Black-Box Audit of Current Statistics in the IARPA.org Planetary Atlas

Full report

On this page

Executive summary

This audit was conducted as a strictly black-box review using only browser-visible behavior, accessibility-visible text, public pages, and public JSON artifacts exposed by IARPA.org. The strongest confirmed finding is that the site publicly defines Current Statistics as a read-only research mode that combines active borders, preserved Factbook data, annual historical series, and simulation world-state changes, and that it exposes only two world modes in the Globe UI: Current Statistics and Simulation Earth. I did not find a separate “Research Layer” selector in the observable world-mode interface, which is consistent with the product definition you supplied.

The public record also confirms several target behaviors that matter for your audit. The Atlas states that Current Statistics presents “up to 20 front-facing entities nearest the center of the globe” as connected summary cards, and that selecting a card or point opens the complete dossier. It also states that future source values are marked as carried forward rather than shown as fresh observations. Those claims are reinforced by the About page, which says successor states in fictional branches can aggregate prior jurisdictions while keeping source years and uncertainty visible.

Where the audit becomes more cautious is functional verification. In the accessible fallback visible to this browser, the homepage loaded in Simulation Earth state, not Current Statistics: the page showed the Simulation Earth “Selected synthetic cell” panel, while the Current Statistics status fields in the mode panel were still placeholders represented by dashes for political state, statistics source, entities, and shown. Because the button itself was not operable through this harness and the globe canvas behavior was not directly inspectable, I could not directly verify switch latency, card recalc during rotation, border persistence after toggling outlines off, or DOM-stability during pointer interaction. Those remain unverified, not disproven.

The publicly exposed data artifacts are substantial enough to support a strong architecture recommendation. The site publishes a 2025 Factbook manifest with 260 entries and 258 globe points, a historical manifest with 275 exact country-value series spanning 1990–2025, and per-entity JSON records with stable IDs, hashes, coordinates, and source ordinals. However, the currently exposed globe-points JSON contains compact fields such as population, area, capital, government type, and route, but does not expose GDP or life-expectancy keys in that summary payload. That gap is important because your required basic card contents include GDP and a major health indicator. In other words, the public evidence suggests the data foundation exists, but the compact Current Statistics card layer likely needs an enriched summary index or deterministic joins at runtime to satisfy the required card payload without excessive per-card fetches.

My overall conclusion is that the Atlas’s stated product model already points in the right direction for your requirements, especially around unified research mode, “20 nearest front-facing entities,” dossier opening, and carried-forward labeling. The biggest open risks are not conceptual; they are interactional and state-management risks: immediate mode activation, empty-state diagnostics, border-layer override on mode switch, stable card identity across ordinary updates, and an explicit successor-state aggregation model that can collapse the world into a single 2090 China dossier without implying observed 2090 statistics.

Test environment and evidentiary basis

Environment used

The Atlas opened in a browser-accessible fallback state with the explicit message that WebGL was unavailable, and that the site was using an accessible two-dimensional globe renderer while preserving research routes and simulation controls. That matters because all observations below come from the accessible/fallback experience and public data endpoints, not from a WebGL canvas inspection.

The homepage visibly exposed the world-mode control with exactly two mode buttons, “Current Statistics” and “Simulation Earth,” plus adjacent Current Statistics metadata placeholders for political state, statistics source, entities, and shown. The same page also exposed view controls including country outlines, event symbols, and information density, plus keyboard shortcuts indicating that R and S switch modes.

Public artifacts inspected

I inspected the public home page and help/accessibility/privacy/about pages, the Factbook 2025 landing page and source/method page, the Historical Factbook landing page plus China historical series and historical JSON, the China Factbook page and JSON, the China institutional profile, the public events JSON, and the public globe-points JSON. Those artifacts exposed release numbers, generated timestamps, counts, stable hashes, routes, coordinates, and read-only classifications.

The public network-visible data is explicitly identified as read-only and reviewable. The Factbook import page states that the integration preserves the exact uploaded CSV, source records, source cells, deterministic field anchors, per-row hashes, per-entry hashes, and a machine-readable manifest. The historical source/method page likewise states that the complete historical CSV is preserved byte-for-byte and that the daily replay uses interpolation only as a visual guide between annual values.

Limitations of this audit

This audit did not inspect source code, repository files, build logs, the server filesystem, or hidden/private APIs. It also did not directly manipulate the globe canvas or inspect WebGL draw calls. As a result, any statement below about switch timing, rotation-time recomputation, DOM replacement, or border drawing order is clearly labeled as either observed, partially evidenced, or proposed/inferred from public behavior and data. That distinction matters because the site’s public material is very rich, but it still does not substitute for a true interactive session with a browser devtools recording. The site itself also signals that some technical artifacts remain behind access control, such as the importer source deployment copy.

