Research history

Inert source text, including its original formatting. It is not executed as page markup.

**Article type: design proposal. Peer review: not completed.** The contract below is an editorial method for testing, not an established intervention or a claim about what most charities do.

## Start with a decision

A report becomes useful when it can change a decision. Specify who will decide, what options they can choose, when they must choose and what they will do if the evidence changes. “We will learn about impact” leaves all four questions open.

A small organization might decide whether to continue a weekly service, change its delivery schedule or test a referral partnership. A donor might decide whether to renew an unrestricted grant. A community group might decide whether a promised benefit matches what participants actually value. The decision belongs to the people with the authority to act, but the reporting process must also hear those affected by it.

## Seven linked records

1. **Resources:** what money, time and equipment went in, in which period and currency.
2. **Activity:** what was actually delivered and the public evidence for that record.
3. **Reach:** who could access it and who was excluded. Avoid publishing personal records.
4. **Outcome:** what changed, including a unit, observation date and a defensible denominator.
5. **Counterfactual:** what might have happened without the activity and why that comparison is credible.
6. **Experience and harm:** what participants say mattered, how dissent can be raised safely and what unintended effects were observed.
7. **Revision:** the threshold, review date and named organizational role responsible for changing the decision.

Missing links stay missing. A count of services delivered is not automatically a causal effect. A testimonial can supply information about experience without estimating a population average. A model can clarify assumptions without becoming an observation.

## A worked, fictional example

Suppose a community learning program can fund one more term. Its team records sessions delivered and attendance, then learns that the existing timetable excludes caregivers. A simple continuation decision based only on attendance would miss that access question. The group therefore considers a changed schedule alongside continuation and closure.

Before choosing, it defines what it needs to know: whether the alternative time is acceptable, what participation would cost caregivers, whether learning outcomes differ and what other services may be displaced. A pilot could be appropriate, but a small pilot would not by itself establish a durable population effect.

No numerical benefit is invented here. The example demonstrates how the contract can expose an omitted decision dimension. Whether it improves real decisions is an open empirical question.

## Publish the revision rule first

A rule might say: review the program after one term; consult participants through a specified, accessible process; compare outcomes with the predeclared baseline; document harms and missing data; then record why the organization continued, adapted or stopped. The appropriate thresholds depend on context and should not be copied from this example as universal standards.

Changes to the rule belong in a dated log. Distinguish a justified amendment from an outcome selected after seeing the result. Publishing the original plan lets another contributor reconstruct that difference.

## Depth and values remain separate

The contract asks both where work acts and what different moral lenses consider important. A rights concern may create a constraint rather than an extra point in an average. Care may require attending to relationships that a cost model misses. Future generations may change the time horizon. These questions belong in the decision record, not behind a single hidden score.

## Use and test

The [Impact Measurement Toolkit](/tools/impact-measurement-toolkit/) can turn the mechanism into an observation plan. [Charity OS](/tools/charity-os/) can retain local project, budget and evidence records. The [Moral Frameworks desk](/programs/moral-frameworks-program/) can record disagreements, while the [Failure Lab](/programs/failure-lab/) can document what changed after an unexpected result.

A field test should compare decisions made with and without the contract, define what improvement means before analysis, track burden on staff and communities and publish failures as well as successes. No such test is claimed here. The contract earns its place only if evidence shows that its benefits justify its costs.

