An open view.
Visible sources.

From observation to investment research.

Terra Signals joins dated source observations to three editorial research frameworks: storm restoration (Quanta Services), forest monitoring (Planet Labs), and electricity generation (Constellation Energy). These are documented industry connections, not evidence of a contract at the selected location. The company desk retains these three frameworks. The free question workbench also calculates country electricity comparisons, national forest-area changes, event rankings, city PM2.5 and selected World Bank indicators. It routes supported keywords to explicit calculations; it is not an unrestricted conversational AI.

Each brief separates facts, a research hypothesis, counter-evidence, and the next information to check. No share-price prediction, trade recommendation or expected return is calculated. Robinhood instrument availability is checked against the official registry. A registry entry does not establish liquidity, eligibility or an executable trade. Wallet connection and purchases are not enabled.

Free question workbench

Suggested questions execute server-side calculations using Terra’s existing sources, without a paid model API. Electricity shares divide generation by source by total generation; import dependence divides net imports by demand. Missing components and zero denominators are excluded. Forest change uses the latest national value minus the value exactly five years earlier, deduplicated by country. It covers the countries represented by the forest study anchors, not every country. GDP is in current US dollars per person, not income or a city salary. Event and air-quality answers rank available records, not global risk.

Country-name matching is available for electricity, forest and country indicators. Country filtering of events and city-air readings is not yet supported. Source years can differ and coverage can be partial. Each answer exposes the calculation, source dates, original links and a JSON export. The interpretation and next-check text use topic-specific editorial rules, not a language model. Unsupported questions are explicitly rejected. No ChatGPT or Claude API is active.

Optional on-device AI

Find relevant evidence downloads the open all-MiniLM-L6-v2 sentence-embedding model from Hugging Face and runs it in a browser worker with Transformers.js and WebAssembly. It compares your question with the available source facts and editorial research passages, then selects three relevant passages. Their original wording is preserved. Scope and counterpoints are always included; the model does not write new numbers, forecasts or company claims. Similarity is a reading aid, not a measure of investment confidence.

Your question and evidence are processed locally. Hosting providers receive requests for model files; no inference API receives the research text. The first use downloads the model and runtime. Browser support, memory and connection speed affect availability. Cancel stops the worker. Source accuracy and the choice of editorial company connections still need your review.

Model and Apache 2.0 license ↗ · Transformers.js ↗

A research notebook, not a performance record

Saving keeps the original observation, hypothesis, retrieval date and optional AI reading in local browser storage. Notes can be edited. Checking a source does not overwrite the saved facts. A record leaving a feed is not evidence that the hypothesis was false. Export your notebook before clearing browser data. There is no account sync, public immutable log, portfolio tracking or claimed backtested return.

The public research journal contains three dated editorial launch snapshots with source facts and counterpoints. It is maintained with the site, not an immutable or independently audited archive. No financial outcomes have been evaluated.

Access

The globe, research briefs and local AI are currently free. No token holding or payment is required. Advanced monitoring and connected execution are future product directions, not services being sold today.

Typography: Host Grotesk by Element Type, licensed under the SIL Open Font License. The existing NASA globe remains the visual foundation.

Terra Intelligence: product direction.

Terra is an event-driven investment research workspace. Alternative data means observations outside financial statements, such as environmental events and electricity statistics. A transmission channel describes a possible link between an event and a sector; it does not establish a company contract, revenue impact or expected share-price move.

Available today

Event Intelligence combines dated observations with three editorial sector frameworks. AI evidence selection ranks passages by semantic relevance, not by expected returns. Thesis Tracking stores research snapshots and notes locally, with manual source checks and export. There is no automated investment recommendation engine.

Proposed: Scenario Lab

The proposed next stage is historical analogue research and conditional scenarios. A useful implementation would separate base, upside and downside assumptions, make sensitivity visible, and test results on held-out periods. Event studies would need point-in-time price data, benchmark-adjusted returns, transaction costs, and controls for look-ahead and selection bias. These market models, backtests and probability forecasts are not implemented or validated.

Proposed token utility

One possible design is access to advanced monitoring, research credits and batch exports, with the core explorer remaining functional without a token. This is a product proposal, not an active entitlement or token sale. Supply, access thresholds, pricing and any fee policy are undecided. No yield, revenue share or token appreciation is promised.

Tokenized instruments are a separate product from a possible Terra access token. Registry matching does not establish trade eligibility, liquidity or investment suitability. Terra does not execute purchases. Product rights and restrictions must be checked with the issuer: Robinhood Stock Tokens.

