Static snapshot over live query: the fault model behind the domain map

The 3D map of the portfolio renders from a versioned catalogue manifest built at deploy time. The WebGL ocean is a layer over a plain HTML index that works without it. We rejected a live API that would read the portfolio database on every visit, and this record states that option on its merits, explains why it failed the fault analysis and names the condition that would reverse the call.

The option we turned down

The rejected design served the catalogue from a live API that read the portfolio database on every visit and returned exactly the domains currently for sale, with current prices and cluster tags. Its case is strong. There is one source of truth, a closed sale vanishes from the map the moment the record changes, and nobody maintains a build pipeline that can drift from the database. For a portfolio whose whole business is the offer, showing a sold name as available in the middle of a negotiation is a real error, not a cosmetic one, and the live API removes that class of error entirely.

Everything in series

The cost is that the database, the API process and the renderer sit in series. The map is only up when all three are up, and their availabilities multiply rather than average. The 3D scene makes this worse, because it needs the whole catalogue to lay out the clusters in depth, and a partial or slow response yields an empty abyss rather than a degraded one. The failure that settled it was more mundane than an outage: WebGL context loss on low-end mobile GPUs, which leaves a black canvas with no navigation. In the live design the data existed only in JavaScript memory behind that canvas, so a lost context meant a page with nothing on it at all.

What stays up when a part fails

The chosen design builds a versioned JSON manifest at deploy time, renders from it a plain HTML index of every domain grouped by cluster, then mounts the WebGL scene on top as an enhancement. On webglcontextlost, or when WebGL is unavailable, the canvas is torn down and the list is revealed with the same anchors, so deep links keep working. Web accessibility guidelines such as WCAG already demand a text equivalent of the map, so the fallback costs almost nothing extra, and real users exercise it every day instead of leaving it to rot untested. The manifest is cached at the edge for 600 s with stale-while-revalidate, so an origin outage serves the last good snapshot instead of an error. The inquiry form posts to a queue that tolerates the backend being down and rechecks status when processed, and that is how we pay for the remaining weakness: a sold domain can stay visible for one cache lifetime, and it gets corrected in the reply to the buyer, not on the map.

The condition that reverses this

Go back to the live API the day status changes arrive faster than one rebuild plus the 600 s cache lifetime, which by our estimate means moving from a few changes per week to timed auctions or live bidding, because at that point a stale snapshot stops being a lag and becomes a wrong offer.

Why not keep the static layout and add a live status call per domain?

That hybrid is the migration path if the reversal condition gets close, but it is not free. The status call is a new runtime dependency, and the page still has to render correctly when that call fails, which puts you back on the static design with a badge on top. Until status changes outpace the rebuild, it adds a failure point without changing what the buyer sees.

Back to Deep Tech Assets · Domains

← Back to Deep Tech Assets
Engineering & Architecture Brief

Static snapshot over live query: the fault model behind the domain map

The 3D map of the portfolio renders from a versioned catalogue manifest built at deploy time. The WebGL ocean is a layer over a plain HTML index that works without it. We rejected a live API that would read the portfolio database on every visit, and this record states that option on its merits, explains why it failed the fault analysis and names the condition that would reverse the call.

The option we turned down

The rejected design served the catalogue from a live API that read the portfolio database on every visit and returned exactly the domains currently for sale, with current prices and cluster tags. Its case is strong. There is one source of truth, a closed sale vanishes from the map the moment the record changes, and nobody maintains a build pipeline that can drift from the database. For a portfolio whose whole business is the offer, showing a sold name as available in the middle of a negotiation is a real error, not a cosmetic one, and the live API removes that class of error entirely.

Everything in series

The cost is that the database, the API process and the renderer sit in series. The map is only up when all three are up, and their availabilities multiply rather than average. The 3D scene makes this worse, because it needs the whole catalogue to lay out the clusters in depth, and a partial or slow response yields an empty abyss rather than a degraded one. The failure that settled it was more mundane than an outage: WebGL context loss on low-end mobile GPUs, which leaves a black canvas with no navigation. In the live design the data existed only in JavaScript memory behind that canvas, so a lost context meant a page with nothing on it at all.

What stays up when a part fails

The chosen design builds a versioned JSON manifest at deploy time, renders from it a plain HTML index of every domain grouped by cluster, then mounts the WebGL scene on top as an enhancement. On webglcontextlost, or when WebGL is unavailable, the canvas is torn down and the list is revealed with the same anchors, so deep links keep working. Web accessibility guidelines such as WCAG already demand a text equivalent of the map, so the fallback costs almost nothing extra, and real users exercise it every day instead of leaving it to rot untested. The manifest is cached at the edge for 600 s with stale-while-revalidate, so an origin outage serves the last good snapshot instead of an error. The inquiry form posts to a queue that tolerates the backend being down and rechecks status when processed, and that is how we pay for the remaining weakness: a sold domain can stay visible for one cache lifetime, and it gets corrected in the reply to the buyer, not on the map.

The condition that reverses this

Go back to the live API the day status changes arrive faster than one rebuild plus the 600 s cache lifetime, which by our estimate means moving from a few changes per week to timed auctions or live bidding, because at that point a stale snapshot stops being a lag and becomes a wrong offer.

Technical FAQ: Why not keep the static layout and add a live status call per domain?

That hybrid is the migration path if the reversal condition gets close, but it is not free. The status call is a new runtime dependency, and the page still has to render correctly when that call fails, which puts you back on the static design with a badge on top. Until status changes outpace the rebuild, it adds a failure point without changing what the buyer sees.