An evidence atlas.
A plan to make it useful.

AVAILABLE NOW

A working geographic workspace

• Twelve satellite watch areas, multiple Sentinel-2 and Landsat passes, date comparison, acquisition metadata and source links.
• An always-visible Map / Satellite switch, OpenStreetMap context and four regional NASA imagery options, Sentinel-1 catalogue records and optional four-product NASA FIRMS thermal observations.
• Reported city-control points in Sudan, with dated sources, opposing-evidence notes, expiry and historical selection. Sudan, Ethiopia and Yemen now have thirteen community-mapped faction categories, separate from those city assessments, with a published source revision date and historical revisions.
• Forty-five source-linked faction profiles and filters, including separate political identities and coalition members. City evidence links directly to imagery and reporting.
• Report filters for region, topic, period, group, language and publisher. Indexing activity is clearly separated from incident counts.
• Local bookmarks, shareable views, evidence-preserving GeoJSON export, feed health, source inventory and ten public read-only API endpoints, including attributed territorial GeoJSON.

News collection runs every 30 minutes; imagery availability is checked every six hours. Control assessments are public-source syntheses added explicitly by an operator. They are not automatically refreshed or ground-verified. Unknown areas stay unknown.

COUNTRY NAVIGATION · DELIVERED

The Horn, country by country

Ten country views, more detailed locally hosted reference borders, highlighted selection, shareable country state and filtered imagery/reporting. Reports have chronological date sections and numbered watch-place references. Country borders and missing territorial coverage remain separate from armed-group areas. How to read the map.

Future original territorial coverage needs a first-party evidence database, independent cartography, editorial review, contributor IP assignments and source-specific redistribution permissions. Current community shapes retain CC BY-SA. Data rights.

NEXT · REVIEW AND PROVENANCE

Claims that can be challenged

Build a persistent observation and claim database, with separately stored publication, event, acquisition, retrieval and review times. Identify copied posts, syndicated articles and repeated videos before counting corroboration. Add a private reviewer queue, supporting/contradicting evidence relations, audit history, corrections, retractions and contributor moderation.

Acceptance: every published assessment resolves to retrievable source records; a correction changes the revision and appears in history; an expired or retracted record cannot appear as current. Independent reviewers should be able to reproduce the reasoning.

Multilingual entity resolution needs Arabic, Somali, Amharic, Tigrinya, Oromo, English and French aliases, transliteration, ambiguous-place handling and administrative gazetteers. Translation keeps the original text and model/version beside it. Translation does not create an additional source.

REQUIRES EVALUATION

Change candidates, with image quality

Ingest suitable optical bands and terrain-corrected radar through documented catalogues. Co-register passes, mask clouds/shadows, preserve native resolution, compare seasonally appropriate baselines and report gaps. Roads, flooding, vegetation, construction and differing radar geometry can resemble conflict damage.

Acceptance: a change proposal exposes baseline/current scene IDs, acquisition times, masks, algorithm/version and quality score. Benchmark false positives on non-conflict changes and reviewed historical cases before showing public classifications. A thermal anomaly or image difference cannot alone establish an attack or who controls a city.

Free optical imagery has cloud and revisit limits; free radar metadata is not a processed image. Very high resolution or frequent commercial tasking requires a procurement and redistribution budget. The atlas currently offers manual comparison, not automated combat detection.

REQUIRES DATA AND EDITORIAL CAPACITY

Dated territorial coverage

Delivered: coarse Sudan areas from an openly licensed community map, three initial source revisions, group filtering, clickable provenance and fill opacity. The Somalia Commons source is disputed and dated 2025, so it is excluded. Ethiopia adds ENDF/allied areas, TDF, OLA, Fano and ANURF. Yemen adds PLC/allied areas, Ansar Allah and a source-mapped AQAP pocket. Other countries remain explicit coverage gaps.

Obtain lawful territorial datasets or build reviewed, coarse geographic assessments from multiple evidence streams. Store validity intervals, competing claims, uncertainty, location precision and source dependence. Publish coverage and the age of each assessment. A city point must never be expanded into a surrounding polygon by guesswork.

Acceptance: every boundary has a documented basis and date; conflicting claims remain inspectable; updates preserve history and revisions. Broad area context must avoid publishing individual positions or actionable military movements. Missing observations cannot be labelled uncontested.

REQUIRES PERMISSION OR CREDENTIALS

News, social evidence and economic context

The source registry covers a maintained source inventory and records active, prospective and permission-dependent access. Prioritise regional publishers, humanitarian data, public agency notices, maritime advisories, documented social APIs and lawful public-media submissions. Add explicit licence, retention, rate budget and deletion obligations before enabling each connector.

Telegram, X, Reddit, YouTube and Meta have different access and reuse rules. Public visibility does not grant unrestricted bulk collection, translation, AI processing or resale. Private chats and access-control bypass are outside this product. Approved social records need original links, publication time, language, lawful media provenance and source-protection controls.

Economic context should distinguish shipping disruption, port/airport availability and diplomatic announcements from conflict control. AIS, market data and paid imagery need provider-specific licences. A price move and a news claim do not validate one another.

COMMERCIAL OPERATIONS · PLANNED

One claim, all its evidence

A future API needs authentication, source-specific response rights, pagination, stable IDs, schema versioning, quotas, webhooks, correction events, monitoring and a service-level agreement. A client should receive record count and publisher/source groups separately, plus explicit uncertainty about independence.

{
  "claim_id": "stable identifier",
  "place_id": "gazetteer identifier",
  "claim": "attributed statement",
  "status": "reported | corroborated | disputed | retracted",
  "as_of": "assessment date",
  "expires_at": "validity limit",
  "record_count": 32,
  "publisher_groups": 8,
  "independence": "partly unresolved",
  "supporting_evidence": [],
  "opposing_evidence": [],
  "revision": 1
}

Illustrative schema only. The free API exposes the current atlas observations; no paid service, trading signal or claimed probability of truth is running. Investor and prediction-market users should inspect the original evidence and the relevant market’s resolution rules.

RESEARCH AND IMPLEMENTATION CHOICES

Use established tools deliberately

Liveuamap demonstrates a map with chronological reports and an editorial verification stage. Public conflict-monitor projects show the value of toggled layers and source status. PharoSec keeps dated evidence and disclosed coverage at the centre; the Sudan layer now adapts licensed Commons cartography with visible credits and CC BY-SA attribution. Revisions are checked every six hours. This is community cartography, not independent territorial verification.

Lucide supplies the consistent interface icons, pinned and self-hosted with its licence. Public Sans supplies the typography. The original bronze aperture mark represents intersecting observations around an unknown centre.

MapLibre GL JS and PMTiles are candidates for larger vector datasets and temporal polygons. Tile content needs its own licence. The present atlas retains Leaflet for the existing georeferenced satellite overlays. A renderer migration is justified by dataset scale, not appearance alone.

Floating UI can support complex popovers. shadcn’s official coding-agent skill helps agents work with its React component system; it is not a GPT-6-specific frontend runtime. This release uses native dialogs and restrained custom CSS rather than switching frameworks.

Laya remains a text-classification candidate. It is not an image comparison or factual verification model. Activation requires a labelled regional-language evaluation, event/source-separated test splits, calibration, abstention and published error rates.