Methodology

Purpose

Vigil's entire value is credibility. It is a context platform, not a news service: each conflict module explains what happened, where, who is involved, why it matters and what to watch, and is reviewed on its stated cadence. This page is the rulebook that every record, grade, date and map on the site is built to. It is written to be held against the product, and it says, in each section, what Vigil deliberately does not do as well as what it does.

The vocabulary the rules are written in — proxy war, de facto authority, grey-zone pressure, state capacity, observable — is defined separately, with the common misreadings noted. Read the glossary →

1. The analytical model

Vigil keeps three things apart that reporting routinely collapses into one sentence: the event — that something happened; the attribution — who did it, or who says who did it; and the assessment — what Vigil judges the evidence to show and what it means for the system. A real-world event can be established without its attribution, its intent or its strategic significance being established, and each of the three is separately dated, separately sourced and separately graded.

What a grade is attached to

Vigil grades the claim, not the speaker. A government's statement is not high-confidence because a government made it, and a claim by a party to a conflict is not low-confidence because that party made it. Who said something is part of the evidence; it is not the grade. Where a statement is the fact — a declaration, a formal position, a legal assertion — the statement itself is what is recorded, attributed to its author.

An event, a claim about it, and an assessment of it are three different records. That something happened, that a party says a particular actor did it, and that Vigil judges the evidence one way, are separately dated, separately sourced and separately graded. Collapsing them is the most common way a conflict record becomes wrong: the event survives, the attribution turns out to be contested, and a single sentence carrying both takes the attribution down with it.

A confirmed physical event can coexist with a disputed attribution, and Vigil publishes both states at once rather than waiting for one to resolve or picking a side. An installation was struck and stopped operating; who struck it is contested between credible accounts; both facts are published, the dispute is described, and Vigil adjudicates nothing it cannot.

Absence of verification is not confirmation of anything.Where an inspection regime, a monitoring mission or an observation channel has stopped, what follows is that the state of the thing is unknown from that date — not that it continued, and not that it stopped. Vigil records the date access ended and the last verified position, and says explicitly that the interval establishes neither outcome.

Incompatible measurements are not the endpoints of a range.Where two figures count different things — a documented individual count and an unpublished estimate, a launch count and an impact count — they are published side by side with what each measures, and never as a low and a high of one number. A range implies a distribution, and there is no distribution across two different questions.

2. One corpus, four surfaces

Vigil publishes one body of canonical records and exposes it through four surfaces — the four ways of using Vigil Research, the public research system. No surface is a separate truth layer. Each resolves back to the same records, with the same dates, grades and sources, and a test suite asserts that the surfaces cannot disagree: none may label, grade or classify an object the others do not.

• Read — authored context: the publications (Weekly Briefs, Analytical Notes) and every canonical record rendered as a page.
• Explore — the same records made selectable, in a set of modes per module (profile, actors, diplomacy, forces, logistics, economy, humanitarian, geography, timeline, watch, evidence), each panel linking back to the record it came from.
• Map — geographic access to the same canonical geography, infrastructure and location records, and, where a module publishes one, the recorded activity tempo at those markers over time with the dated developments beside it. A marker is a published location record; it does not create a reality the records do not carry, and a count of records is not a measure of what happened.
• Watch — the standing analytical questions, their observables and their review history, over the same system.

Aggregation adds no claim. The module Profile, the Economy and Humanitarian surfaces and the Explore panels compose records; they do not score, rank, total or average them. A grouped object shows the records behind it with their own dates and grades, and a figure that appears both as a module headline and inside a pressure is one record at two altitudes, not two numbers. Explore's Evidence mode lists each record's grade, review date, gaps and source count and computes nothing from them. Nothing on any surface updates itself.

3. Publication types

• Weekly Brief — the scheduled review output: a dated developments brief whose every item carries its own reporting window, the specific reports behind it, its registered sources and its confidence grade. Those requirements are enforced by the schema rather than by editorial memory. A brief reports material change or meaningful quiet: a quiet review lists the standing files examined and states that none moved. One cross-conflict edition accompanies the module briefs each cycle.
• Analytical Note — triggered by a question or development that needs deeper mechanism analysis than a brief item can carry. Deeper than a brief item, narrower than a module; published event-driven, so it does not follow the weekly cadence and does not replace a module's brief. A Note carries its own sources, confidence and assessment date, and the current standard is a structured body where the analysis supports it: key findings; what changed and what did not; what the evidence does not establish; the canonical records the analysis connects to; its uncertainty; and what it means, what it does not mean, and the observables that would move the picture. Where a Note authors a map, the map draws only located records the module already publishes.
• Standing record — a canonical object updated over time rather than published once: an actor, an event, a relationship, a formation, a supply system. It carries its own review date, and a dated brief that changes it says so on the record.
• Watch review — a dated entry in a standing question's ledger, stating what one review found and the brief, event or Note it rests on. Defined in full under Watch.
• Module baseline — the first brief a module publishes, covering a longer window because it establishes where things stand. A module that joins coverage mid-cycle enters with one, and owes its first ordinary Weekly Brief in the next cycle.

