Editorial Integrity

Publication Validation Protocol.

Every claim published by Impact Lab — on the website, in research reports, in outgoing emails, in product descriptions — must be verified, sourced, or explicitly qualified as an estimate. No exceptions.

Version 1.0Effective March 10, 2026All channels
Website·Research reports·Outgoing emails·Public announcements·Product descriptions·Internal documents

Foundations

Three governing principles.

This protocol rests on three non-negotiable principles. They apply to every publication, every channel, every cycle.

01

Verifiability

Every factual claim must be traceable to its source.

No figure can be published without an identifiable, accessible, dated source — or an explicit qualification as an estimate.

02

Prudence

When in doubt, qualify or do not publish.

If data cannot be verified, it must be removed or presented as "estimate," "approximation," or "order of magnitude." Wording must match actual certainty.

03

Traceability

Every publication must be auditable.

The source of every key claim must be documented and accessible. Post-publication corrections must be logged and visible.

Data Tiers

Three-tier data classification.

Every piece of data must be classified before publication. The classification determines what can be published and how it must be framed.

Publication authorized

Verified Data

Peer-reviewed academic publications
Official statistics (WHO, World Bank, governments)
Reports from cited organizations (GiveWell, Rethink Priorities, ACE)
Directly measured internal metrics (page count, emails sent, Stripe revenue)
Mandatory qualification

Qualified Estimates

Cost-effectiveness model results
Extrapolations from existing data
Uncertainty ranges and confidence intervals
Forward projections and scenarios

Required phrasing

"Estimate based on…""According to [source], approximately…""Order of magnitude: … (range: X–Y)""Per our models, … (uncertainty: ±X%)"
Publication forbidden

Non-Verifiable Data

Fabricated or invented figures
Data with untraceable original source
Statistics copied without primary source verification
Model results with undocumented assumptions

Pre-Publication Gate

Compliance checklist.

Every publication must pass this checklist before release. Every item must be validated — no partial compliance.

1

Data & Figures

  • Every figure is sourced (identifiable, accessible, dated)
  • Estimates explicitly qualified as such
  • Uncertainty ranges indicated where applicable
  • Data less than 3 years old (or age disclosed)
  • Units and currencies clearly specified
2

Claims & Wording

  • No uncertain claim presented as certain
  • Causation distinguished from correlation
  • Comparisons made on equal basis (period, methodology, scope)
  • No promises or guarantees implied
  • Superlatives justified with source
3

Sources & References

  • Each source identified by name, author, and date
  • Primary sources preferred over secondary
  • Sources accessible to reader (URL or full reference)
  • Every cited source actually consulted first-hand
4

Outgoing Emails

  • No unsourced figures in email body
  • Claims about Impact Lab are factually accurate
  • No unconfirmed commitments or promises
  • Tone is factual and measured, not promotional
5

Product Descriptions

  • Description matches the actual delivered product
  • No unbuilt features announced as available
  • Prices and conditions accurate and current

Hard Constraints

Prohibited cases.

The following are categorically forbidden in any Impact Lab publication. Each category includes concrete examples to eliminate ambiguity.

Prohibited

Fabricated Figures

  • "Impact Lab analyzed over 1,000 sources" → if the real count is not verified in the system
  • "This program saves 10,000 lives per year" → if no source confirms this exact figure
  • "Our research shows a 50x ROI" → if the calculation is not documented and reproducible
Prohibited

Unverified Sources

  • Citing a GiveWell report without verifying the figure actually appears in it
  • "According to a 2024 study" with no link or full reference
  • Attributing a figure to an organization without checking their publication
Prohibited

Ambiguous Wording

  • "Impact Lab is recognized by GiveWell" → if GiveWell has not explicitly stated this
  • "Our recommendations are the most rigorous in the sector" → without documented comparison
  • "Thousands of donors trust us" → if the real donor count differs
Prohibited

Unconfirmed Promises

  • "Your donation will have 10x more impact" → unprovable promise
  • "We will publish a report every month" → if capacity is unconfirmed
  • "100% of your donation goes to research" → if allocation rules include other items
Prohibited

Unmarked Extrapolations

  • "This program costs $0.50 per DALY" → if this is a modeled estimate stated as fact
  • Correct: "Per our model, estimated cost is $0.30–$0.80 per DALY (based on [source])"

Procedures

Validation by publication type.

Each publication type follows its own validation sequence. Steps are mandatory and sequential.

01

Research Publications

Impact Index, Deep Dive Reports

  1. 1Draft with systematic source citation for every claim
  2. 2Full checklist pass (Data, Claims, Sources — sections 1–3)
  3. 3Cross-verification: each key figure checked against at least one independent source
  4. 4Red-team review: adversarial read to identify unsupported assertions
  5. 5Publish with methodology and sources accessible to readers
02

Outgoing Emails

Outreach, replies, newsletters

  1. 1Factual drafting — claims only, no promotional language
  2. 2Email checklist pass (section 4)
  3. 3Every cited figure verified against internal data
  4. 4Tone check: factual, measured, no exaggeration
03

Website Updates

Pages, components, displayed metrics

  1. 1Verify all numerical data displayed on the page
  2. 2Claims checklist pass (section 2)
  3. 3Consistency check: site figures match current internal data
04

Public Announcements

Forum posts, social media

  1. 1Factual draft with inline sources
  2. 2Full checklist pass (sections 1–3)
  3. 3Verify every assertion is supported by published site content
05

Product Descriptions

Stripe listings, checkout pages

  1. 1Verify exact match between description and actual deliverable
  2. 2Product checklist pass (section 5)

Required Language

Qualification vocabulary.

When data is not a verified fact, the following qualifiers are mandatory. Using the wrong certainty level is a protocol violation.

Certainty Level
Required Phrasing
Verified fact
Direct assertion + source citation
Modeled estimate
"Estimate based on [model]" or "Per our model…"
Approximation
"Approximately" or "Order of magnitude"
Range
"Between X and Y" with source of range
Projection
"Projection" or "Estimated scenario"
Opinion
"We estimate that…" or "Our analysis suggests…"
Dated data
"Data from [year]" — if >3 years: "potentially outdated data"

Error Response

Post-publication corrections.

If a factual error is identified after publication, the following sequence is mandatory. Speed of correction is a measure of institutional integrity.

1

Immediate correction

The incorrect data is corrected or removed from all channels where it appeared.

2

Correction notice

A visible notice is added: "Corrected on [date]: [description of change]."

3

Notification

If the error was communicated via email or announcement, a correction is sent to all recipients.

4

Log entry

The error and its correction are recorded in the Lab Log with full details.

5

Root cause analysis

The cause of the error is identified and the protocol is strengthened to prevent recurrence.

Pipeline Integration

This protocol is embedded in the research pipeline as a guardrail. The Red-Team stage (step 4) includes protocol compliance verification. The Publish stage (step 6) requires full checklist passage before any release. The protocol is revised each research cycle (quarterly).

This protocol is part of Impact Lab's governance architecture. It is public, auditable, and constitutive of our commitment to rigor and transparency.