{
  "adapterVersion": "research-content-v1",
  "authorship": {
    "actorIds": [],
    "complete": false,
    "refs": [],
    "state": "unknown"
  },
  "hash": "d77cd44724fa64b2468ff24f67e36545f51f1b57859cf1ea867ec0f10a81cfe6",
  "id": "version:manuscript-the-feedback-contract-v1",
  "previousId": null,
  "recordId": "manuscript:the-feedback-contract",
  "sources": [
    {
      "selector": {
        "kind": "entry"
      },
      "sourceId": "source:current-journal-the-feedback-contract"
    }
  ],
  "subject": {
    "body": "\n**Article type: design proposal. Peer review: not completed.** The contract below is an editorial method for testing, not an established intervention or a claim about what most charities do.\n\n## Start with a decision\n\nA report becomes useful when it can change a decision. Specify who will decide, what options they can choose, when they must choose and what they will do if the evidence changes. “We will learn about impact” leaves all four questions open.\n\nA small organization might decide whether to continue a weekly service, change its delivery schedule or test a referral partnership. A donor might decide whether to renew an unrestricted grant. A community group might decide whether a promised benefit matches what participants actually value. The decision belongs to the people with the authority to act, but the reporting process must also hear those affected by it.\n\n## Seven linked records\n\n1. **Resources:** what money, time and equipment went in, in which period and currency.\n2. **Activity:** what was actually delivered and the public evidence for that record.\n3. **Reach:** who could access it and who was excluded. Avoid publishing personal records.\n4. **Outcome:** what changed, including a unit, observation date and a defensible denominator.\n5. **Counterfactual:** what might have happened without the activity and why that comparison is credible.\n6. **Experience and harm:** what participants say mattered, how dissent can be raised safely and what unintended effects were observed.\n7. **Revision:** the threshold, review date and named organizational role responsible for changing the decision.\n\nMissing links stay missing. A count of services delivered is not automatically a causal effect. A testimonial can supply information about experience without estimating a population average. A model can clarify assumptions without becoming an observation.\n\n## A worked, fictional example\n\nSuppose a community learning program can fund one more term. Its team records sessions delivered and attendance, then learns that the existing timetable excludes caregivers. A simple continuation decision based only on attendance would miss that access question. The group therefore considers a changed schedule alongside continuation and closure.\n\nBefore choosing, it defines what it needs to know: whether the alternative time is acceptable, what participation would cost caregivers, whether learning outcomes differ and what other services may be displaced. A pilot could be appropriate, but a small pilot would not by itself establish a durable population effect.\n\nNo numerical benefit is invented here. The example demonstrates how the contract can expose an omitted decision dimension. Whether it improves real decisions is an open empirical question.\n\n## Publish the revision rule first\n\nA rule might say: review the program after one term; consult participants through a specified, accessible process; compare outcomes with the predeclared baseline; document harms and missing data; then record why the organization continued, adapted or stopped. The appropriate thresholds depend on context and should not be copied from this example as universal standards.\n\nChanges to the rule belong in a dated log. Distinguish a justified amendment from an outcome selected after seeing the result. Publishing the original plan lets another contributor reconstruct that difference.\n\n## Depth and values remain separate\n\nThe contract asks both where work acts and what different moral lenses consider important. A rights concern may create a constraint rather than an extra point in an average. Care may require attending to relationships that a cost model misses. Future generations may change the time horizon. These questions belong in the decision record, not behind a single hidden score.\n\n## Use and test\n\nThe [Impact Measurement Toolkit](/tools/impact-measurement-toolkit/) can turn the mechanism into an observation plan. [Charity OS](/tools/charity-os/) can retain local project, budget and evidence records. The [Moral Frameworks desk](/programs/moral-frameworks-program/) can record disagreements, while the [Failure Lab](/programs/failure-lab/) can document what changed after an unexpected result.\n\nA field test should compare decisions made with and without the contract, define what improvement means before analysis, track burden on staff and communities and publish failures as well as successes. No such test is claimed here. The contract earns its place only if evidence shows that its benefits justify its costs.\n",
    "collection": "journal",
    "frontmatter": {
      "author": "Charitology editorial project (AI-assisted draft)",
      "date": "2026-10-25",
      "issue": 3,
      "sources": [],
      "summary": "A proposed reporting contract linking a giving decision to outcomes, uncertainty, affected communities and a public revision rule.",
      "title": "The feedback contract: designing a giving decision people can correct",
      "volume": 0
    },
    "type": "entry"
  },
  "subjectHash": "40b575ce4271462dbf2ca4e6abff3f5c5cf15d9a2bf1e39a7d5241aedd56427e",
  "version": 1
}