Status and cadence language

Every conflict module describes itself in the same vocabularies, so a label on one page always means what it means on another.

Publication status — where a module is in its life:

• Published — live and usable, and still being expanded and reviewed. Every record inside it carries its own confidence grade, status date and stated gaps, so how settled any individual claim is never depends on this label.
• Preview — in development; reachable for editorial review only, never listed publicly.

Update cadence — how often a module is reviewed:

• Weekly review / event-driven updates — the module is reviewed every week and publishes a dated Weekly Brief for that cycle; a review that finds nothing material says so, rather than going silent. A quiet week is reported as a quiet week, not padded. Currently all eight published modules.

A cadence is both a review commitment and a publication commitment. Reviewing a module weekly means looking every week and publishing the finding, including when the finding is that nothing material changed. What it never means is filling a quiet week with material that does not meet the evidence bar: a quiet review states what was examined and what it found, and is the same length as it needs to be. All eight published modules are reviewed every week. Each receives a dated weekly brief; quiet weeks explicitly record when no material change was identified.

Baseline reference pages are under progressive source review: records are re-checked source by source, and each shows its own review date. Each module additionally states its analytical type(interstate war, conflict system, regional system, regional transition, strategic flashpoint) and a small set of thematic tagsdrawn from one shared vocabulary. Tags describe the kinds of content a module holds (energy, sanctions, displacement, corridor…). They are themes, never sides: no factions, no flags, no symbols.

4. Evidence and sources

Vigil synthesises from public material: official statements, established media, OSINT and geolocation communities, satellite-derived reporting, international organisations, NGO and civil-society data, think-tank analysis, and market data. It does not use leaked, hacked, or operationally sensitive material.

Every citation on the site references a registered source by a stable id; the build fails if a record cites a source that is not in the registry. A source's type classifies how a claim is grounded — an official statement, an international-organisation dataset, an OSINT geolocation — not how true it is. How strongly the evidence supports the claim is a separate judgement, expressed by the confidence grade. Browse the full source registry →

5. Evidence states

Beside the confidence grade, several kinds of record carry a state saying how firmly one specific relationship or step is established. Vigil does not use one universal state vocabulary; it uses a small one per kind of claim, because "confirmed" means something different of a subordination, a map point and a causal step. In every case the state describes the evidence for that claim, never the standing of whoever made it, and a reported state is never rendered as a confirmed one.

• How an organisation sits in a force structure — confirmed, reported, disputed, historical or unknown. A reported subordination is drawn differently from a confirmed one.
• How equipment relates to an operator — documented, reported, historically associated, captured or prospective; and how firmly the system's presence in a conflict is established — confirmed, documented, reported, associated or unverified.
• How an organisation relates to an event — participant, reported participant, attacking, defending or supporting force, affected organisation, historical association, and predecessor or successor involvement.
• A step in an impact chain — confirmed, plausible or unconfirmed. A plausible step is a scenario-like claim and is marked as one; a chain is never shown as more established than its weakest step.
• A campaign's frame — an observed operation, a declared plan or an assessed contingency, which are three different kinds of claim and are labelled as such wherever they appear.
• An actor's red line — stated, inferred or disputed; an actor's objectives are kept as stated (sourced, ungraded — someone said it) and assessed (graded — Vigil says it), and no surface may merge the two.
• A point on an authored Note map — confirmed by an official statement, reported by credible reporting, or withdrawn.
• A Weekly Brief item carries no state of its own beyond its confidence grade and its reports; where it describes a contested attribution, the item says so in words.

6. Confidence

Material quantitative claims are presented with an as-of date, a confidence grade, and a source wherever applicable — a value is shown with the context needed to weigh it, not on its own. The grade is one of three:

• High — multiple independent, credible sources agree; little serious dispute.
• Moderate — credible sourcing with known gaps, wartime distortion, or meaningful disagreement between estimates.
• Low — contested, unverifiable, or single-sourced; published only when the uncertainty itself is informative, and labelled as such.

Confidence is not importance, and it is not severity. It reflects the quality and agreement of the evidence, its recency, and the gaps that remain. A moderate grade beside a displacement figure describes how well the figure is established, not how bad the situation is, which is why the site prints the grade as a worded label rather than a bare word beside a number. Where a figure cannot be responsibly estimated, Vigil says so — "not reliably estimable" is a legitimate published value here.

Why high confidence is the quietest colour

Confidence grades are coloured the opposite way round from most publications. High is the calmest grey on the page, Moderate is the site's amber, and Low is the only one that reaches for a warning colour. Most dashboards do the reverse — certainty in green, doubt in red — which trains a reader to skim for the green and to treat the qualified claims as noise.

