Thirty years of United States history, rebuilt for the next thirty.
One of the internet's oldest free history libraries, moved onto digital infrastructure built to carry it forward, and handed back with its audience, its addresses and its search standing intact.
ushistory.org has taught United States history to the internet since 1995, and more than 600,000 people visit every month. We moved all 2,434 pages onto a single modern platform with one template system, kept every address the last thirty years had handed out working, and launched without a minute of downtime. The archive is now on infrastructure the organization can maintain, extend and plan against.
One of the most widely read free history resources in the world
The Independence Hall Association is a Philadelphia nonprofit founded in 1942 to advance the preservation of the city's historic district. In 1995 it put United States history on the web, and ushistory.org became one of the most widely read free history resources in the world: complete online textbooks, the founding documents in full text, the people and places of the nation's founding, and teaching material used in classrooms across the country.
Three decades of continuous publishing produced something genuinely rare. A domain authority of 70. More than 600,000 monthly visitors. Millions of words of original scholarship. And an inbound link profile built one citation at a time since the Clinton administration.
It also produced a website that had been added to, and never rearchitected, for thirty years.
A site built incrementally over three decades, without a unifying architecture
The infrastructure had outlived its support window by fifteen years
The site ran on Windows Server 2003 R2 with IIS 6.0. Microsoft discontinued the operating system in 2009 and ended support for the web server in 2015. Neither had received a security patch in over a decade. It ran from a single instance with no failover. The organization held its own copy of the site files, so the exposure was interruption and a slow return rather than loss of the material.
Four technologies, twenty stylesheets, no template system
Pages were rendered across Classic ASP, PHP, .htm and .html, with no consistent logic governing which was used where. Some ASP pages called Microsoft Access databases. At least twenty distinct stylesheet sets were distributed unevenly across sections. There was no unifying template system, so every page was, in effect, hand built. The site was built incrementally over three decades without a unifying architecture, and each layer carries the conventions of the year it was added.
Most of the library was not reachable from search
Of the 2,434 pages, only around 70 were generating meaningful organic search traffic. The rest held the same scholarship. Structural inconsistency, four competing URL prefix formats with no enforced canonical, and largely absent metadata made those pages difficult for a search engine to tell apart. That put most of the archive out of reach of a search result.
One domain was really several sites
A prior assessment had described the site as one domain behaving as four, meaning the four combinations of protocol and hostname that all served content with no canonical version enforced. That was true, and it was only part of the picture.
We ran a live DNS and certificate audit host by host rather than working from the inherited crawl, and found six subdomains beyond the main site. Most were infrastructure or aliases. Two held content, and neither appeared in any prior inventory. One was a blog that had been dead for years: confirmed defunct and retired. The other was a full microsite about Octavius V. Catto, the nineteenth century Philadelphia civil rights leader, running on a separate host and a different content management system. Roughly twenty sections including a civil rights timeline, teaching material, lesson plans and educator resources. Real, substantial content belonging to the organization.
It had been invisible to every previous assessment for a structural reason worth naming. A crawler begins at the main host and follows links within that host, so a subdomain living on a separate server falls outside the crawl boundary unless it is explicitly configured to jump hosts. This one was also serving over an invalid certificate, so browsers and crawlers were stopped before they reached any content. Off the crawl path, and erroring out.
We took it to the client with a recommendation and a clear scope boundary, because it was genuine additional work and it deserved its own decision. It also happened to align exactly with the direction the organization had told us it wanted to grow.
Four technologies do not merge quietly
Bringing Classic ASP, PHP, .htm and .html under one content management system is not a matter of moving files. Each technology had carried its own naming conventions for three decades, and folding them into a single namespace surfaces collisions that do not exist while they are apart.
And the migration had to be invisible to everyone else
The organization depends on this site. It carries advertising revenue that funds the mission, it is cited in classroom curricula, and its search position is an asset built over thirty years that cannot be rebuilt if it is lost. Every legacy address, including bookmarks and inbound links ending in .html, .htm and .asp, had to keep working on day one.
Do not migrate the problem. Rebuild past it.
The brief we were first given was a server migration. Get the site onto something modern and stable, then think about the rest. We advised against it, and made the case in full.
A migration transfers a system's problems rather than resolving them. Classic ASP does not run on Linux, so moving to modern, cost effective infrastructure would have required a line by line rewrite of every ASP file regardless. The twenty stylesheet sets, the four page types, the absent redirect logic and the canonicalization problems would all have arrived intact on the new server, with the added disruption of having just been moved. And the post migration maintenance cost would have stayed unpredictable, which for a nonprofit budgeting against a fundraising runway is its own kind of harm.
Our recommendation was to bypass the migration entirely and rebuild from the ground up on a unified platform, protecting the search equity through disciplined URL mapping and redirect architecture. Done once, done properly, the organization ends up with a system it can maintain and forecast against, rather than paying twice and still needing the rebuild.
They agreed, and that is the project we delivered.
Seven phases, in the order that removes the most risk first
Before any reconstruction, one narrow intervention: take a second independent copy of the legacy site into our own custody, alongside the organization's own archive and confirm the CDN in front of it was caching aggressively, so that a failure of the aging server during the rebuild window would not take the site off the internet. Low cost, no changes to the existing environment, and it meant the rebuild could not take the live site down with it.
We conducted our own current state assessment rather than working from the audit already in hand: server configuration, page type inventory, database dependencies, stylesheet structure, redirect logic, canonicalization patterns, metadata completeness, and content distribution across the directory hierarchy. The deliverable was twofold: a technical findings report, and an information architecture blueprint organized around how readers use the site.