From the globe to your destination.

Search for a city and select the matching country and region. The globe flies to your destination. Dragging interrupts the flight; reduced-motion preferences remove the spins.

The study gathers current modelled air quality, earthquakes, recent nearby events and country indicators. Earthquakes cover ten years, within 300 km, at magnitude 4.5 or above, up to the 200 most recent. The twelve-month EONET storm catalogue is partial, limited to 500 global records. A nearby track position does not prove that a city suffered damage.

The country score is a baseline shared by cities in that country. Terra does not pretend that a national dataset can rank neighbourhoods. Check housing costs, visas, insurance, parcel-level flooding and local services separately.

Forests, history and outlooks.

The map opens with the latest available records for each source. The recent fire view combines NASA EONET and the public GDACS (EC-JRC / UN) feed, retaining reports whose last activity date is within 72 hours and excluding named prescribed burns. GDACS current status alone is insufficient: stale and future-dated activity is excluded. Reports from the two providers within 10 km and 72 hours are treated as a possible duplicate and the EONET record is retained. This spatial heuristic can merge distinct nearby fires. Forest reference layers and clickable study locations also appear in Active planet. It is an incomplete catalogue, not a complete live fire map; symbol sizes are not burned areas. Volcano catalogues can contain older open records; open status is not confirmation of current activity. Event profiles provide Now, History and Outlook. USGS local history covers the past 365 days within 100 km, M ≥ 2.5, capped at the latest 200 records. EONET history contains reported positions of the same event, not daily measurements or burned-area estimates.

The forest globe uses UMD / Global Forest Watch tree-cover density in 2000, v1.8, at a 30% canopy threshold, and 2025 tree-cover loss, v1.13. Cartographic tiles are reprojected and reduced for the globe; the native loss dataset is approximately 30 m but the globe is not a 30 m analytical surface. The southern Antarctic tile rows have no loss coverage. Green is a historical reference, not a current remaining-forest mask. Pink or amber loss is not automatically permanent deforestation and does not measure net change including regrowth.

Twelve named locations are study anchors, not forest boundaries. Local history uses GFW annual loss tiles from 2001–2025 around each anchor. A blank tile is not evidence of no change. FAO / World Bank forest-area series (AG.LND.FRST.K2) refer to the entire country, not the selected forest. Source estimates may interpolate between survey years.

The Outlook scenario extends the national average absolute annual change over the latest five-year interval for five further years. The result is floored at zero. It is an illustrative arithmetic scenario, not a validated forecast, and has no probability or confidence interval. Summaries use deterministic rules and source values; no language model invents observations or forecasts. Official cyclone forecasts retain their issue date and uncertainty. A link to GWIS provides official fire-danger outlooks; Terra has not connected a local fire-spread forecast.

GFW spatial-statistics queries require an API key and are not used in this keyless version. No company exposure or token availability is inferred from proximity. Financial execution and token gating are not enabled.

Companies and tokenized instruments.

Selected profiles link documented operators or regional companies: EDF, Constellation, Weyerhaeuser and Suzano. Each relationship states whether it is site-level, regional or only country-and-sector context. Proximity does not establish asset ownership, damage or financial exposure.

The official Robinhood asset registry is checked hourly for the matched ticker and a deployment on chain 4663. Missing entries remain unavailable. Issuer status does not establish eligibility, liquidity or redemption rights. Read issuer product terms. No trading or token gating is enabled.

Nuclear sites use WRI’s 2021 inventory, not a current operational census. Fessenheim is excluded following its 2020 closure. Remaining capacities and operating status must be checked against the linked IAEA source.

Related news uses publisher headlines through Google News, sorted by publication date with duplicates removed. Date and query scope are shown; a related article is not proof of an exact event match. Location photographs and satellite images retain separate dates and attributions.

A score you can inspect.

Terra Habitability v1 combines seven national criteria into a score from 0 to 100. Higher is more favourable within this model. The weights and curves below are editorial choices by Terra, not an official ranking or a scientifically validated prediction of personal safety.

Normalisation

Natural hazards, conflict and healthcare deficits use the published INFORM 0–10 values: component score = 100 − 10 × value. Water and electricity use the population access percentage directly. Homicide score = 100 × [1 − ln(1 + rate) / ln(21)], clamped to 0–100: zero reported homicides maps to 100 and rates of 20 or more map to zero. Reporting differences remain a limitation.