Vigil inverts it because the colour should track how much is being claimed, not how comfortable the claim is: the louder the colour, the less Vigil is claiming. A high-confidence line needs no emphasis; it is the ordinary state of a published claim here. A low-confidence one is where a reader's attention is actually worth spending, so that is where the emphasis goes. The same scale and the same three colours are used everywhere a grade appears — briefs, notes, records, maps, Explore and exported images.

7. Dates and freshness

Vigil keeps several dates apart on purpose, because a single "updated" stamp hides more than it tells. Publication date, event date and review date are three different facts, and a page shows only the dates its own record authors:

• Published — the date a publication object went out: a Weekly Brief, a global edition, an Analytical Note. It is not revised afterwards.
• Assessed as of — the date of the newest evidence supporting the current status or claim. Often older than the review date, and deliberately so.
• Last reviewed — the date Vigil most recently checked the record. It says when someone looked, not when anything was true.
• Changed in brief — a dated Weekly Brief explicitly recorded a change to this record, and links to it. Only an authored change counts; being mentioned in a brief is not a change.
• Coverage reviewed — the date a whole coverage area was reviewed and briefed. It does not claim every record inside it was revalidated.
• Reporting window — the period a developments brief covers.
• Data-specific dates — leadership, economic statistics, formation status and equipment status each carry their own as-of date, which a page review never silently advances.

A recent review date does not make older evidence current.Vigil preserves the newest genuinely sourced status date rather than refreshing it to look up to date, and a quiet review cycle is recorded as a review rather than left to be mistaken for inattention.

8. Canonical records

Vigil uses typed records rather than prose alone because analysis should link back to durable objects rather than become isolated text. Every conflict module is a set of typed records, each with its own dates, confidence grade and sources, validated against one schema at build time. A page renders records; it never authors a claim of its own. The families:

• Module — the standing judgment about a conflict system: its lede, key facts, historical phases, the questions Vigil is following, and the economy and humanitarian sections described below.
• Actor — a party as it exists in one module: its objectives (stated and assessed, kept apart), constraints, leverage, red lines and standing analysis. The same real-world state has a separate actor record in each module it appears in, because what it is doing in one theatre is not what it is doing in another.
• Country — identity only: name, formal name, codes, capital, alternate names. Never a profile; every claim about a country lives on its actor records.
• Office — a named office of state and the dated succession of its holders.
• Relationship — a typed, dated, graded arrangement between two actors: alliance, defence agreement, security cooperation, economic partnership or dependency, material support, sanctions, basing, institutional membership, territorial dispute, adversarial posture, or a relationship credible sources dispute. Each has one authored record, read from either party's side; the absence of relations is a status on the record, never a separate type.
• Event — a specific occurrence with a defined date or narrow time window, with its participants and their roles graded.
• Formation — an organisation in the Force Atlas: its place in a tree, its lineage, its equipment associations and, where reviewed, its doctrine. A stub states structure and nothing else, and says so.
• Equipment — a family-level system record shared across modules, with its operators on the module edge.
• Location and zone — a place on the theatre map with its own summary, and a deliberately coarse area of status or influence. Zones are never published as documents, because a generalised polygon presented as a page would read as a precise claim.
• Terrain context — why the ground constrains what is possible. Strategic logistics — a supply system, its nodes, what depends on it and its dated disruptions. Campaign — a sustained effort, carrying its frame: an observed operation, a declared plan or an assessed contingency.
• Impact chain — a mechanism from trigger to effect in four stages, each step graded as set out under evidence states.
• Economic exposure and humanitarian pressure — filing objects over the module's own economy and humanitarian sections. An object names which commodities, analysis blocks, figures and observables belong to it and which records in other collections it runs through. It carries no figure, no date and no grade of its own beyond an exposure class, and it claims nothing twice.
• Watch question — defined under Watch. Profile object — defined under Profile.
• Historical phase, Weekly Brief, Analytical Note, glossary term, source, media entry and correction — as described elsewhere on this page.

Events and relationships

Records are linked to each other with typed, graded relationships, so a broad association never masquerades as a precise one. A link binds to an event (a specific occurrence), a campaign(a sustained sequence of related activity), a campaign phase(a distinct period within one) or a theme (a recurring issue connecting several events). No link is preferable to an unsupported link. Every edge can carry its own confidence, date and sources; where an association cannot be supported, it is left out.

Provenance at every level

Sources are registry ids, resolved at build; an unregistered id fails the build. They are carried at the level the claim is made: on every standing record; on each block of economy and humanitarian analysis; on each humanitarian headline figure and on leadership figures; on each step of an impact chain, each holding of an office and each edge that names a formation's equipment or an event's participants. A block or figure without its own source rests on the section's or the record's sources, and the audits report it rather than fill it in. A citation opens the registry entry first — the source's type, role and stated limitation — with the original site one click beside it.