Observed black-box findings

Current Statistics activation findings

The observable UI strongly supports the intended information architecture. The home page states that Current Statistics “combines the active political map with preserved Factbook and historical source values,” and the instrument guide expands that to “active borders, the preserved Factbook snapshot, annual historical series, and simulation world-state changes.” The About page adds that Current Statistics combines dated Factbook material, historical series, institutional records, and the political boundaries active at the selected date, and that successor states in fictional branches can aggregate prior jurisdictions while source years and uncertainty remain visible. That is a close conceptual match to your project definition.

The same public text confirms several specific behavior expectations: up to 20 front-facing entities nearest the center receive connected summary cards; selecting a card or point opens the complete dossier; and future source values are marked as carried forward rather than presented as new observations. Those product claims are observable and should be treated as the public contract for Current Statistics mode.

What I could directly observe on first load, however, was Simulation Earth, not Current Statistics. The page showed a “SIMULATION EARTH · FICTIONAL” selected-cell panel instructing the user to switch to Current Statistics for the active political map, while the Current Statistics status fields in the mode panel still displayed placeholders rather than populated state values. That means the accessible page exposes the shell of Current Statistics status reporting even before activation, but in this harness I could not trigger the button to see the population of those fields or the removal of simulation-cell primacy.

For audit purposes, that leads to a precise verdict: the Atlas publicly claims the right Current Statistics behavior, but this browser session could not directly verify immediate activation behavior. In a release gate, that should be treated as a must-verify interaction, not as a documentation-only requirement. A passing build must populate mode status values and replace Simulation Earth’s marker-first interface on the same interaction cycle that activates Current Statistics, without relying on a hidden later refresh.

Border-rendering findings

The public UI exposes a dedicated “Country outlines” layer toggle, and the globe text says borders are a simplified orientation layer, with international administration shown as dotted and disputed status shown as dashed. The simulation clock text also states that political outlines and country names follow the exact simulated day. This is good evidence that border rendering is both date-aware and layer-driven.

The site’s language is especially important for your fictional-branch requirements. The About page says Current Statistics uses “the political boundaries active at the selected date,” and that successor states in fictional branches can aggregate prior jurisdictions while keeping source years and uncertainty visible. Read together, those claims imply that fictional branches are expected to alter political composition and active boundaries, not merely overlay statistics atop unchanged modern borders.

What remains unverified in this harness is the exact draw-order and override logic when switching from Simulation Earth to Current Statistics after the user has disabled country outlines in Simulation Earth. The requirement you supplied is correct and necessary: if Current Statistics is the active political-research mode, its borders must not disappear simply because a Simulation Earth cosmetic boundary toggle was previously disabled. The public UI exposes enough shared layer controls that this is a credible risk area, but I could not directly reproduce it here.

Country-borders-versus-coastlines distinction is also only partially evidenced. The UI clearly distinguishes border classes like international administration and disputed status, but it does not publicly expose a textual style contract for coastlines versus sovereign borders. In the absence of a direct screenshot of Current Statistics active, the correct audit position is that border distinctness is product-critical, publicly implied, but not fully verified.

Card, point, and leader-line findings

The public card/point model is strongly suggested by both the instrument guide and the points JSON. The guide says that in Current Statistics, up to 20 front-facing entities nearest the center receive connected summary cards, and that selecting a card or point opens the complete dossier. The points JSON shows a compact per-entity orientation record with stable ID, title, approximate latitude/longitude, snapshot year, area, population, capital, government type, route, and related profile route. That is almost exactly the shape expected for a fast card layer or point anchor layer.

There is, however, an important gap between the public points payload and your required compact card contents. The points JSON exposes population, area, capital, government type, and summary text, but a string search of that payload did not find gdp or life_expectancy keys. In contrast, the full Factbook entry JSON and page for China do include population, area, life expectancy, and real GDP per capita, while the historical record JSON includes annual GDP fields for years with data. This strongly suggests that a production-ready Current Statistics card cannot rely on the compact points file alone if it must display GDP plus a major health indicator on every summary card.

The public data also reveals an edge case that affects card/point integrity. The Factbook manifest lists 260 entries but only 258 coordinates, and the separate globe-points file likewise contains 258 records. That means at least two Factbook entries lack geographic point output in the public orientation layer. For Current Statistics, that is acceptable only if those non-pointable entries are excluded from visible top-card selection or shown through an explicit diagnostic/fallback rule. It is not acceptable for the mode to silently include a visible entity without a point anchor or to fail into an empty globe.

Leader-line behavior is publicly documented for selection states in the globe generally: dots or events draw a leader line to a compact label, and offscreen event pointers terminate on the visible globe edge. That is useful precedent for how Current Statistics leader lines should behave, even though I did not directly inspect Current Statistics leader-line drawing in motion.

Dossier information-architecture findings

The public dossier/data foundation is richer than the summary-card layer. A full Factbook record JSON contains stable entry identity, snapshot year, source metadata, key facts, coordinates, and category-and-field structures with source-record ordinals and per-row hashes. A historical record JSON contains year ranges, field-presence counts, raw annual rows, numeric annual rows, and stable row hashes. An institutional profile page adds a distinct research cutoff, evidence labels, confidence vocabulary, and source/report routes. That means the public data architecture already supports a three-layer dossier model: political entity identity, statistical baseline and history, and institutional/provenance context.

