Meeting minutes
Wendy reviews agenda
Agenda-curation, introductions
Wendy: Do we have any introductions? Anyone new to the call?
FalconCNS https://lists.w3.org/Archives/Public/public-web-adv/2021Aug/0003.html
Wendy: Jay, you are queued, up so let's take up the Falcon project
Jay: Hi I'm with Privacy Coop
… for past several years and they are working on approach to consent management that scales across uses and information, and supports all tech
… we believe as we work through use cases with affiliates
… this has become an "Assumption Blaster"
… we start new projects and ist assumptions we cannot solve for
… start with cannots that are off the table
… for example, consumers can't know what 3rd parties are using data
… solutions can't adhere to all regulations
… cannot support asynchronous identity
… operational or big data cannot be stored same as transactional data
… consistent legal language not possible
… frequency capping not possible
… What is Privacy Coop? We're a non-profit; we have members, open to public
… we represent our members' needs
… consumers find value in their rights and we help businesses find opt-ins
… Don [Marti] pointed me to the Consumer Reports link
… they set up an agency and did a report on it [URL on slide]
… we publish in public domain
… been through two iterations of working groups
… if we break this into simple sentence, people understand
… Subject is a person, place or thing
… C-suite folks
… we say person and they hear phone number of email
… but it's a human, and can be a phone number, email; but can also be a thing
… Subject is one of those things
… Predicate is compound
… has to have one of three things: one organization, one program
… legaleeze
… program data, so we swing with it; program is legal use of data
… Jurisdictions, a sovereign region that regulates data, passes laws
… Focus on Organization and Program in the predicate
… If you ever played the game "Battleship"
… we have set up businesses/organizations and the legal uses of data across the columns
… R4 is GM using your data for highway safety studies
… either 0 or 1
… value can be set by regulators, or whoever writes jurisdictional statement
… they could say I think GM should use your data for highway safety studies
… but GM has attorneys could say it should be 0
<garrett_johnson> This looks a lot like IAB TCF
Jay: you could say yes and it flips back to 1
… there is a lifecycle that starts with lifecycle, moves to business, and consumer
… we maintain integrity of upper lefthand quadrant
… if use of data goes out of vogue, we leave it as a dead column
… in this way we can add organizations forever and add legal uses of data to righthand side as more columns
… this helps us blast one of first assumptions
… can deal with any state in time
… can collect data 50 years from now and be in compliance with Programs existing today
… Take this two-dimensional plane
… make a perfect copy and layer it over the first one
… now take this new layer and assign it to the predicate, the jurisdiction
… if GDPR wants to front-load consent elections to be opt-in, they can do it
… or if California wants to do opt-out, they can do that
… We can do this on and on with the Z axis
… we can address jurisdictionality forwever
… LGB and Brazil may be a new layer
… I drew an access line down J4
… data used by FannieMae
… and traffic studies
… look down the axis
… and populate values
… if a calling party doesn't specify a jurisdiction
… if 0, you cannot use data
… if party says they are collecting data, and only one that applies is US74
… then they get a 1 and they can use the data
… how Privacy Coop works, if a member licenses their data, we can flip 0s to 1s
… and conversely to opt-out we flip all those bits to 0
… and a column for all out and all in
… you can get as granular as you need to
… AT&T for example has nine consent for all programs
… Look at all of the strata, it forms a kind of wafer
… we can compress into a binary to be platform specific
… in our test, we can support 20 years of growth
… in programs, regulatory...
… and size is still less than 84kb
… very addressable
… can be dated in time
… we use lazy provisioning
… remain sparse bit maps
… can get back an answer quickly
… What we are proposing
… to this austere group
… We work with IAB, others
… we propose that the metadata around the scheme can become a public commons
… much the way Privacy Coop manages its domain name
… or we can delegate to GoDaddy, etc.
… and DNS as an open stack could be replicated
… and you are just replacing internal guts...pass in the address and get back a 1 or 0
… only thing needed to change in DNS stack
… we are proposing a sibling to DNS
… a new proposal called the Consent Name Service, CNS
… If I cannot got Facebook without going to DNS
… might be good to ask if can use data collected
… this is the general proposal
… before I pass to Alan, my colleague who will show a sandbox of how it works
… Left side is an example
… we made decision that we use GLEIF
<AramZS> So this does not sayyy map to domain... ?