9. Profile

A module's Profile is its strategic orientation as a set of analytical objects: what this system is, what state it is in, what the parties are trying to do and what constrains them, where it is moving, and what matters now. It is not a biography or an encyclopedia entry, and it is not a summary written for the purpose. Every object is a view over records the module already publishes, dates and grades; where a module authors none of a family, the object is absent rather than empty.

The object families, each backed by a record family the corpus already carries:

• Current state — the module's standing judgment and key facts.
• Objectives — what each actor is trying to do, stated and assessed kept apart.
• Constraint — what limits each actor, grouped by kind: domestic, economic, military, external, legal, institutional.
• Trajectory — the current historical phase and the latest reviewed brief.
• External relationships — the typed relationship records the module holds.
• What matters now — the pressures the module's indicators and standing questions point at.
• Economy and Humanitarian — the module's exposure and pressure sections.

Each item in an object carries its register: stated(a party said it; sourced, ungraded), assessed (Vigil judges it; graded) or observed (a dated record of what happened). An object shows a confidence grade only where a record authored one for the object itself; a container over individually graded items says "graded per item" rather than rolling six judgments into one. The same objects, with the same labels and grades, are rendered as a written dossier in Read and as a selectable system map in Explore; a test asserts they cannot disagree. Actor profiles — the per-actor records that feed several of these objects — are the Profilerecords described in the glossary.

10. Watch: questions, observables, reviews

Watch is Vigil's monitoring layer. A Watch question is an enduring analytical question whose answer could materially change an assessment and which can be tested against observable evidence. Each is a record of its own, with a statedcurrent condition dated to the review that stated it, theobservables Vigil monitors for it, what would change the assessment, its sources and confidence, its information gaps and alternative explanations, and the canonical records it runs through.

The grammar is four concepts kept apart: the question(what Vigil is trying to determine), the observable(what evidence would bear on it), the observation (what a dated review actually found, and the record it rests on) and the assessment (whether the standing reading changed).

An observable is something that could be seen to happen, and that would bear on the question if it did. It is written so that a reader can check it themselves: a delivery rather than an announcement about production, an enactment rather than a bill being proposed, a corroborated movement rather than a claim of one. When an observable is met, the record says on what date and against which review — "met" describes evidence, not success.

Review cadence and outcomes. Each question is reviewed on the Weekly Brief cycle and can be updated between briefs when material evidence appears. Each review is a dated entry in the question's ledger with one of five outcomes:baseline (the question was opened), no material change,movement observed (new evidence; the assessment holds),observable met, or assessment changed (the condition line advanced). An entry names the brief, brief item, event or Analytical Note it rests on; no entry is written for a review that did not take place. A quiet week is a finding — the file was asked and the answer recorded — and where a question states why continued absence matters, the run of quiet reviews is shown as the measure.

A question has a lifecycle, separate from its reviews. It is active, resolved (its answer arrived) orsuperseded (a better question replaced it, and the record names the successor). A closed question is not deleted: it keeps its page, its ledger, its basis and its lineage, because how a question was answered is analytical history.

Watch does not predict and does not alert. There is no probability, no expected date, no threshold, no severity, no threat level, no status light and no ordering by urgency — none of those has a field, and none may be added. Nothing updates itself: a question moves when an editor reviews it and records what was found. What Watch makes explicit is the evidence that would cause Vigil's understanding of the present to change, and then whether that evidence appeared. A watch item in a Weekly Brief is the same review seen from the brief's side.

A notification is delivery, not detection. Vigil may tell a subscriber that a Watch question has moved — that an observable was met, that new evidence was recorded, or that an assessment changed. What is sent is an editorial outcome Vigil has already reviewed, recorded and published, arriving by a different route. It is not an independent detection, not a forecast, not a warning and not a threat alert, it is generated by no model, and it changes nothing about how the question is reviewed or what the review may conclude. A question that nobody is notified about moves on exactly the same evidence, on the same cadence, to the same recorded outcome.

Standing Watch, and paid monitoring

A Standing Watch, part of Vigil Professional, is a selected set of canonical Watch questions maintained around one defined exposure — a responsibility, a decision context, a route, a sector. The selection and the stated relevance to that exposure are what a scope adds. Everything under it is the public corpus: the same questions, the same observables, the same dated records, the same sources and the same review outcomes, produced by the same weekly cycle. A scope consumes that work; it does not run a second research process.

One editorial standard, for everyone. Vigil Research and Vigil Professional are two environments over one set of records, not two evidence systems: material published for subscribers meets the sourcing, attribution, evidence-state, confidence, review and corrections rules set out on this page, without exception and without a variant. There is no subscriber-grade claim, no relaxed threshold for material fewer people will see, and no finding that would not survive being published openly. A correction is issued and logged the same way whether the error appeared in a public brief or in paid material.