The China institutional profile is especially instructive. It shows a visible research cutoff of 2026-07-20 UTC, an explicit warning to recheck current-sensitive facts after that date, a visible evidence-label vocabulary, and a separate statement that the Factbook snapshot contributes 158 read-only fields without changing the profile’s claims or cutoff. That is exactly the kind of temporal and provenance honesty your required dossier needs, except your dossier should make those distinctions even more prominent when the selected political state is fictional or successor-aggregated.

The public data therefore supports a dossier that can show: simulation date, political-state date, source year, whether a value is observed or carried forward, present entity identity, predecessor jurisdictions, institutional context, source links, and field-level provenance. What is not yet publicly evidenced is a unified, branch-aware dossier JSON that merges all of those into one Current Statistics record for a fictional successor state such as a one-entity 2090 China. Within the public endpoints I discovered, I found factual snapshot records, historical annual records, event context JSON, and points JSON, but not a published branch-world-state schema for fictional current political composition.

Failure reproduction and network observations

Reproduction steps for empty-mode and missing-border failures

Because I could not directly activate Current Statistics in this harness, the following are the best black-box reproduction scripts I would use in a real browser session to confirm the highest-risk failures. They are grounded in the observable UI structure and public data model.

Candidate empty-mode failure reproduction

1. Open the Atlas home page and wait for the globe and mode panel to render. Confirm that the page starts in Simulation Earth and that the Current Statistics metadata fields are placeholders.
2. Set the globe to a date that should still have entity coverage within the public source period, such as 2025-12-31 CE or a date close to the current snapshot boundary. The public source period ends at 2025-12-31 for historical annual evidence, so this is a valid control date.
3. Activate Current Statistics through the mode button or the documented keyboard shortcut. The instrument guide states that R and S switch modes.
4. Observe whether, within the same interaction cycle, the mode panel populates political state, statistics source, entities, and shown; whether summary cards appear; and whether the primary interface ceases to be marker-first. If the globe remains visually empty, or if the mode panel remains at placeholder dashes without a readable diagnostic, log a failure. This is the clearest black-box manifestation of “data missing or not loaded should show a diagnostic, not an empty globe.”

Candidate missing-border failure reproduction

1. Start in Simulation Earth with the outlines toggle visible. The page exposes a “Country outlines on” control in the shared layer UI.
2. Turn country outlines off.
3. Without re-enabling them, switch to Current Statistics.
4. If Current Statistics opens without visible current political borders, log a failure. The correct behavior is that Current Statistics asserts its own border layer even if a Simulation Earth boundary toggle was previously disabled, because Current Statistics is defined publicly as the active political map plus preserved source context.

Candidate one-entity successor-state failure reproduction

1. Create or load a fictional branch dated in 2090. The site publicly states that any continuation past 2025-12-31 or user intervention creates an explicitly fictional branch.
2. Apply a branch state in which China absorbs every prior country and territory and becomes the sole global political entity. The About page says successor states can aggregate prior jurisdictions in fictional branches.
3. Open Current Statistics.
4. Expected black-box results: one active entity in state, one visible summary card, one point anchor, one dossier, no former internal country borders, and obviously carried-forward or source-dated statistical labeling rather than implied observed 2090 data. If any predecessor borders remain or if more than one active entity appears, log a composition/border-dissolve failure.

Network and data-loading observations

The Atlas exposes a notably reviewable first-party data surface. The Factbook 2025 manifest reports release 1.6.34-wip, generated at 2026-07-25T20:50:09Z, with a 2025 snapshot year, 260 entries, 32,594 source records, 258 coordinate entries, and stable per-record routes and hashes. The China Factbook record JSON and historical China JSON likewise expose generated timestamps, stable IDs, coordinates, source ordinals, and raw-versus-numeric values.

The derived globe-points JSON appears to be a later build artifact from that source set. It reports release 1.6.38-wip, generated at 2026-07-26T20:05:55Z, with source_release 1.6.34-wip, the shared source SHA-256, and 258 point records. That strongly suggests a deterministic derived artifact specifically for globe orientation/card anchoring.

The historical side is equally explicit. The historical manifest reports an 810,253-byte CSV preserved byte-for-byte, 8,932 rows, 36 annual years from 1990 to 2025, and 275 exact country values, while the historical China JSON exposes field-presence counts showing that some metrics are absent for some years. For China, annual GDP exists for 32 years, life expectancy for 35 years, military percent GDP for 25 years, and internet users for only 11 years. That missingness is important because any successor-state dossier or current-state card must surface whether a shown value is observed, interpolated, carried forward, or unavailable.

