Our Methodology
How Pledge Tracker Rates Campaign Promises
An initiative of the African Development Documentation Initiative (AfDDI)
Version 1.0 · Published January 2025 Last updated 22 May 2026
Why this document exists
Pledge Tracker aims to address a specific gap in Nigerian democracy: the lack of a sustained, evidence-based public record of elected executives’ promises, their delivery, or lack thereof. A rating system that cannot account for its conclusions amounts to merely another opinion. This document outlines our methodology in detail. It is intended for citizens, journalists, researchers, and the officials whose promises we monitor, so all can understand the criteria by which we evaluate promises as Kept, Compromised, Not Done, Stalled, In the Works, or Not Yet Rated.
We adapt our methodology from two internationally recognised promise-tracking institutions: the Pulitzer Prize–winning Obameter developed by PolitiFact at the Poynter Institute, and the Trudeau Meter developed by the Polimeter Project at the University of Toronto. Both have tracked thousands of executive promises over more than a decade. Our adaptation respects the core discipline they pioneered — verifiable outcomes over stated intentions — while accounting for the specific evidentiary realities of Nigerian governance.
What we track, and what we do not
Pledge Tracker monitors explicit, identifiable, and verifiable campaign promises made by elected executives at national and sub-national levels in Nigeria. Our flagship project, Tinubumeter, tracks 250 promises made by President Bola Ahmed Tinubu during his 2022–2023 campaign, drawing on his Action Plan for a Better Nigeria (October 2022) and major campaign speeches delivered before the February 2023 election.
A statement qualifies as a tracked promise when it meets all three of the following tests. It must be specific— describing an identifiable action, target, or outcome rather than a general aspiration. It must be attributable— sourced to a public document, recorded speech, or verified campaign communication. And it must be verifiable in principle— meaning that evidence either for or against its fulfilment could exist and be assessed.
We do not include aspirational rhetoric (e.g., “we will make Nigeria great again”), values statements (e.g., “we believe in justice”), or commitments dependent entirely on factors outside the executive’s mandate (such as promises requiring constitutional amendment by the National Assembly, where the executive lacks both constitutional authority and a legislative majority). When a promise is partially within the executive’s control, we track it and explain any constraints in the rating.
We also do not track promises made after the executive takes office. Pledge Tracker focuses specifically on campaign promises—commitments made to citizens in exchange for a mandate to govern. Policy announcements made during office hours are separate and including them could diminish the clarity of the evaluative tool we aim to provide.
Our six rating categories are defined precisely.
Every promise on Pledge Tracker is assigned one of six ratings. The definitions below are the standards by which our editorial team must justify each rating and by which any reader can hold us accountable.
Kept. The promise has been substantially fulfilled as stated. For promises with a numerical target, that target has been met or closely approached. For action-based promises, the action has been completed in a form consistent with the original statement. Evidence must show outcomes, not just effort. A signed policy without measurable results is not Kept; it is, depending on details, In the Works or Compromise. This rating is distinguished from Compromise by full or substantial completion as promised, and from In the Works by the existence of completed outcomes.
Compromise. The government has taken meaningful action toward the promise, but the result is materially narrower in scope, slower in pace, or different in form from what was promised. This rating signals genuine partial fulfilment—more than effort but less than full delivery. Differentiate Compromise from Kept, which requires substantial completion, and from In the Works, where active implementation is ongoing but not completed. Compromise is not used if the government just does less and claims success.
Not Done. No meaningful action has been taken toward the promise within the period being assessed, or any action is too minimal to distinguish from inaction. Differentiate Not Done from Stalled, which involves interrupted progress; from In the Works, where active implementation exists; and from Not Yet Rated, for which evidence is insufficient. Not Done reflects the present state and does not close future action.
Stalled. Implementation began but halted, with no documented progress over a sustained period (twelve months is our standard threshold, adjustable as needed). ‘Stalled’ differs from ‘Not Done’ by the process’s initiation, and from ‘In the Works’ by the absence of ongoing progress. Unlike ‘Compromise’, ‘Stalled’ signals suspension rather than partial fulfilment or a narrower outcome.
In the Works. Documented, active implementation is underway, but the promise has not yet been fulfilled. There must be evidence that work is actually occurring — not just an announcement. In the Works differs from Kept and Compromise in lacking completion, and from Stalled by evidence of ongoing progress. Implementation includes a chain of actions (e.g., appropriation followed by execution), not just pronouncements.
