Family Trees Grow Wide, Not Tall

I toggled Fan on this lab and watched six generations of ancestors collapse into concentric rings around my name instead of a hedge of boxes marching off the right edge of the screen. Same people. Same marriages. A different information architecture problem entirely. I built it to answer a narrow craft question: if kinship graphs are really graphs, what does the visualization owe the visitor before you ask them to pan? The demo ships three synchronized views (hourglass Tree, node-link Pedigree, radial Fan) over one public subset, with focus-is-viewport navigation, marriage as a unit, and inspect-before-travel separation between refocusing the graph and opening a profile. The fan chart is ego-centred: my name in the middle, ancestors in seven concentric rings, pill-shaped arc segments with rounded fillets instead of thin radial wedges, pink and grey only where sex is the encoding. That is not a skin on Ancestry. It is an argument about legibility you can toggle against the industry default in the same frame.

When is a family tree an information architecture problem rather than a heritage mood board? When the interface treats pan-and-zoom as the primary navigation skill and nobody has ranked the jobs. Three tasks recur in every product review I read: orient (who surrounds this person?), locate (find one named relative), and edit (add parents, resolve merges, attach sources). A node-link pedigree (boxes, connector lines, root at the top or left) is often optimised for edit while locate and orient pay the width penalty. An ego-centric chart puts one person at the centre and draws kinship outward from that focus. Genealogy products call the whole domain a "family tree," then default to one layout for all three tasks. That is the same attention failure I described in Design Is Attention Control, Not Aesthetics: if nothing decides what you see first, you perform triage yourself, and you leave.

Three Tasks, Three Layouts

The honest product map pairs layout to task, not layout to brand. Fan and hourglass views suit orient and storytelling: depth accumulates in rings or stacked bands while the ego stays centred. Node-link pedigree suits structured edit and print export: boxes align with add-parent flows genealogists already know. Locate is the task the chart alone rarely wins. Search, index sidebars, and record hints do the locative work in Ancestry and FamilySearch; the graph supplies context after the name is found. My lab ships search-to-focus for that reason. It is not a replacement for a hint engine. It acknowledges that find-a-person is often a retrieval problem wearing a tree costume.

I would revise the default-chart argument if a moderated locate task showed pedigree beating fan on time-to-target with real kinship data, including cousin branches and duplicate ancestors. I have not run that study. This essay and lab are a design hypothesis with a testable prototype, not a heuristic audit with pass/fail metrics. Until the tasks are measured separately, recommending fan as the category default is directionally right for orient and wrong as a universal replacement for edit or search.

The Chart Everyone Ships Is the One That Breaks First

Ancestry, FamilySearch, MyHeritage, and Geni all converge on a node-link pedigree as the default mental model. Boxes for people. Lines for parentage. Hints, leaves, and merge prompts hung on that scaffold. It is familiar. It also punishes depth when the product expands every generation to full width instead of collapsing collateral lines. The worst case grows wide: two parents, four grandparents, eight great-grandparents. Many products hide cousins until you ask. The sprawl argument applies to full expansion, not to every pedigree on screen. Draper and Riesenfeld, at the BYU Family History Technology Workshop, argue that fan charts save space for genealogical exploration draper-fan-charts. McGuffin and Balakrishnan analyse why classical graph layouts struggle to use canvas efficiently for multitrees and dual trees mcguffin-genealogy. That is graph-browsing literature, not a spec for filleted pill arcs.

Nielsen Norman Group's tree-testing research nng-tree-testing transfers partially. Label competition and depth fatigue on hierarchical paths apply to pedigree scanning. It does not transfer cleanly to cycles, duplicate ancestors, or emotionally loaded hereditary data. Real users still complain without the academic vocabulary. Comparison threads between FamilySearch and Ancestry recur on a single theme: the collaborative tree is powerful and hard to use uxmag-hierarchy. Ancestry's shaking-leaf hint pattern is brilliant discovery UX built on a brittle canvas. FamilySearch's single shared tree is merge governance on the same canvas. The canvas is the bottleneck when locate and orient are forced through edit-shaped geometry.

There Is Already a Taxonomy. Products Just Ignore It.

Genealogy visualization is not an undiscovered country. It is an understaffed museum basement. McGuffin and Balakrishnan catalogue hourglass charts, bowtie variants, dual trees, and multitree browsing mcguffin-genealogy. Fan charts as radial ancestor rings are a separate line of work draper-fan-charts. Ferdio's Data Viz Project entry ferdio-fan defines the structure: inner circle is the focus; each outward ring doubles the segments. Depth becomes radius, not acreage. MyHeritage adds a sun chart with photo-capable segments and an hourglass pitched at relatives who want past and future in one frame. FamilySearch's fan editorial familysearch-fan uses four gradient colours, one per grandparent line. The literature already knows the menu. The shipping default ignores it.

