. Agent subscription/API usage: not quoted; verify current plan and commercial data terms.

2. Human intake, rights review, QA and revisions: unknown until timed pilot.

3. Storage, hosting, integration and payment-processing costs: excluded unless separately agreed.

4. Customer acquisition and support time: not measured.

Caveat

No margin or income estimate. Price hypotheses are editorial offer tests, not observed market rates; taxes, scope and effort require a real quote.

Proposed service concept — earnings not verified or guaranteed

Buyer

Small nonprofit programme or communications lead

Deliverable

Claim-to-evidence workbook, caveat list and editable short impact-report draft.

Charging basis

Fixed-fee evidence-pack preparation; optional repeat reporting separately scoped.

Acquisition channel

Permission-based discovery with existing professional contacts, referrals or an opt-in sample page; none executed.

Assumptions

1. Customer has this unresolved problem and values a reviewed deliverable above existing built-in alternatives.

2. A consenting pilot confirms data access, achievable quality and economically tolerable review time.

3. Provider can deliver within a written scope and supports the result.

Revenue guarantee

No

Canadian adaptation — proposed

Proposed Canadian pilot: use the nonprofit’s own reporting-year definitions and approved English/French programme labels; keep Canadian programme cohorts separate from other regions when definitions differ. Confirm aggregate-data handling and small-group suppression with its privacy owner. No charity reporting, funding eligibility or legal-compliance certification.

Limits

1. Original proposed plan; untested.

2. Problem and capability references are evidence of context only, not demonstrated end-to-end success, customer demand, willingness to pay or profitability.

3. Compatibility, workload, delivery effort and availability must be verified in a real authorized pilot.

4. Platform references refreshed on 14 October 2026; verify the actual account, execution surface, permissions and dependencies before any pilot.

Candidate platforms — compatibility untested

1. Name

   Claude (Cowork capabilities)

   Required capabilities

   1. Read supplied files and produce editable documents and spreadsheets.

   Capability source urls

   1. https://support.claude.com/en/articles/13345190-get-started-with-claude-cowork

   2. https://support.claude.com/en/articles/16761823-claude-cowork-and-chat-are-one-claude

   Workflow compatibility

   Unknown; verify in the proposed first test.

   Historical candidate name

   Claude Cowork

   Selection basis

   Editorial candidate from general capabilities; exact end-to-end fit untested.