Methodology

Remains records insecurity incidents in Nigeria — kidnappings and abductions, armed attacks, killings, and communal violence, including attacks involving armed groups such as ISWAP and Boko Haram.

We record incidents down to Local Government Area and community level across all 36 states and the FCT. This granularity is deliberate — state-level data obscures where the crisis actually falls.

We record the human cost of each incident: people killed, abducted, and injured. Where abduction cases remain active, we track days in captivity. Where those taken are released or rescued, the record says so.

We do not record political events, protests, foiled attacks with no victims, rumours, or incidents outside Nigeria. Our scope is defined and we do not expand it without stating so publicly.

Every entry is built from published news reports, and every entry names its sources — you can always see exactly where we got our information, including what each source claimed when they disagree.

Outlets carry different weight, judged on documented reporting practice rather than reputation:

Tier 1 — outlets whose confirmed reporting may stand alone: Premium Times, Daily Trust, HumAngle, Punch, TheCable, ICIR, Channels TV, ThisDay, FIJ, and international wires (BBC, Reuters, AP, Al Jazeera).

Tier 2 — established outlets: Guardian Nigeria, Vanguard, Tribune, Leadership, Blueprint, and PM News. Two independent Tier 2 outlets are required for an entry to count. An incident carried by a single Tier 2 outlet may appear in the record marked reported — awaiting corroboration: visible so that it is not invisible, but its figures enter no total on this site until a second, independent source confirms them.

Lead-only — aggregators and outlets with documented reliability concerns may surface a story but can never confirm one, however many carry it. Syndicated copies of one report count as one source, not several.

Institutional datasets (ACLED, UCDP and similar) are used only as benchmarks to check what we may have missed. They are never ingested, never republished, and never processed by our tools — their records are their own.

Every published entry carries a source grade, visible on its page: ⬛⬛⬛⬛ confirmed by a leading outlet · ⬛⬛⬛⬜ two independent outlets · ⬛⬛⬜⬜ single confirming outlet, reported but not counted · ⬛⬜⬜⬜ surfaced but not corroborated. Only the first two grades enter the headline figures.

No incident enters the record without human review. This is non-negotiable. AI assists with reading, screening, and drafting — it does not publish, and it cannot: nothing reaches the public record except through a reviewer’s decision.

1. Collection. Reports are gathered continuously from outlet feeds and targeted searches. The full text of every article we assess is archived at that moment — links die; evidence should not.

2. Screening. Each article is read in full and screened against our admission rules — foiled attacks without victims, statistics round-ups, rumours, and follow-ups carrying no new facts are set aside, with the reason recorded. What we reject is audited, because a record that publishes a floor must know what it is missing.

3. Matching. Every candidate is compared against existing entries using the details that identify an event — named people, the specific community or road, the perpetrator description, dates — so the same event reported by many outlets becomes one entry with many sources, and follow-up coverage becomes an update to the entry it belongs to, never a second count.

4. Human review. A reviewer checks every candidate before anything is published: sources, location, figures, and a set of editorial red-line checks — on ethnic attribution, on naming individuals, on government-only sourcing — that must each be answered before publishing is possible. The conservative-figure rule applies wherever sources disagree.

These are not guidelines. They are hard limits that do not bend under any circumstances — not under editorial pressure, funding relationships, government requests, or time constraints. Every person on this platform is named, numbered, or recorded. They deserve no less than this.

Data integrity

01We never inflate figures. When sources disagree on casualty numbers, we publish the lower confirmed figure and show the higher claim beside it with its source. We never average conflicting figures. We never round up. And an absent figure is recorded as not established — never as zero, and never as a guess.

02We never count the unverified. A report that cannot be confirmed is held or set aside — it is never counted, never graded, and never presented alongside verified entries without clear separation. No exceptions.

03We never publish the same incident twice. Every new record is checked against existing entries by the details that identify an event — named people, place, perpetrator, date. Potential matches are held for human review before publication. A double-counted incident is a false number, and false numbers betray every person in the record.

04We never name an individual without documented sourcing. Every named person on this platform is named only where their identity is confirmed by a Tier 1 or Tier 2 source — and naming a person who may still be in captivity requires an explicit editorial decision that the naming is safe. Getting a name wrong in a memorial record is an irreversible harm.

Attribution and framing

05We never use ethnic group names as collective perpetrator labels. Public entries use neutral descriptions — “armed group”, “armed herders”. Where attribution is contested, we say so. Where it is unconfirmed, it does not appear.

06We never treat government attribution as independently verified. When an entry rests solely on government statements — regardless of how many outlets carried them — the entry says so. Government attribution is a data point, not a confirmation.

07We never state causation without direct evidential grounding. We record what happened. Political, climatic, or structural explanations of conflict are matters of ongoing research; we do not encode contested frameworks into individual incident records.

08We never present analysis as documented fact. Where Remains offers context or trend analysis, it is clearly labelled as analysis. The line between what happened and what we think it means is always visible.

Records and independence

09We never delete a record silently. If a record is found to be false, it is marked as retracted with a full public explanation — not quietly erased. The history of what we published and changed remains visible.

10We never accept editorial influence from funders or partners. No funding relationship, data partnership, or commercial arrangement gives any external party influence over what is published, classified, corrected, or removed. Any relationship that seeks such influence ends immediately.

These ten limits are the floor, not the ceiling, of our editorial standards. If our practice ever falls short of them, our corrections record will say so.

Uncertainty is not a failure. It is an honest feature of conflict documentation.

Conservative figures always. Published figures are “at least” numbers — the most conservative count the evidence supports, with higher claims shown beside them, unverified and uncounted.

Absence is not zero. When no casualty figure was established, the entry says “not established”. A zero on this platform means a confirmed zero. This is a deliberate position in a live methodological disagreement: some conflict datasets record an unknown death toll as zero; casualty-recording standards keep “unknown” as its own state, distinct from any number. We follow the latter — a blank here is information, not a gap waiting to be filled with a 0.

Location honesty. When we know only the state, the map says so — markers without confirmed coordinates are placed at the centre of their state and labelled approximate. We do not invent precision we do not have.

We acknowledge openly that our record is incomplete. Nigeria’s insecurity crisis is systematically underreported, particularly in areas with limited press access, and most of all for private cases — kidnappings settled quietly between families and abductors without police involvement, media coverage, or any institutional record. The incidents on this platform represent what has been documented — not the full scale of what has occurred. This gap is itself part of the story we are telling. If you know of an incident that is not here, tell us.

We make mistakes. What matters is what happens next. Corrections are recorded on the entry itself — what was wrong, what changed, and when — and no record is deleted, only updated, with its history preserved. If you believe an entry is wrong — the figures, the location, the classification — write to corrections@remains.ng. Every request is reviewed by a person. We will respond.

Remains is an independent platform. We are not affiliated with any government, political party, armed group, or commercial interest. Our editorial decisions are made by our team alone.

[Legal structure — to be stated on confirmation. Funding list published from the first grant.]