Above ego-centric charts sits population-scale work the defaults barely ship. Dual trees generalise hourglasses for linked families mcguffin-genealogy. TimeNets timenets-gramps put time on the primary axis. GeneaQuilts geneaquilts matrix thousands of individuals. User studies on hierarchical quantitative charts found sunburst among the least preferred layouts for browsing; treemap worse still woodburn-hierarchy-viz. That evidence cautions against importing generic sunburst or treemap idioms into kinship without a matched task. It does not prove fan beats pedigree for locate. It proves trendy hierarchy widgets underperform when the task is browse-and-compare.

What the Lab Gets Right

The data visualisation work is where the hypothesis becomes inspectable in the interactive demo below. Tree view is an hourglass neighbourhood for orient: parents as a marriage unit above, children below, sibling bands hidden until you ask so the waist stays readable. Pedigree is the ancestor chain for visitors who still think in boxes. Fan follows Draper's radial space-saving logic draper-fan-charts with generation guide rings, seven ancestor depths tuned to the viewport, and filleted arc segments instead of sharp wedges. Filters for sex and parental line dim non-matches rather than deleting structure. Search jumps focus for locate. That pairing, chart for orient plus search for locate, is closer to how NN/g separates encoding from navigation nng-tree-testing than a legend-heavy tree alone.

Interaction matches the viz. Tap a card to refocus; use the profile control to inspect without travelling. Couples stay paired in a marriage unit because splitting spouses across branches is how node-link trees lie about how families are remembered. Sex is encoded with pink or neutral grey plus name labels on every segment, not colour alone; unknown sex stays neutral. That is a demo encoding, not a claim that pink is the right semiotics for every culture or every product. Pan and zoom centre on the pointer; cinema mode and popup window support desktop study. Screen readers get focus announcements and region labels per view, though curved fan labels remain weak for non-sighted locate. I argued in Waiting Is a Design Surface that interstitials deserve intention; here the chart-type switcher lets you compare sprawl against rings without pretending they share one cognition.

Where the Industry (and I) Still Owe Work

Commercial products bury fan and hourglass as secondary tabs when orient deserves first-run parity, with node-link reserved for edit and print. FamilySearch's four grandparent gradients familysearch-fan solve lineage side at the cost of a cognitive four-way split. A genealogist at Organize Your Family History oyfh-fan-praise praised Ancestry's fan beta for clarity while mourning lost leaves and ThruLines. My lab omits hints entirely. That is a demo trade. Merge-conflict UX on a shared tree and emotionally sensitive DNA presentation remind you that chart geometry is not separable from governance or biology: a misplaced line hurts differently when the line is hereditary.

Radial typography remains unsolved at depth: outer-ring names shrink and curve. MyHeritage's sun chart shows what photos on segments buy; I have not shipped that. Neither my lab nor the majors offer TimeNet axes timenets-gramps or quilt-scale matrices geneaquilts for thousand-person reunions. Mobile is honestly secondary here: pedigree and fan carry desktop hints because small screens punish both sprawl and radial labels. A composite cousin scene, patterned on repeated forum reports rather than one verified interview, spent twenty minutes panning Ancestry for her grandmother's sister, then found the name in the record sidebar. The graph was decorative. The index did the locative job the chart was never asked to optimise for.

A River Does Not Climb a Trunk

Picture a river delta from the air. Water does not honour a single vertical trunk and polite bilateral forks. It splits, rejoins, abandons channels, and returns. Genealogical depth behaves like distributary flow: each generation adds capacity sideways, including cousin marriage and duplicate lines tree tests never modelled. Forcing that flow into one upright pedigree is like asking the delta to climb its source hill in a straight line. You can draw it. You cannot read it without zooming until the names blur.

Fan and hourglass layouts accept the delta shape for orient. They let depth accumulate in rings or stacked bands so the eye stays near the ego who cares. That is not decoration. It is hydrology applied to kinship, with search still doing locate when the channel forks beyond one screen.

Give the Familiar Tree Its Full Weight

The objection deserves its full force, including against my lab. Everyone learned node-link pedigrees first. Genealogists build muscle memory for add-parent flows on box charts. Ancestry's hints are genuinely useful when they fire. Researchers export pedigree PDFs for archives. Switching default charts risks alienating the constituency that funds the category. If fan charts hide discovery features, power users are right to call them pretty toys. Ship search, hints, and merge before you ship radial novelty as the only front door.

Take that seriously. Familiarity is a real asset. It is not proof of fit. Tree testing shows hierarchies fail when depth and breadth compound nng-tree-testing, but genealogy also needs card-sorting work on how people name sides and lines before you hard-code filters. User comparisons praise fan clarity while naming feature losses oyfh-fan-praise. The strongest products will pair layouts to tasks and measure each separately. I would abandon the orient-default claim if locate tasks favoured pedigree in a moderated study. Familiarity should not veto a layout that solves orient sprawl the category names in its own marketing copy.

The Chart You Default To

Genealogy is not failing because families are complicated; it is failing because one layout carries orient, locate, and edit while search and hints do the work the chart claims. I built three views and a search jump so rings, marriage units, and dimming filters could be judged against sprawl without pretending one shape fits every task. My composite cousin closed Ancestry and texted me the name the sidebar found after the tree refused to show her where to look.