The site’s network posture is intentionally quiet. The privacy page says there is no account requirement, no advertising, no analytics, no remote fonts, and no embedded third-party forms, and that the international atlas search runs with a local on-page index rather than an external search provider. That means most observable data-loading issues are likely to involve first-party JSON/assets and client-side state composition, not third-party network dependencies.

A final network/data-loading observation is important for performance and packaging: the public globe-points file is a manageable summary payload, while the field-index JSON linked from the Factbook source page was too large for this retrieval tool to fetch in one pass. That does not prove a user-facing problem, but it does reinforce the architectural value of keeping Current Statistics cards on a compact summary index rather than issuing many large record fetches during globe rotation.

Proposed target architecture and algorithms

Proposed entity-composition data model

The public data suggests four distinct ingredients already exist: a factual snapshot layer, a historical annual layer, an institutional profile layer, and a branch-aware political-state concept. The correct Current Statistics model should therefore not be “a list of cards.” It should be a composed world-state object with two separate concerns: all active political entities in the selected branch/date, and the visible top-20 subset for the current camera view. This separation is required because the public product contract already says Current Statistics composes active borders and world-state changes, while only the nearest 20 front-facing entities receive cards.

A practical entity object should include: a stable world_state_entity_id; current display name; entity status such as observed, renamed, merged, annexed, restored, dissolved-successor, or fictional aggregate; branch ID; simulation date; political-state effective date; statistical source year by field group; predecessor jurisdiction IDs; boundary geometry ID; anchor point; visible-card eligibility; card summary payload; dossier payload; provenance flags; and a compact “value status” map that distinguishes observed, historical, interpolated, carried forward, aggregated, overridden, or missing. This model is an inference from the public structure of stable IDs, record URLs, routes, source ordinals, hashes, yearly statistical rows, and explicit research cutoffs.

The compact summary payload should not depend solely on the current points JSON, because that file does not publicly include GDP or life expectancy. Instead, the product should materialize or cache a Current Statistics entity-summary index per active entity containing the exact basic card fields you require: name, source year, population, GDP, area, one major health/social indicator, and status label. This can be assembled from the Factbook snapshot, historical series, and institutional metadata once per entity-state change rather than every visual frame.

Proposed successor-state aggregation rules

The site already states publicly that successor states in fictional branches can aggregate prior jurisdictions while keeping source years and uncertainty visible. The missing piece is a published rulebook for how that aggregation should happen. My recommendation is to make the aggregation regime explicit and field-specific.

For stock metrics, the rule should be additive unless a field explicitly states otherwise. Population, area, and GDP should be summed across represented predecessor jurisdictions when the successor completely replaces them in the selected branch state. GDP per capita should not be summed or averaged directly; it should always be recalculated from aggregated GDP divided by aggregated population, with the quotient labeled as “recalculated from aggregate GDP and population.” This matches your required 2090 China case and avoids displaying impossible inherited per-capita values. The historical and Factbook sources already expose population, area, and GDP-related values with numeric structure suitable for this.

For rate metrics, the rule should be denominator-aware and documented in the dossier. My recommended hierarchy is:

• population growth rate: convert each predecessor’s rate into implied annual change using its population denominator, then recompute against total population;
• birth rate and death rate: convert to implied births/deaths using each predecessor population denominator, sum those implied counts, then divide by aggregated population and rescale to the published unit;
• infant mortality: convert to implied infant deaths per birth cohort using births as denominator where possible, otherwise fall back to a documented approximation and downgrade confidence;
• internet use: convert to implied internet-user counts using population denominator, then divide by aggregated population;
• unemployment: use labor-force denominator if available from the full Factbook field set; if unavailable, show “aggregation approximation” or withhold the value rather than pretending population-weighting is exact;
• military expenditure as percent of GDP: aggregate military spend by converting each predecessor percent to implied spend with that predecessor GDP denominator, sum implied spend, and divide by aggregated GDP. The public Factbook records and historical rows already distinguish which fields are raw values and which are numeric machine parses, making this kind of derived presentation feasible.

For non-additive descriptive fields, the successor entity should not silently inherit one predecessor’s capital, government type, or leader-route. Instead, the dossier should show one of three states: explicitly overridden by branch logic, carried from a named predecessor, or “not yet assigned in this branch.” This is particularly important for a complete world replacement like one-China-2090, where the political composition has changed radically but the inherited factual baselines remain source-dated.

For predecessor lineage, the dossier should list every represented predecessor jurisdiction with stable names and dates, drawing from the historical manifest’s exact-name series, which already includes discontinued entities such as Czechoslovakia and the Netherlands Antilles. In the one-China-2090 case, the predecessor list would be world-scale and should therefore support truncation with an “expand all predecessors” control, not silent omission.

Border-merging and internal-border suppression rules

The public contract implies that active political boundaries in the selected branch/date are what matter, not baseline modern boundaries. Therefore, when two adjacent predecessor polygons belong to the same active current entity, their internal shared boundary must be suppressed in Current Statistics. This is not cosmetic; it is the core visual test of whether the political composition has truly changed.