More depth, not a different truth. What a paid layer may add is quantity and instrumentation: further monitoring depth, additional observables, more scopes, more activity series, and tools for history, comparison and personal workflow. What it may not add is a separate analytical authority. Nothing is assessed differently, weighed differently or stated with more confidence because it is behind a subscription.

Personalisation changes delivery, never analysis. Following or saving a question changes what a reader is shown and what reaches them. It must not change what counts as movement, what an observable requires, what confidence a finding carries, how evidence is weighed, which questions Vigil opens or closes, or what Vigil chooses to research. No question is reviewed more attentively because somebody is following it, and no reader's selections are an input to an assessment.

Comparison stays descriptive. Activity series placed side by side remain what §13 defines them to be: counts of dated, reported occurrences under separately stated inclusion rules. Putting two of them on one axis lets their timing be compared and does not make them commensurable measures of anything. Vigil publishes no universal escalation score, threat score, normalised risk number, composite index or predictive signal on any surface, to any reader — and will not, unless it builds such a thing openly and writes its method on this page first.

Payment buys none of it. A subscription buys access to Professional functionality and material. It does not buy influence over assessments, sourcing, confidence language, coverage or corrections — the same commitment Vigil makes about sponsorship and voluntary support, stated here about the mechanism that could most easily erode it.

The Professional tools have their own method. The rules that bind Vigil Professional's observable records, Tempo Lab, comparison, Scenario Lab and briefing tools — what each is and is not, what it uses, what may be inferred from it and how staleness shows — are published at /professional/methodology/, readable without an account and subordinate to everything on this page. In particular, a scenario there is a conditional exploration of the recorded system under stated assumptions, and never a forecast. What the product itself is, what it costs and what it refuses to do is set out at /professional/, and one published Professional record is shown in full, free, on the walkthrough linked from it.

11. Force Atlas

The Force Atlas is a module's structured catalogue of the military and security institutions in it: state armed forces and interior forces, their services and branches, commands and formations, non-state and private military organisations, the organisational relationships between them, their lineage, and the equipment families they are documented to field. It is representative rather than exhaustive, every parent link carries its own evidence state, and a record that is only a structural stub says so rather than carrying generic prose.

It does not publish current tactical positions, unit readiness, live strengths, movement or routes, or any targeting-quality detail. Force information is broad, delayed and historical, and the record model has no field for the excluded class — the omission is architectural, not a matter of restraint on the day.

Identity doctrine. A mark on a record is identity, not decoration, and Vigil draws no heraldry of its own. Identity resolves in one fixed order: a sovereign state shows its state flag; an international or regional institution shows its own mark and is never given a country's; an organ of a state — an armed forces, a ministry — shows the flag of the state whose organ it is, as orientation rather than as a claim; a service or formation shows its own insignia only where the mark can be established from an official description that fixes the geometry or from a file that is itself free of copyright, and a formation's insignia is never inherited downward; and where none of that can be defended the record shows a neutral Vigil category symbol, labelled as such. A flag on a formation answers "whose force structure is this" on a page that shows two armies at once; nothing is coloured, ranked, grouped or counted by it.

12. Non-operational by design

Vigil describes conflicts structurally and historically. It does not publish current unit positions, force concentrations, base coordinates, movement tracking, readiness, live strengths, routes, detailed vulnerabilities, or anything designed to support targeting. The content model has no fields for such information — the rule is enforced by the site's build system, not by editorial memory — and no schema, surface or map layer is designed in a way that would let the site become a tactical tracker. Front lines on maps are approximate and labelled illustrative. Military and nuclear sites are shown at area level only. Humanitarian coverage stays at broad regional level: there are no fields for facility locations, aid routes, convoy timing or operational coordination.

13. Maps and geography

If Vigil cannot defend the geometry, it does not draw it.A map on Vigil draws real coordinates of published location records, located infrastructure, and authored areas only where the area itself is defensible. It draws no invented influence areas, no intent arrows, no strike or weapon radii, no targeting detail and no false precision, and where a location is uncertain the record says so rather than placing a confident marker.

Map zones. Some maps carry broad coloured zones. Zones are approximate, contextual and non-operational: every zone is source-dated, confidence-labelled, and drawn with deliberately coarse geometry — the build system caps how much detail a zone outline may carry, so a zone cannot be drawn precisely even by mistake. Zones are not a live control map, and their meaning differs by conflict. For Ukraine–Russia, broad control zones do not show tactical boundaries or live unit positions. For the Sahel module — which covers Mali, Burkina Faso and Niger — zones show broad influence, activity, access pressure, or administrative context; influence and activity are not the same as territorial control, and no Sahel zone claims otherwise. For regional-system modules such as the Caucasus, zones show de facto administration, historical conflict areas and broad corridor bands: a de facto zone is a status description, never a recognition claim; a historical zone marks where a past conflict was fought, never a current line; and old ceasefire lines are never presented as current control. For regional-transition modules such as Syria, an "administered" zone shows the broad administrative reach of a transitional government — deliberately not a consolidated-control claim — and all other zones are influence, activity or integration-status context; the vocabulary structurally cannot draw a front line.