That blueprint reorganized thirty years of accumulated structure into six content pillars:
A full crawl produced an inventory of every live URL, its position in the hierarchy, its inbound link profile, and its metadata. We pulled twelve months of search data so that every subsequent decision could be weighted by what people read rather than by the order the crawl returned. That weighting set the order of the information architecture, the redirect priorities and the recovery work.
Content moved onto a modern, unified publishing platform with a single template system. Migration is where a rebuild of this age either succeeds or quietly fails, because a thirty year old site does not surrender its contents cleanly. Three recovery efforts ran alongside the main migration:
The site is an archive and a library, and the design says so. A serious, archival register throughout: Playfair Display for headings against PT Serif for body text, a palette drawn from brick, tan and parchment, and generous measure on running text because these are pages people read rather than scan.
Every landing surface was built from a single template pattern rather than individually, so the site now has a consistent visual system that the organization's own staff can extend without a developer.
A brand refresh, thirty years in the making
The rebuild was also a rebrand. We gave the organization a complete new identity: a US History mark drawn as a founders era monogram, a star anchored between two rules, and a palette of rustic browns, brick and parchment taken from the materials the archive is made of.
It reads as confidently stamped on card stock or cut into wood as it does in a browser tab, and it gives thirty years of publishing a mark drawn from the same materials as the archive.
PT Serif carries the reading. Generous measure, quiet contrast, and a bookish rhythm chosen for pages people sit with rather than scan, set against parchment drawn from the archive itself.
Variants for light, dark and busy grounds, with a compact mark drawn for avatars, favicons and social profiles, so the brand holds together wherever the archive travels.
This is where a rebuild of this age lives or dies, and it deserves its own section. See the technical deep dives below.
No cutover without proof. Seven gates, thirty six individual checks, run against the production path before DNS moved and again through the edge after it.
The work that is invisible when it is done right
What was verified at launch
The rebuilt site went live on 30 August 2026. What follows is the day one snapshot: early by design, because day one is the right moment to assess launch impact, and it will be extended and restated as longer stretches of performance data close. A rebuild changes the address of every page, so an honest read has to control for everything that changed alongside it. We compared Sunday against Sunday, because this site's weekends and school days behave nothing alike. We looked only at readers arriving from search, the largest and most consistent audience. And we used only the measures that were recorded the same way on the old site and the new one, setting aside the ones the change of platform had made incomparable. On that basis, the audience came through the transition intact and read considerably more of the site than it had before.
Every address on the site changed. The clicks did not notice.
A rebuild changes the address of every page, and the risk that matters is that search results stop resolving. Search Console counts clicks at Google's end rather than ours, which makes it the one instrument that spans a cutover without changing.
Across the five days from launch the site took 9,050 clicks from search, against 9,813 in the five equivalent days before it and 6,818 in the five days before that. Impressions rose 4.3 percent over the same window, from 889,253 to 927,628. On launch day itself, a Sunday, clicks ran ahead of both of the two Sundays before it.
| Clicks from search | 16 to 21 Aug | 23 to 28 Aug | 30 Aug to 3 Sep |
|---|---|---|---|
| Sunday | 825 | 985 | 1,114 |
| Monday | 1,413 | 1,745 | 2,192 |
| Tuesday | 1,491 | 2,143 | 2,121 |
| Wednesday | 1,532 | 2,491 | 2,029 |
| Thursday | 1,557 | 2,449 | 1,594 |
| Five day total | 6,818 | 9,813 | 9,050 |
There is no step change at the switch. That is what 6,212 verified redirect rules are for, and it is the single result this project was most at risk of not achieving.
Measured on a Search Console domain property, which covers every hostname and protocol form the legacy site served. Thursday 3 September is provisional: Search Console finalizes its most recent days over the following 48 hours and that figure will revise upward.
The archive is moving up while Google reassesses it
Changing the address of every page on a thirty year old site sends Google back to reassess all of it, and that reassessment runs for months rather than days. It is still running. The direction it is running in is already clear.
Net movement, meaning after everything that moved down has already been subtracted. The gains by page below add up to more than the net figures for that reason.
| Bill of Rights and later Amendments | 2,961 | |
|---|---|---|
| Twelve other pages, between them | 304 |
| The Pottawatomie Creek Massacre | 82 | |
|---|---|---|
| The Townshend Acts, lesson plan | 49 | |
| Betsy Ross, flag questions answered | 43 | |
| Thomas Paine, The American Crisis | 30 | |
| US textbook chapters | 25 | |
| Benjamin Franklin, his life | 22 | |
| The Declaration of Independence, Congress's draft | 17 | |
| Betsy Ross and the US Flag | 13 | |
| Four smaller pages | 23 |
The page carrying most of that movement is the one the rebuild had the most to fix. Before the switch it held no navigation markup of any kind, and asked for a stylesheet that did not exist. The material was always there. What it did not have was a structure a search engine could read.
Visibility is not readership. A page that has moved up has not yet been clicked, and the value is only realized when it converts into visits and then into pages read, which is how this site earns. Search Console will settle that, and it needs ninety days or more before the answer means anything. Term level detail stays in the client reporting, where it belongs.
We ran a traffic weighted sweep of the site the morning after launch, checking the highest earning pages against the live site rather than waiting for reports to arrive. It surfaced a set of legacy address patterns worth 51,315 annual visits between them. All were diagnosed and corrected the same day, and re verified individually.
The sweep is part of the engagement, not a follow up. Day one is when a broken legacy address costs the least to find and the most to leave in place.