The correct dissolve rules are straightforward in plain language. First, build the set of all polygons currently assigned to each active entity. Second, merge all polygons that belong to the same active entity, including enclaves and exclaves. Third, remove every shared interior edge between polygons of the same active entity. Fourth, render only the exterior shell and coastline, plus any still-active disputed or internationally administered boundaries that remain distinct in the selected world state. The public UI already distinguishes ordinary outlines from dotted and dashed political edge classes, so the same classing should remain after dissolve.

In the required 2090 one-China test case, the result should be visually simple: there is only one active political entity, so there are no inter-entity borders left to render. Former country borders must disappear entirely. The only remaining lines should be the exterior political outline where relevant and coastlines. Any surviving internal country lines would be a hard failure of border suppression. This is consistent with the site’s own claim that fictional branches can aggregate prior jurisdictions in Current Statistics.

Center-nearest entity-selection algorithm

The public product text already gives the core rule: show up to 20 front-facing entities nearest the center of the globe. The simplest robust implementation in plain language is as follows. Begin with the full active entity set for the selected world state, not merely today’s visible cards. Use each entity’s geographic anchor point, which the public data already exposes as an approximate orientation latitude/longitude. Convert those anchors into globe-space vectors. Remove any entity whose anchor lies on the back hemisphere relative to the current camera. Project the remaining anchors into screen space. Measure each projected anchor’s distance to the visual center of the globe viewport. Sort ascending by that distance and choose the first 20.

To keep the UI stable during ordinary rotation, the selection should use hysteresis. In plain language: if entity 20 and entity 21 are nearly tied, do not swap them instantly on every tiny camera movement. Require a small superiority margin before a new entity displaces an already-visible one. This is not directly stated by the site, but it is strongly implied by the public promise of connected cards and by the need to avoid DOM replacement during ordinary updates.

Entities without valid anchor points must never enter the visible-card set silently. The public manifests already show that not every Factbook entry has a coordinate-derived point. Therefore the selection algorithm needs an explicit fallback rule: either exclude such entities from visible-top-20 selection and surface them in a diagnostic “inactive for globe placement” list, or assign a branch-specific synthetic anchor with visible provenance. What it must not do is produce a card with no point or produce an unhelpful empty field.

Card-placement and collision-avoidance algorithm

The public UI is panel-heavy. On the same screen, the Atlas can show the world mode panel, timeline controls, playback controls, HUD, consequences, dossiers, event blurbs, and movable panels. The site explicitly says panel title bars can be dragged and that HUD can show, move, or collapse panels. Because of that, card placement cannot be done with static “left safe column / right safe column” constants. It must compute against live occlusion rectangles derived from the current layout.

A workable plain-language algorithm is this. After selecting the visible top 20 entities, estimate each card’s preferred side from the projected point’s x-position. Split cards into left and right queues, but rebalance if one side becomes overloaded. Build a set of forbidden rectangles from the current UI surfaces: mode panel, date controls, playback controls, HUD, any open dossier, and event/callout surfaces. Then place cards top-to-bottom in the left and right lanes using a score that penalizes overlap with forbidden rectangles, overlap with already-placed cards, excessive leader-line length, and crossings with existing leader lines. If no candidate position is collision-free, progressively shrink card density, then collapse low-priority cards into a compact mode-specific list rather than drawing them over controls. This inference is driven by the site’s movable interface model and your explicit avoidance requirements.

Leader lines should connect point-to-card using the shortest non-occluding route available, with endpoint clamping so that offscreen or nearly offscreen points terminate on the visible globe edge rather than leaving floating card stems in empty UI space. The public globe behavior already uses edge-terminated offscreen pointers for events, which is a good precedent for Current Statistics entities as well.

Stable-element and interaction requirements

Your DOM-stability requirement is correct and should be treated as non-negotiable. The current public text does not describe DOM replacement behavior, so this section is a proposed release rule rather than an observed site fact. The right behavior is: when an entity remains in the visible set and only its card position or displayed values update, the card node and point node must keep the same stable identity and must move via transforms or attribute updates rather than being destroyed and recreated. This avoids separating pointer-down and pointer-up across refreshes. The mode already promises that selecting a card or point opens the complete dossier, so that interaction must survive ordinary visual updates.

Concretely, stable UI elements need stable keys rooted in a branch-aware entity ID, not in array index or render order. Recalculation of the visible-top-20 list should preserve node identity for entities that remain visible. Replacement is acceptable only when an entity truly enters or exits the visible set, or when its underlying active-entity identity changes because the world state changed. The dossier route opened from a card and the dossier route opened from its point must be the same route or same state object.

Dossier information architecture

A passing Current Statistics dossier should be organized into a small number of highly legible blocks.