Four things a fill can mean, kept apart. Physical control, a declared restriction, commercial behaviour and assessed influence are different claims about the same ground or water, and one colour cannot carry them. Where they diverge the module says so in words and the map states none of them.

No inferred route geometry. Where Vigil publishes the nodes of a corridor — a pipeline, a supply route, a transit lane — it publishes the nodes. It does not draw the line between them unless the geometry itself is sourced. An ordered list of places is not a route, and a line drawn through them would be an inference presented as a survey: it would state a path, a length and a set of intermediate locations that no source supports. Where a corridor is shown as an area, the area is the published geometry and is deliberately coarse. The absence of a line is intentional, and a corridor record says so rather than leaving a reader to wonder whether the map failed to load.

Sea areas. Some modules are about water rather than ground, and their zones are broad sea areas: an enclosed sea, the approaches to a strait, the open ocean beyond it. A sea area is orientation geometry only. It never depicts naval positions, patrol areas, transit routes, mined water or strike zones, and it never encodes whether a waterway is open — legal access, physical access and commercial access to the same water can differ at the same moment, and one fill cannot carry three answers.

Authored figures. A locator on a record page and a map in an Analytical Note are drawn from the module's own public-domain outline and the module's already-published location records; they can draw points and names, and they have no means of drawing an area fill, an arrow, a route or a front. The interactive map, with its own layers, provenance and as-of date, is always one click away.

The Map surface: activity tempo and recorded events

A module with the map capability has a Map surface of its own, and it asks four questions in one frame: where the module's published geography is, when recorded activity in a defined population intensified or paused, which dated developments explain the pattern, and which standing Watch questions it bears on. Nothing geographic, dated or editorial is authored on that surface: it composes location records, event records, Watch questions and one collection of its own — the activity series.

An activity series is a defined population, stated in full. Each series declares in writing what it counts, what it deliberately does not count, which of the module's published markers it counts at, where its records come from, and the gaps it knows about — and the surface prints all five beside the count rather than behind a link. A record in a series is a dated, reported occurrence at a published marker. It is not a measure of damage, duration, output lost, casualties or effect, and no field exists on the schema for any of those.

What a recorded-event count does not mean. A series counts what Vigil has recorded at markers Vigil publishes. It is therefore a lower bound on the published record, never the total activity in a theatre: activity at an unmapped site is not counted, and each series names the specific cases it is known to be missing. Nor is the count a measure of intensity, escalation, threat, danger, operational tempo or momentum. A busier bin means more records met the series' stated inclusion rule in that period; a quieter bin means fewer did, and may mean that reporting thinned rather than that activity did. Counts of different series are not commensurable with each other, and none of them is commensurable with a figure published by anyone else.

The tempo band is arithmetic, not a surface. Records are placed in weekly or monthly bins by date and counted; the count is drawn as one accent at five fixed steps. There is no smoothing, no interpolation, no second colour, no per-row normalisation and no density estimate between points — and there is no heatmap anywhere on Vigil, on the map or beside it. A period a series does not cover is drawn as no coverage and is visibly distinct from a covered period with zero records, because "we were not counting" and "we counted none" are different statements. Every series a module publishes is a row on one shared time axis so their timing can be compared; exactly one row is active, and the map, the developments row and the inspector follow it. Selecting a period narrows the surface to the records inside it, and the inspector lists them, so a bin can always be opened back into the dated, sourced records that produced it.

The developments row is a chronology, not a feed. The dated developments on the axis are editorially selected key events with pages of their own; they explain the pattern rather than restating it, and the row states how many of them fall inside the selected period. It is not a front line, not a reconstruction of a continuous state, and not a live picture — the Map surface shows no current deployments, positions, routes, readiness or targeting information, under the same rules as every other map on this site (§12).

A series is maintained, and says when it was last reviewed. Each carries its own status-as-of and review dates and a dated review ledger tied to the Weekly Brief that made each entry, exactly as a Watch question does. A series whose subject has ended is closed and keeps its records rather than being deleted.

Activity tempo is not maintained for every module, and that is a decision rather than a gap. A series exists only where Vigil has a defensible, repeatable inclusion rule — a population it can define, a set of published markers to count at, and reporting it can apply the rule to consistently. Current coverage is Ukraine–Russia, Sahel, Caucasus, Syria, Taiwan Strait, Iran–Gulf and NATO Eastern Flank, of eight published modules. The others have no activity series, and Vigil does not create one for consistency of presentation: a module without Tempo is not a quiet module and not a lesser one, and an absent series never means an absent theatre. Counts it could not defend would be worse than none.

