When 18-year-old African-American Michael Brown was shot six times by police in Ferguson, Missouri, Ed Summers and his colleagues quickly began collecting tweets.
The controversial 2014 killing of Brown had become a focal point of the Black Lives Matter campaign: Mr Summers' team were looking to use the social media platform as a new tool for documenting abuse.
Within two weeks, more than 13 million tweets had been collected, and the experience had given birth to a new archiving initiative: Documenting the Now, a specifically Twitter-related human rights project.
"[Tweets] are transient and there are so many of them, it's sometimes difficult to think about them in the context of an event," says Mr Summers, the project's technical lead.
Each tweet on the platform, once sent, gets its own webpage in the form a discrete URL. While tweets can be deleted by users, they can't be edited after publication. For a platform characterised by its impermanence, this presents an irony.
"In some ways it's very stable," Mr Summers says. "It has this fixed nature to it. But at the same time, it can be deleted at any time."
Consent, anonymity and volume
Documenting the Now is a joint initiative between three institutions: Washington University, where Mr Summers works as a researcher, the University of California, Riverside, and the University of Maryland.
Mr Summers admits it's a "work in progress", as archivists and activists struggle with the complexity of big data. One of the core issues, he says, is gaining consent. Many people who use a service like Twitter don't want their comments to last, and Mr Summers says it's important not to disrespect that.
"There are expectations that perhaps tweets will not last a long time," he says.
"If you send a tweet with an image of protest, you may want it to be available in the short term to raise awareness about the protest, but you may not want it to live in perpetuity in an archive."
Consent is doubly important, Mr Summers says, if the use of someone's tweet for campaigning purposes could risk placing them in danger.
"The other problem is that a platform like Twitter offers a fair degree of anonymity," he adds. "Unlike Facebook, there is no requirement for your account to reflect your actual identity."
Not everybody, especially protesters, want to be online in that way, he says. "So to engage with them as individuals raises a lot of challenges."
A more basic consideration for Mr Summers' team is one of volume. Modern technologies allow for the gathering of vast amounts of information, but for that information to be of use, it has to be adequately assessed for meaning.
Because of this, archivists in the digital age need to be selective. But how do you decide what to hold on to, and what to discard?
"There is no one answer to that question," Mr Summers says. His team have used a few strategies: giving priority to the tweets of hashtag creators, or to people whose posts have garnered significant retweets.
"Someone sent a tweet and it happened to get retweeted by a celebrity and it made a lot of impression, so it got retweeted tens or hundreds of thousands of times. So that would be a good candidate for wanting to archive it."
Keeping an eye on evil
For Jay Aronson, director of the Centre for Human Rights Science at Carnegie Mellon University, this is nothing new. "There have always been accounts of human rights violations," Professor Aronson says. "I think what's new here, what's really exciting, is the volume of data that we have."
Professor Aronson credits the arrival of the smartphone with a change in the dynamic of human rights monitoring. Today's protests, he says, often have "hundreds or even thousands" of people recording audio and vision.
"The amount of information that we have about events that take place in public is vastly larger than it was," he says.
This allows archivists (and journalists) to piece together several accounts. For Professor Aronson, the aim isn't finding a neutral or authoritative account.
"If you can gather enough accounts, you can understand the biases in each particular view of the event," he says. "You can better understand what you know, and what you don't know."
Aside from helping to establish context, the more cameras there are, the greater the chance of witnessing specific evidence of a violation. Professor Aronson is currently involved in a Ukrainian case focusing on deaths during the Euromaidan protests, where video was "crucial" in making the case against a particular group of riot police.
YouTube, Professor Aronson says, has become the archive of the public present. But online, he says, videos and images are easy to manipulate.
"When we think about the human rights context, one thing that I'm always worried about is the selective use of video to tell a particular story that may not be the truth," he says.
A case of forensics and counter-forensics
When it comes to the monitoring of potential human rights violations, verification is as important as collection and collation.
Richard Matthews from the University of Adelaide's School of Electrical and Electronic Engineering says there's an arms race between forensics and counter-forensics.
"Every time we release a new technique, somebody very shortly afterwards will release a counter-forensics technique, and we've got to then release a counter-counter-forensics technique," he says.
Dr Matthews works with governmental and non-governmental agencies to help establish what's real and what's not, with a focus on determining the unique identifiers that are left on an image by a digital camera's image sensor — the device that records the image as it is taken.
These unique identifiers can only be seen after running images through complex signal processing algorithms.
"We can extract this difference and get a unique fingerprint for each image sensor," he says. "Not just between model and model of camera — it's actually between each individual image sensor."
Everybody has a phone; not everybody knows how to use it
Taking a video or photo in a war zone or any politically or racially charged environment carries significant risk. The organisation Witness works to sharpen the skills of those keen to help document abuse.
"We train on understanding where video can be used in cases, what aspects of the situation are most useful to capture," says Yvonne Ng, a senior archivist with the New York-based non-profit.
"We train on how to capture important information about a video so that it can be authenticated and verified by users down the line."
The organisation has also developed a video evidence field guide for lawyers to help them better understand the ways human rights videos can be used as evidence in legal settings.
"Even outside of the courtroom, documentation needs to be verified for it to be used by journalists or investigators," Ms Ng says.
"This is particularly important when the video is coming out of conflict zones or heavily censored areas, where eyewitness video is sometimes the only source of information."
One organisation Witness has partnered with, the Syrian Archives, has developed a workflow to identify, analyse and publish thousands of citizen videos collected from the war-ravaged stage.
"They make those verified videos available online with contextual information so that people can identify and understand what's going on," Ms Ng says. "Their hope is that these videos and reports can eventually be used to achieve justice and accountability because they have been verified."
Protecting your sources
But once you've built an archive, how do you safeguard it from the hackers, saboteurs or even subpoenas?
Professor Aronson says there's no simple answer. That it will depend on the level of threat faced by activists, archivists and their sources. Particularly heinous video — of torture, or sexual violence, say — might be "air gapped": stored on an internal server not connected to the internet.
Another suggestion is the establishment of an "evidence locker" — a central repository where important human rights related data from all over the world can be stored in safety. But according to Ms Ng, the idea is a contentious one.
"There is a question of transparency: who will be the organisation that owns this evidence locker and whose interests do they serve?" she asks.
"Why do they get to make those decisions about what gets collected, who gets access and how it can be used?"
In the meantime, she says, it's vital to build awareness: to make people understand that YouTube is not a safe archive and that if they have important material to safeguard, they should also think about making back-up copies.
Witness have a general guide to archiving video for activists. The aim is to provide small groups with fewer resources with the knowledge to have their own videos on hand when they need them — not relying on social media.
"Empowering people to manage and preserve their own videos, it is an approach that makes sense."