The identity block should show current entity name, branch name or fictional-state note, simulation date, political-state effective date, and a one-line plain-language status such as “fictional aggregate successor” or “observed entity, carried-forward statistics.” The site already publicly distinguishes factual current statistics from fictional branches and warns that future values are carried forward rather than newly observed.

The provenance block should then separate three dates: statistical source year, institutional profile research cutoff, and branch/political-state date. The China institutional profile already demonstrates that the site can surface a research cutoff separately from the Factbook snapshot year, and the historical/factbook sources already surface their own year boundaries. That separation should be mandatory in the dossier header, not buried in footnotes.

The quantitative block should show population, area, GDP, recalculated GDP per capita, life expectancy, population growth, birth rate, death rate, infant mortality, internet use, unemployment, and military expenditure percent of GDP. Each field should have a compact badge such as Observed, Historical, Interpolated, Carried forward, Aggregated, Overridden, or Missing. The historical record JSON’s raw/numeric split and field-presence map already demonstrate the need for field-level provenance rather than a single dossier-wide status label.

The lineage block should list prior jurisdictions represented, with expansion UI for long lists. In the one-China-2090 case this must be exhaustive in content, even if collapsed by default for readability. Public historical-series coverage across 275 exact names shows that the site already preserves predecessor-like exact country values across discontinued and renamed entities.

The source block should provide deep links to the full Factbook entry, the historical series, the related institutional profile where present, and the specific full report or profile JSON if available. The public data model already includes route, record URL, related profile route, report routes, and stable field anchors, making this practical without private data.

Source-year and temporal-honesty rules

The Atlas already publicly commits to temporal honesty. The historical source page says the observed evidence period ends at 2025-12-31 CE, and the home page says everything after that is a projection or explicitly fictional branch. The Current Statistics guide also says future source values are visibly marked as carried forward rather than presented as new observations.

The strict rule for a 2090 one-China branch should therefore be: never display a bare “2090” statistical year for inherited values unless a specific field was actually overridden with 2090-native branch data. Instead, display the simulation date as 2090, the political-state date as 2090, and the statistical source year as 2025 or whatever the latest field-specific source year actually is. If a field is inherited from the 2025 Factbook snapshot but internally states “2024 est.” in the preserved content, the dossier should preserve that fact rather than flatten everything to a false “2025 observed” presentation. The current China Factbook record already demonstrates that different fields within a 2025 snapshot can themselves point to 2024 estimates.

The same rule should apply to institutional context. If the dossier includes institutional profile material with a 2026 research cutoff, that cutoff remains separate from the statistical source years and separate from the 2090 fictional branch date. Nothing in the dossier should imply that a 2026 institutional research cutoff somehow validates observed 2090 geopolitical composition.

Desktop, mobile, keyboard, touch, and reduced-motion requirements

The site’s accessibility page gives a clear baseline: keyboard use, visible focus, text enlargement, mobile reflow, reduced motion, semantic headings, printable reports, and a non-WebGL fallback are all intentional product goals. Reduced-motion preferences are supposed to stop automatic globe rotation; the globe supports pointer, touch, wheel, pinch, and keyboard controls. Those are not peripheral claims here; they directly constrain how Current Statistics cards and recomputation should behave.

On desktop, the mode switch should preserve pointer stability, dynamically avoid open panels, and allow keyboard focus to move from world mode to card to dossier without crossing hidden or offscreen elements. On mobile, the site’s own reflow standard means cards must compress into a smaller visible set, scrollable side sheet, or focus-driven single-column layout rather than trying to draw 20 floating cards over a narrow screen. The accessibility page explicitly says pages are designed to reflow at narrow widths and high zoom, and that long identifiers and links should wrap instead of forcing horizontal scroll.

For keyboard, mode switching, point selection, card activation, dossier opening, and dossier closing must all be operable without a pointer. The home-page help already documents keyboard shortcuts including arrow keys, zoom, space for playback, and R/S for mode switching. Current Statistics should extend that same discipline to entity navigation.

For touch, point hit targets and card hit targets must be large enough that leader lines and animated recalc do not interfere with touch-up on the intended target. For reduced motion, auto-rotation should stop, but manual drag/keyboard rotation must still trigger visible-entity recomputation and card repositioning. Reduced motion is a motion preference, not a request to freeze Current Statistics into stale spatial state. That requirement is an inference anchored in the accessibility commitment to reduced-motion behavior plus the fact that Current Statistics is center-nearest and front-facing.

Acceptance tests and remediation backlog

Given/When/Then acceptance tests

The following acceptance tests translate the public product contract plus your audit brief into black-box release criteria. The tests are especially motivated by the publicly visible two-mode UI, the “20 nearest front-facing entities” contract, the carried-forward labeling requirement, the existence of 258 globe points versus 260 snapshot entries, and the site’s explicit fictional-branch model.