Jay: global entity identifier database
… rationalization needs to be efforted to a group such as this
… how to put in new identifiers, what are those global standards, what do they look like, how modified over time
… that's what we want to get out of this
… Thank you in binary
… stop sharing and go to Alan to share his screen
Alan Nekhom: what we have done for Privacy Coop
… we have implemented what Jay showed as conceptualization
… we used with our affiliates for consent elections
… one partner gets consent one by one upon user login
… they have ID and authn stack to our version of CNS
… we have list of jurisdiction; 136 with privacy policies yesterday
… they pop up right and left
… First thin is that we have a number of dummy companies
… If we pick 1LLC
… would address commons only for their particular row
… to determine if I, Alan, have chosen these rows
… but contrary to spirit of data privacy
… Let's say 1LLC wants to use GDPR and Australia
… and use data for relevant advertising
… I have pre-populated information
… we can support any ID methodology that is needed
… we build JSON strings for API port
… we prep and submit it
… and here we have results for each of subject end points, we got a 0
… consider BBC, a media company, and say GDPR
… and stick with relevant advertising
… rerun and our subject matters have all consented to that use
… if we add personalized services and recheck, not a lot of shifting
… Jay mentioned that our affiliates and members have particular agreements
… let's say an affiliate wants all uses and all jurisdictions
… can see here, all approved
… go to a large entity, individuals have made consent elections that vary by jurisdiction, subject
… Two emails for example, there is no data to say to whom these email addresses belong
… my gmail and his aol could both belong to Jay
… Next step is the ability to update
… Let's say by company
… In US, say external marketing
… leave same subjects
… get compressed binary of the entire modified election
… dummy data we have been updating all at the same time
… wind up at same state
… Go back to data seller
… Choose: USA, external marketing, and query
… we are opted out
… and come back here and opt in with same selections
… now you can see....
… it is very fast, Sandbox runs in AWS in T2micro
… can handle blocks in other 5 seconds
… Finally we can come over here and do transactional consent
… I pre-populated this; chosen US @ external marketing
… my email at gmail and non participant
… non-participant has not consented
… run again and a fourth time, setting meter
… my email now says you cannot use it anymore; let you use three times and that's it
… we can support metering, blends of jurisdictions; we do logical in and org
… that's the gist of it
… anyone have questions?
<wseltzer> [showing https://
Wendy: thank you, Jay and Alan
James: thank you both for the presentation
… how do you know that the person who enters email and telephone number is legit?
Jay: Let's say person making request is AT&T
… they will have email address that they pass in
… they can verify whether that person is that person
… but they get back a default value per jurisdictionality
… quick way for corporation to get back response
… Second aspect of question
… if person is logged in through authorized agent, how do we know who they say they are
… what we propose is to follow the existing DNS framework
… how would GoDaddy verify that I am owner of Privacy Coop before they make changes
… there is security in place for that
… we would leverage the existing infrastructure
James: So if I'm AT&T
… and if you have not registered on platform and I ask about GDPR
… you are dealing with all those jurisdictions
… you are trusting the DNS
Jay: correct
James: if you later put your email in, you can personalize by org based on the 3D matrix you showed earlier
… you asked me question; email addresses and phone numbers are verified by sending info to those emails and numbers
Jay: yes, that is what we do
… and you highlight an additional feature, James
… If AT&T can pull effectiveness of their programs and support with non-repudiation
… as they are concerned with potential violations
… and they have data for the first time
… so could owner of that email address owner; who's been asking questions about my data
… and lastly jurisdictions themselves
… GDPR can go and collect stats on the effectiveness; same for CA and Brazil
… we provide them an opportunity for data to understand effectivenss of decisions they make
… Idea that Romania cannot datamine US vice versa
Alan: let me add some other tidbits
… we can put in open source concept of data map and bit source
… network think of physical layer
… multiple layers that are we are not making suppositions about
… so many companies that are expert on how to handle that
… unified ID is one we picked because an affiliate wanted to support it
… Start looking at account numbers, which we support internally
… Ouch, may not be unique across companies
… doesn't matter, would be unique with in the company and generate an XYZXXX coordinate
… in the hyperkey
… Each of you has a cube today of your own
… you have cube for each of your input
… three cell numbers, you have three cubes
… but very sparse because we don't have companies asking about phone numbers
… I'll assume you have a 1.214.555.1212 phone number
… if Amazon reaches in, CNS has no idea that is Angelina's phone number
… it will lazy provision a row agains that phone number against jurisdicitions, so cube is sparse
… If Angelina comes a week later and pays for licensed use of her data and can see that Amazon asked to use my cell phone for advertising
… that is power for individuals to have control
… Data privacy is a hot topic among consumres
… At same time, we enable companies to use data legally
… Most companies cannot shrug off the fines
… need for data use for business, and allow people to throttle it
… and we want to see our members be paid for use of the data
<Zakim> AramZS, you wanted to say Is there a technical specification here? Could I take your idea an implement it myself or does the current proposal require interaction with your co-op?