Not Yet Rated. Available evidence is insufficient to support any of the five substantive ratings with the confidence Pledge Tracker requires for publication. This rating is not a process failure but indicates a lack of data. Distinguish Not Yet Rated from Not Done (active evidence of inaction) and Stalled (evidence of halted progress). Promises under this rating are re-examined each quarter as evidence develops.
What counts as evidence, and how we weigh it
Pledge Tracker rates promises on the basis of verifiable outcomes, not on the executive’s stated intentions, effort, or rhetoric. This is the single most important methodological commitment we make, and it shapes how we weigh every source.
We organise sources into three tiers. Ratings must rely mainly on evidence from Tiers 1 and 2.
Tier 1 — Primary sources. Federal budget documents and Appropriation Acts. National Assembly records, Hansards, and committee reports. Government gazettes, signed legislation, and statutory instruments. Statistical releases from the National Bureau of Statistics, the Central Bank of Nigeria, the Nigerian National Petroleum Company, the Nigerian Communications Commission, and other official agencies. Reports from the African Development Bank, the World Bank, and the International Monetary Fund in their Nigeria-specific work. Court rulings from the Supreme Court, the Court of Appeal, and the Federal High Court.
Tier 2 — Reliable secondary sources. Established Nigerian newspapers and broadcasters with documented editorial standards: Premium Times, The Cable, Daily Trust, Punch, BusinessDay, Channels Television, Arise News. Established international wires and outlets with Nigeria desks: Reuters, the Associated Press, Bloomberg, the BBC, and the Financial Times. Specialist Nigerian data and analysis platforms: Dataphyte, Stears, BudgIT’s published reports, and SBM Intelligence. Peer-reviewed academic work on Nigerian governance and development.
Tier 3 — Sources used with explicit caution. Government press releases and official social media accounts are recorded as the government’s claim but never as standalone evidence of the outcome. Partisan publications and political-party communications. Single-source social media reports. Opinion columns and editorials are used for context only. Specialist sources outside our established library are evaluated on a case-by-case basis.
When our editorial team encounters a conflict between sources, we resolve it according to a documented hierarchy. Tier 1 evidence prevails over Tier 2 unless the Tier 1 source has been credibly challenged. Independent corroboration across two or more Tier 2 sources outweighs an uncorroborated Tier 1 claim where the Tier 1 source is a government self-report rather than statistical or budgetary documentation. Where Tier 2 sources disagree, we report the disagreement transparently and reflect the uncertainty in our confidence rating rather than picking a side.
A claim made only by a government spokesperson, only by a partisan opponent, or only by a single source of any tier is, by definition, an unverified claim. Pledge Tracker treats unverified claims as exactly that.
How a promise becomes a rating: our editorial workflow
Every rating on Pledge Tracker passes through five sequential stages before publication. The stages exist to separate research from judgment from verification from publication, so that no single act — and no single tool, human or AI — can produce a published rating on its own.
The first stage is monitoring. We maintain rolling monitoring of all 250 tracked promises through a combination of automated news surveillance and scheduled monthly sweeps across primary and secondary sources. Monitoring identifies which promises have material new developments warranting assessment and which can continue at their current rating until the next quarterly review.
The second stage is dossier construction. When a promise is selected for assessment — either because monitoring surfaced new evidence or because it has reached its scheduled quarterly review — a research analyst produces a structured evidence dossier. The dossier collects all relevant sources from the review period, organises evidence into actions taken, measurable outcomes, non-actions, and contradictions, and proposes a draft rating with reasoning. We use AI tools to accelerate retrieval and structuring; we do not use AI to make rating decisions.
The third stage is verification. Every source cited in a dossier is independently verified by a human reviewer before the dossier proceeds to editorial judgment. URLs are opened, dates confirmed, and quotes checked against the original document. Citations that cannot be verified are removed, and the dossier is returned for further research. This stage exists specifically to prevent the credibility risks introduced by AI-assisted research, including fabricated citations and misattributed quotations, which we treat as a known and serious hazard.
The fourth stage is editorial judgment. A managing editor reviews the verified dossier, applies the rating definitions above, writes the public-facing summary, and assigns a confidence level (High, Medium, or Low) that reflects the strength of the evidence base. Ratings drawn from converging Tier 1 evidence receive High confidence. Ratings drawn from contested or partial evidence receive Medium or Low confidence, with the reasoning made explicit in the published rating.
The fifth stage is publication and revisability. Ratings are published with full source citations, publication date, and the version of methodology under which they were issued. Every rating is open to revision when new evidence emerges; revisions are dated and the prior rating preserved in the public record, so that readers can see how our judgments have changed and why. We do not delete or quietly amend ratings. Transparency about our own reasoning is part of the standard we apply.