1. Given the Atlas is open in Simulation Earth, when the user activates Current Statistics, then simulation-cell markers stop being the primary interface in the same interaction cycle.
2. Given the Atlas is open in Simulation Earth, when the user activates Current Statistics, then current political borders are visible immediately without waiting for a later hidden refresh.
3. Given Current Statistics activates successfully, when the mode panel updates, then political state, statistics source, entity count, and shown count are populated with non-placeholder values.
4. Given Current Statistics activation cannot complete because required entity data is absent, when the mode switch occurs, then the interface shows a visible diagnostic message rather than an empty globe.
5. Given Current Statistics activation encounters malformed entity data, when the mode switch occurs, then the diagnostic names the failure category and preserves the rest of the UI.
6. Given country outlines were disabled in Simulation Earth, when Current Statistics activates, then current borders appear anyway.
7. Given Current Statistics is active at a date with multiple active entities, when the globe is stationary, then only the 20 nearest front-facing entities are carded.
8. Given the selected world state contains fewer than 20 active front-facing entities, when Current Statistics is active, then only the available entities are carded and no placeholder ghost cards appear.
9. Given an entity is shown in Current Statistics, when its card is visible, then a geographic point for that same entity is also visible.
10. Given an entity is shown in Current Statistics, when its point and card are visible, then a leader line connects that exact point and that exact card.
11. Given the globe rotates slightly, when the set of nearest front-facing entities does not materially change, then the existing entity cards keep stable DOM identity.
12. Given the globe rotates far enough to change the visible top 20, when one entity leaves and another enters, then only the entering/leaving entities receive new DOM nodes.
13. Given a visible entity remains in the top 20, when its card is repositioned, then pointer-down and pointer-up on that card still resolve to the same entity interaction.
14. Given Current Statistics is active, when the user clicks a card, then the same dossier opens as if the user clicked that entity’s point.
15. Given Current Statistics is active, when the user clicks a point, then the same dossier opens as if the user clicked its card.
16. Given a dossier is open for an entity, when the globe rotates but that entity remains visible, then the open dossier stays bound to that entity.
17. Given cards are placed on screen, when there is room on both sides of the globe, then cards distribute broadly across left and right columns rather than clustering on one side.
18. Given the world mode panel is open, when cards are laid out, then no card overlaps the world mode panel.
19. Given date controls are open, when cards are laid out, then no card overlaps the date controls.
20. Given playback controls are visible, when cards are laid out, then no card overlaps the playback controls.
21. Given HUD panels are visible, when cards are laid out, then no card overlaps any visible HUD rectangle.
22. Given a dossier side panel is open, when cards are laid out, then no card overlaps the dossier presentation area.
23. Given event blurbs or related overlays are visible, when cards are laid out, then no card overlaps those overlays.
24. Given more than 20 active entities exist in the current world state, when the user rotates the globe, then the visible top 20 are recalculated continuously from the current camera and not held stale.
25. Given an entity hovers near the edge of the front hemisphere, when tiny camera changes occur, then hysteresis prevents card flicker and unnecessary enter/exit churn.
26. Given Current Statistics is active in a fictional branch after 2025, when a card shows values, then the UI labels them as carried forward, aggregated, overridden, or otherwise provenance-qualified rather than implying direct observation in the branch year.
27. Given a dossier is open for a fictional successor state, when statistics are displayed, then the dossier shows simulation date, political-state date, and source year as separate fields.
28. Given a dossier is open for a fictional successor state, when a metric is aggregated, then the dossier labels the metric as aggregated and shows the method summary.
29. Given a dossier is open for a fictional successor state, when GDP per capita is shown, then it is recalculated from displayed aggregate GDP and population rather than inherited verbatim.
30. Given a dossier is open for a fictional successor state, when rate metrics are shown, then the weighting method is documented at field or section level.
31. Given a successor state represents prior jurisdictions, when its dossier opens, then all predecessor jurisdictions are listed or accessible through an expand control.
32. Given the branch is the required 2090 one-China world, when Current Statistics activates, then the underlying active-entity count is exactly one.
33. Given the branch is the required 2090 one-China world, when the globe centers that entity, then exactly one basic card is visible.
34. Given the branch is the required 2090 one-China world, when the globe centers that entity, then exactly one geographic point is visible.
35. Given the branch is the required 2090 one-China world, when the user opens the entity, then exactly one current dossier is addressable for that world state.
36. Given the branch is the required 2090 one-China world, when borders render, then former country borders are absent and only the exterior political boundary or coastline remains.
37. Given the branch is the required 2090 one-China world, when the dossier shows population, area, and GDP, then each equals the aggregate of represented members according to the documented composition set.
38. Given the branch is the required 2090 one-China world, when the dossier shows the source year, then it explicitly states that the values were carried forward from 2025 or another latest field-specific source year and were not observed in 2090.
39. Given the user prefers reduced motion, when Current Statistics is active, then automatic rotation stops but manual rotation still recalculates visible entities.
40. Given the user navigates with a keyboard only, when they switch to Current Statistics, then they can reach the visible cards, points, and dossier controls without a pointer.
41. Given the Atlas is used on a narrow mobile viewport, when Current Statistics is active, then cards reflow into a non-overlapping mobile presentation rather than clipping behind panels.
42. Given a visible active entity lacks a valid geographic anchor, when it would otherwise enter the top-20 visible set, then the UI excludes it from on-globe placement and surfaces a diagnostic or fallback treatment.
43. Given a field is unavailable for a displayed entity, when its card or dossier renders, then the field is marked missing or not available instead of omitted without explanation.
44. Given the user toggles from Current Statistics back to Simulation Earth and back again, when the second Current Statistics activation completes, then the same world state is recomposed deterministically with stable visible counts and no residue from simulation markers.
45. Given a public institutional profile exists for the entity, when the dossier opens, then the dossier links to that profile and shows its separate research cutoff.
46. Given a public Factbook and historical record exist for the entity, when the dossier opens, then source links to both are visible from the dossier.