Annual PM2.5 maps 5, 10, 15, 25, 35 and 50 µg/m³ to 100, 80, 60, 40, 20 and 0, with linear interpolation and clamping. The first five concentration anchors follow the WHO annual guideline and interim targets. Their score values and the 50 µg/m³ endpoint are Terra choices, not WHO scores or safe/unsafe thresholds.

Aggregation and missing data

The base score is the sum of each component score multiplied by its weight, divided by 100. If INFORM human hazard is 7 or higher, the final score is capped at 40. This explicit editorial rule limits compensation for elevated conflict exposure by better service indicators.

A country is ranked only when all seven inputs are valid and their published years are within the last six years. Missing inputs are neither imputed by Terra nor reweighted away. Coverage is the sum of weights with usable inputs; incomplete countries remain visible without a final score. Ties share a rank using the displayed one-decimal score.

INFORM component dates refer to the 2026 edition, not the age of every underlying observation. INFORM itself uses imputation, and its underlying missing-indicator share is shown in each score breakdown. Mixed years and national aggregation limit comparability. This is a September 2026 snapshot, not a live alert system.

What stays outside the score?

GDP, property prices and current hourly air quality remain context. Recent event counts do not measure future hazard, so they are not scored. Housing, personal rights, visa eligibility, climate comfort and neighbourhood services are not fully represented. Do not read a high score as a guarantee that an address is safe or suitable for you.

Where the data comes from.

Named forests: Wikidata (CC0), retrieved 25 September 2026, a partial catalogue of named forest locations, not boundaries. Local loss imagery remains UMD / Global Forest Watch. Volcano sites: Smithsonian Global Volcanism Program, Volcanoes of the World 4.8.3, Holocene list downloaded 23 October 2019, distributed through the IntroQG educational data mirror. Submarine sites are excluded. This historical catalogue is supplemented with dated EONET reports, matched by name and within approximately 40 km to avoid duplicate markers; inclusion does not imply current activity. Consult the linked Smithsonian profiles for updated eruption histories.

Additional wildfire reports: GDACS public feeds. Alert colours express the source assessment, not a live perimeter. Electricity uses the bundled, dated OWID snapshot directly, so selecting the layer does not wait for a network request.

Dates are shown in each profile. Coverage depends on the provider. Free Open-Meteo endpoints are for non-commercial use; commercial operation requires an appropriate plan or self-hosting. World Bank data are generally CC BY 4.0; retain the original provider and series methodology.

INFORM citation: European Commission, Joint Research Centre (2026), INFORM Risk Index, DOI 10.2905/JRC.E5A3SHG. The Terra composite is separate from the official INFORM overall index.

Your browser, your research.

Terra saves your watchlist, research snapshots and notes in local storage on this browser when you choose to save them. These are not synced to a Terra account. You can remove entries in Watchlist or clear this site’s data in your browser settings.

This first-visit notice also stores an acknowledgement locally, so it does not appear on each visit. It is specific to your browser, not your IP address. Clearing site data or using another browser will show it again. Acknowledging this notice does not enable advertising or analytics tracking.

Requests to Terra’s hosting service and external image, map and data providers may transmit network information such as your IP address. Those services process requests under their own policies. Public source links and image credits identify providers. Never put sensitive personal information in research questions or notes.

Images, geography and craft.

Earth surface: NASA Blue Marble Next Generation, a static reference image. Close-zoom imagery: Esri World Imagery, Maxar, Earthstar Geographics and the GIS User Community, a published mosaic rather than a live image. Forest detail: UMD / Global Forest Watch, 2000 canopy cover and 2025 loss. Named forest markers require a Wikidata reported area of at least 50 km²; major forest regions are retained separately. Coverage and reported areas may be incomplete. Additional textures: Three.js and three-globe examples. Relief, landing-page orbits and symbol sizes are illustrative.

3D: Three.js. Orbital propagation: satellite.js. Geography: Natural Earth / world-atlas. Typography: Space Grotesk under SIL Open Font License and system serif fonts. Icons: Lucide.

Photographs come from scientific institutions, Wikimedia Commons or vessel profiles where available. Credits and source links accompany images. The Terra logo is generated illustrative artwork, not satellite observation. Source titles, place names and original documents may retain their original language.

Maritime attribution: Fintraffic / digitraffic.fi, CC BY 4.0; AISHub; data distributed by Norwegian Coastal Administration under NLOD; Data delivered by BarentsWatch; NOAA / NDBC. Original source attribution stays with each report.