Regional systems and transitions

Some modules cover a linked regional security system rather than a single conflict. Each module states its type — interstate war, conflict system, regional system, regional transition, or strategic flashpoint — and groups its content into declared sub-theatres: distinct arenas whose events, locations, profiles and zones are tagged and validated against the module's own list. Grouping is analytical, not political — placing a record in a sub-theatre implies nothing about status or sovereignty.

A regional-transition module covers a post-war transition: the historical conflict is the backbone, the current transition is the subject, and the two are kept separate as claims. Historical control and battle context are dated as historical; current authority is described as status — "administered", "integrating under agreement", "outside state authority" — with confidence grades, never as settled fact; and disputed authority or autonomy is described through each party's position ("Türkiye considers…", "Kurdish-led authorities describe…", "Damascus states…") without adopting any side's terminology as the site's own voice. Foreign military presence is represented at broad structural level only, exactly as elsewhere on the site.

A strategic-flashpoint module covers a system where sustained pressure, deterrence failure or political escalation could produce a major crisis. It is not a prediction and not a countdown, and it does not track live operations: there is no front line, no control layer, no activity layer and no scenario content — and in particular no description of how an attack would be conducted or resisted. Vigil uses the type to connect public signals with geography, actors, capabilities, economic exposure and civil resilience — the structural picture that would matter if a crisis came, described while it has not.

Two consequences follow, and both are rules rather than preferences. A flashpoint module's map shows broad strategic geography only: never live military positions, patrol routes, force dispositions, air-defence locations, base layouts or operational vulnerabilities. And where such a module reports military activity, it reports officially published figures as one government's detections rather than as measured fact, and treats the count as context rather than as a threat metric — a quieter month is not de-escalation, and a busier one is not a timetable. Where the underlying dispute concerns sovereignty, the module describes each party's position and does not adjudicate between them.

Regional-system modules apply this site's separation of historical and current claims strictly: entities that no longer exist are profiled as historical; forces that were disbanded carry their dissolution dates; and where sovereignty is contested, the module describes de facto authority, international recognition and each party's position separately — using formulations such as "de facto authorities", "internationally recognised as part of Georgia", or dual place-naming (Stepanakert / Khankendi) — without adopting any side's terminology as the site's own voice.

14. Documentary imagery and visual evidence

Documentary imagery provides context; it does not silently become evidence. A photograph on a Vigil surface is an orienting image: the caption and credit state what the frame shows and when, and no surface says the photograph supports the assessment printed beside it. A picture near a claim proves nothing about the claim.

An image may appear on Vigil only if it has an entry in the media registry, and an entry cannot be authored without the fields that make an image defensible: creator, source, licence, the place and — to the precision the source states, never tightened — the date the frame shows, a caption that states when it was taken, alternative text, and what Vigil did to the file. Every published frame carries its credit, linked to the source. A candidate is admitted only if it passes five tests — relevance, rights, provenance, context and restraint — and a candidate that fails is recorded with the reason and can never be referenced by a surface.

• No image with unclear rights. The admitted licences are public domain, CC0, CC BY and CC BY-SA, with their obligations met on the asset itself; NonCommercial, NoDerivatives and unlicensed material are refused.
• No generated imagery presented as documentary photography.
• No event implied from unrelated context imagery, and no crop that changes what the frame shows. Permitted treatment is a crop, a resize, restrained tone and exposure correction and a legibility overlay; nothing is added to or removed from the frame.
• No dramatic imagery chosen for mood. A surface with no defensible photograph keeps its authored cartographic ground, and that is the normal state rather than a gap.
• Maps, locators and authored graphics follow the evidence rules of the maps section, not the imagery rules; they are drawn from records, not from pictures.

15. Neutrality

Vigil describes; it does not advocate. It is not neutral about evidence quality, and it does not pretend to a complete philosophical neutrality no analysis can have. What it holds itself to is precise: the grade a claim receives does not depend on which actor made it; the identity of a source is part of the evidence and is never itself proof; an actor's stated perspective is recorded as that actor's and kept apart from Vigil's assessment; and the evidence standard is the same for every party. Actors are labelled by role (belligerent, backer, hedger) and bloc, never as "friendly" or "hostile". Neutral framing does not mean false balance: where facts are documented — an invasion, an annexation, a strike on civilian infrastructure — Vigil names them plainly.

16. Economic and humanitarian analysis

Economic sections explain exposure and transmission: the physical mechanism by which pressure on a war, a border or a strait reaches the systems people trade, burn, eat, insure and budget with. They work at the level of infrastructure, trade, energy, commodities, access and constraint, and at the level of sectors and instrument categories. Exposure and confidence are different things and are never printed as one bare word. Nothing on this site is investment advice, and Vigil never recommends, ranks, or names securities as opportunities.

Five quantities that are not each other

Reporting on infrastructure routinely collapses five different things into one number. Vigil keeps them apart, everywhere, and says which one it means:

• Capacity — what an asset is designed or certified to move. A design figure, often established once and never demonstrated since.
• Availability — how much of that capacity is physically usable now, after damage, maintenance or a constraint elsewhere in the system.
• Flow — what actually moved, measured, over a stated period.
• Dependency — what has no alternative if that flow stops. A route with a substitute and a route without one are different risks at identical volumes.
• Scenario — what would happen under an assumption. Always labelled, never mixed into the four above.