Prioritized remediation backlog

Highest priority

The first priority is to create a Current Statistics entity-summary index that contains the complete compact card payload required by the product contract. Public globe-points data is compact and useful, but it does not publicly expose GDP or life expectancy keys, while the full dossier data clearly contains richer fields. Without an enriched summary layer, the product risks either under-informing cards or issuing too many heavy lookups during rotation.

The next highest priority is to implement an explicit mode-activation completion state with visible diagnostics. The public mode panel visibly reserves fields for political state, statistics source, entities, and shown, but the accessible shell also demonstrates how easily placeholders can persist. Activation should therefore have a hard success/failure contract with a user-readable message whenever entity composition or border data is not ready.

A third top-tier item is to decouple Current Statistics border visibility from any Simulation Earth outline toggle state. The public UI exposes a shared outlines control, while the public definition of Current Statistics makes active borders fundamental rather than optional. Mode switch should therefore reassert political-border visibility whenever Current Statistics is entered.

A fourth top-tier item is to formalize branch-aware stable IDs and hysteresis for visible-entity selection. The public product promise of center-nearest carding combined with ordinary user rotation will otherwise produce card jitter and click instability.

Important follow-on work

Implement a branch-aware successor-state aggregation engine with field-class rules, method labels, and predecessor lineage output. The About page already promises that successor states can aggregate prior jurisdictions while source years and uncertainty remain visible; the backlog item is to make that promise fully deterministic and dossier-visible.

Implement boundary dissolve and internal-border suppression as a first-class topology step in political-state composition, not as a cosmetic afterthought. This is essential for the one-China-2090 case and for any merger, annexation, unification, or complete world replacement.

Implement dynamic occlusion-aware card placement tied to the live HUD/panel layout. Because the site explicitly supports movable panels and focus-globe layouts, any hard-coded card-safe zones will eventually fail.

Strengthen dossier provenance UI so that simulation date, political-state date, statistic source year, institutional cutoff, and field status badges are always visible at first glance. The public components for this are already present in separate pages and JSON formats; Current Statistics needs to unify them.

Quality and maintainability work

Add a browser-visible debug/status overlay for Current Statistics composition in non-production or shareable-debug mode, showing counts for active entities, visible entities, missing anchors, missing metrics, and border-geometry status. Because the privacy posture is explicitly no-account/no-analytics, a local visible diagnostics approach is a better fit than hidden telemetry.

Publish a public branch/world-state schema or at least a shareable exported snapshot format if the project wants external auditors to validate successor-state composition more deeply. The existing public manifests and records are already unusually reviewable; extending that practice to branch state would materially improve auditability.

Known unknowns and required evidence

Several critical points remain unknown because this was a strict black-box audit limited to public pages and public JSON in an accessible fallback session.

I still need a real interactive browser capture of switching from Simulation Earth to Current Statistics, ideally with timestamps or a screen recording, to verify whether activation is immediate and whether simulation markers truly stop being the primary interface without a delayed repaint. The public text says what should happen, but this harness could not directly execute the button path.

I also need evidence of Current Statistics in an active fictional successor branch, especially the one-China-2090 test case. The public site says such successor-state aggregation is supported conceptually, but I did not locate a published branch-world-state JSON endpoint among the discovered public records.

For border auditing, I need at least one screenshot or recording showing country outlines toggled off in Simulation Earth, then Current Statistics activated, so border override and draw-order behavior can be judged directly. The public UI exposes the ingredients of that scenario, but not the result in this harness.

For stable-element auditing, I need a DOM or accessibility-tree snapshot during interaction, or a slow-motion recording of clicking a card while the globe rotates, to confirm that pointer-down and pointer-up are not separated by DOM replacement. The public data and product contract make the requirement obvious, but they do not prove the implementation.

For mobile and accessibility closure, I need narrow-viewport and reduced-motion evidence of Current Statistics while active, not just general site accessibility commitments. The accessibility page promises reflow, keyboard use, reduced motion, and a non-WebGL fallback, but the audit still needs active-mode proof that cards, points, and dossiers honor those commitments together.