How we handle difficult cases
Some promises raise editorial questions that no general rule resolves cleanly. We explicitly address the recurring categories of difficulty.
Promises with passed deadlines. When a promise specifies a date that has passed without fulfilment, we rate against the state of affairs as of the deadline. A promise that called for a target by 2025 and was not met by the end of 2025 is rated Not Done or Compromise as of that point, and the rating is not adjusted upward by subsequent partial action, though continuing action is recorded in the rating’s update history.
Promises affected by baseline methodology changes. Several Tinubu promises reference baselines (such as the 2021 World Bank GDP figures or pre-2023 unemployment statistics) whose underlying methodologies have since changed. We rate against the baseline defined by the promise itself, while transparently noting the methodological complication. Where reasonable analysts could rate differently because of a baseline question, we say so.
Promises dependent on legislation. Many promises require the National Assembly to act rather than the executive alone. We rate the executive’s pursuit of the promise — whether it has tabled the necessary bills, used its political capital, included relevant items in budget appropriations — separately from the legislative outcome. A promise whose legislative path has been blocked despite serious executive effort is rated In the Works or Stalled, depending on circumstances, not “Not Done.”
Politically sensitive promises. Promises touching ethnic, regional, religious, or security-sensitive matters receive the same evidentiary standard as any other promise, applied with additional editorial care. The neutrality of Pledge Tracker’s voice does not waver based on the political weight of the topic. Where evidence is genuinely contested, we report the contest rather than artificially resolving it.
Promises with sub-national dimensions. Where a federal promise depends on state-level implementation (housing, education, health), we evaluate federal-level action and outcomes nationally and note subnational variation in the rating. Future state-level deployments of Pledge Tracker will address sub-national promises directly.
Promises whose fulfilment is contested by independent measurement. When the government claims a promise has been kept and independent sources disagree, we rate based on independent measurement. The government’s claim is recorded in the rating; it does not determine the rating.
Our use of artificial intelligence
Pledge Tracker uses AI tools in our research workflow, and we believe it is important to be explicit about how we do so. AI accelerates evidence retrieval, helps structure complex multi-source information, and enables a small editorial team to maintain rigorous monitoring across 250 promises, a task that would otherwise require a substantially larger staff. AI does not — and will not — make rating decisions.
We treat AI-assisted research as a category of evidence that must be verified more carefully, not less. Language models can fabricate URLs, misattribute quotations, and confidently state facts that are not true, particularly when the subject matter is underrepresented in their training data, as Nigerian sources frequently are. Our verification stage exists specifically to catch these failures before they reach publication. Every citation that appears on Pledge Tracker has been opened and read by a human editor.
We will update this section as the technology evolves and as our use of it changes. Readers who wish to inspect our AI workflow in detail are welcome to contact us; the prompts and verification protocols we use are not secret.
Our independence
The African Development Documentation Initiative is a non-profit organisation registered in the Netherlands. AfDDI accepts funding from foundations, individual donors, and grant-making institutions whose missions align with independent governance research and citizen accountability. AfDDI does not accept funding from political parties, candidates, sitting officeholders, government agencies of the countries we track, or organisations whose primary purpose is partisan political activity. Funders have no editorial input into our ratings. All Pledge Tracker funders are publicly disclosed on our funders page.
Our staff and volunteers are required to disclose any direct personal or financial relationship with the officials whose promises we track. Where a conflict exists, the team member does not participate in assessing the affected promise.
Errors, corrections, and the public record
Pledge Tracker will make mistakes. When we do, we will correct them visibly. Errors of fact are corrected in the affected rating, dated, and noted in a public corrections log. Errors of judgment — where a rating was defensible at the time, but new evidence persuades us a different rating better fits the standard — are revised through the normal update process, with the prior rating preserved in the rating’s history. We do not silently amend. Readers who believe a rating is in error are invited to contact us with specific evidence to support their claim.
Methodology versioning
This document is version 1.0. We will revise it as we learn from real editorial work, and each revision will be dated and archived. Previous versions remain accessible so that any historical rating can be evaluated against the methodology in force at the time it was issued. We commit to publicly announcing substantive revisions before they take effect, along with the reasoning.
Contact
For methodology questions, correction requests, or research partnerships:
Pledge Tracker is an initiative of the African Development Documentation Initiative – AfDDI in Collaboration with Townhall Radio.