An alternative route's capacity is therefore never presented as relief, and production withheld because its outlet is unusable is not a loss of the resource. Where a figure is a forecast it is labelled as one and carries the vintage of the release that produced it: successive outlooks revise each other, and a revision is not a contradiction, but a March estimate quoted against a June forecast measures the release rather than the world.

A denominator is part of a share. "A third of global crude trade" and "a fifth of world petroleum liquids consumption" can describe the same flow and are not interchangeable. Vigil states what each share is a share of, and never mixes two denominators inside one comparison.

Humanitarian analysis

Humanitarian sections answer a question about the people inside a conflict: who has been displaced or harmed, which of the systems a civilian population depends on is failing, and whether anything can reach them. Their object is a pressure — a need and the population and service it lands on — described through dated, sourced figures, the mechanism recorded in the module's impact chains, the access recorded in its logistics, and the geography recorded in its terrain. Each figure carries its own as-of date and grade, and nothing merges, averages, indexes or ranks across them: Vigil publishes no severity scale, no phase classification, no crisis ranking and no needs index, and the record model has nowhere to put one. Where a module's own records state a condition in words, those words are rendered; where they do not, the object carries no state at all.

17. Corrections

Corrections are part of the record. A material error — one that changes what a reader would conclude, as opposed to a typographical fix — is corrected in place and logged publicly on the corrections page, with the date, the page and what changed; the log is never purged. Assessments may be revised as evidence changes, and a superseded assessment remains traceable: a Weekly Brief is a dated snapshot that is never rewritten, a closed Watch question keeps its ledger, and a standing record's recorded changes link to the brief that made them. If you find an error, Vigil wants to know: the route is on the corrections page.

18. Completeness and limitations

These bound what any record on the site can claim:

• Vigil is selective. It covers a small number of conflicts, and within each, the records that have met the publication threshold; absence from Vigil does not prove non-existence.
• Public information is incomplete, and some of what matters is never disclosed. Vigil does not assume access to private information, and absence of public evidence is not always evidence of absence.
• Information environments differ by conflict: what is observable in one theatre is invisible in another, and public-source availability is uneven.
• Wartime deception is deliberate — every side shapes what is visible.
• Organisational structures change faster than they are documented.
• Confirmation is often delayed; early accounts are revised.
• Casualty and equipment-loss estimates are uncertain and contested.
• Economic data in wartime is partial and lagged.
• The available sources carry their own selection biases, and translation across languages loses nuance.
• Some objects are developed in depth and others are deliberately brief; each says which it is, and a confidence grade cannot remove the uncertainty it describes.

Profile completeness

Every organisation and equipment record states how complete it is, in the same four terms used on its badge:

• Full — reviewed prose, links, dates and sources are published.
• Partial — core information is reviewed, but some sections remain incomplete.
• Stub — a minimum structured identity and relationship record is published.
• Research required — the record remains hidden until it reaches the publication threshold.

Vigil prefers an honest stub over generic or invented prose.

19. Source registry

Every source Vigil cites is listed in a single registry, with its type, its role, and its stated limitation. Citations resolve against it by stable id, and the build fails on any unregistered reference. The public registry lists the sources that published records cite. Browse the full source registry →

20. AI assistance

AI tools may assist with search and discovery, source comparison, organisation, drafting, translation, coding, tests and structured-data work. Epistemically their output has one status: AI output is never itself a source. Text a model generates is a draft until it resolves to evidence in the registry; a source a model names is a claim to be checked, including that it exists and says what is claimed; and a model's own confidence, however expressed, is not a Vigil confidence grade and never substitutes for one. Fabricated citations and confident errors are treated as a known failure mode of the tool and are checked for as risks, not assumed away. No model publishes to Vigil, and editorial responsibility and publication approval remain with Vigil. Where AI sits in the workflow is set out in the editorial process; the short disclosure is on the About page.

21. The standard

Vigil's standard is not perfect certainty; in conflict analysis, perfect certainty is usually impossible. The standard is four tests a piece must pass before it is published, and a reader can hold any record to them:

• No claim without a basis. If it cannot be sourced, it is not published — not softened, not hedged; not published.
• No event without context. A development not connected to the system it belongs to is a headline, and Vigil does not publish headlines.
• No assessment without uncertainty. A grade, a stated information gap, and what would change the judgment.
• No forecast without an observable. If nothing could be watched that would confirm or refute it, it is an opinion with a date on it.

What Vigil offers, then, is orientation that can be traced: what is known, what remains uncertain, why a claim is included, when it was last supported, and how it connects to the wider conflict. When the evidence changes, the record changes with it.