Wendy: one more question from Aram
AramZS: I read the document; thank you for the presentation
… this does not seem like a technical specification
… If I wanted to build my own Privacy COOP
… I don't know how to do it in a standard way
… Is proposal here a methodology to access data in your coop?
… or a general for anyone to build on
Jay: neither
… we found effectiveness for big data and transactional requests and frequency capping
… gets thorny in other standardized solutions around consent election
… We propose taking core element, the bit map
… and deploy it
… we are telling you how and why
… idea that I think is available is to take the schema and the limited API
… and publish into the public domain and become a CNS server
… DNS servers today hosted by governments and private corps
… I picture a cloud computer specified as DNS and beside it CNS server
… supplemented with pubhub
… this isn't for Privacy Coop to have copy of data
… we could host and figure out right mapping, and update the public commons
… we are not guarding the data at all
… if I type in Privacy.coop, provides an address...would be for consent elections
… May not be that important to this group
… we had 6 petabytes, to resolve mathematical component of map reduction
… did not have to drain for every single program
… provide a pointer to the address looking for, so resolves to a math componenet
… could run one for advertising
… and not scrub the datalake
… huge for considering data solutions
Wendy: We need to move on
… where should people engage to comment further or ask further questions?
Jay: document was published, or ino@privacy.coop email
Wendy: thank you and feel free to come back with future updates
Jay: we are available for Q&A
Wendy: we need to leave time for Suishi
SUSHI: https://github.com/privacycg/proposals/issues/27
Wendy: Russell, tell us about SUSHI
Russell: SUSHI is way to serve interspaced ads in a privacy preserving way
… after seeing Ben Savage's proposal, seeing vector is confusing
… main goal of this proposal is that is very easer for end user to understand why they saw a particular ad
<wseltzer> https://
Russell: can click and see topics of interest
… and why we thought you were interested in those topics
… Angelina brought up stuff Google is doing with FLoC
… would want to see what they are doing with FloC
… short summary
… when web site wants to show contextual ad, it suggests a set of topics related to page
… from a content hierarchy, something like IAB's content taxonomy
… someone interested in college football, basketball shoes, or three or four topic suggestions
<AramZS> This Falcon idea seems to me to 1. not be a specification of the type we normally discuss and t/f out of context for this group and potentially out of context for the entire W3C. 2. Unclear in its advantage over say... a single user having a browser extension that manages this data and communicates it in some way to downstream systems. 3. Difficult to conceptualize or parse without actionable code. If the hope is that this become some sort of
<AramZS> request layer, I would expect to see a fully open-sourced code base that proves out some of the discussed processing.
Russell: to use those topics on page, browser remembers
… browser can identify topics you are interested in
… when site requests to serve an ad, in addition to page serving
… it can tell ad network other things you are interested in
… football and user interested in topics related to football
… for history of user and related things
… advertisers can make suggestions for topics that may be of interest to a user
… it's a hierarchy
… so adv won't follow you around with particular shoes for example
… basketball shoes may follow you around, but ...
… a little less
… to protect topics shared, don't become general interest unless topic is of general interset to the user
… it would work with PARAKEET
… rather than topics shared directly with publisher or ad network
… contextual topics and the learned interest are shared with an intermediadiary
… to filter to make sure they are not uniquely identified
… so ads cannot be tied back to user or to construct a profile of the user
… on the link shared
… I have a ranking algorithm used ot go in and determine which ads, what topics of interest
… have it phasing out over 30 days
… things more recent v longer ago no longer impact rankings
… can see in browser
… here are contextual topics for this page; topics shared with PARAKEET that seemed to be learned
… and allows user to go in and mark interest for ads, or stop showing ads for this topic; the user-specified portion
… user can go in and indicate their interests; have higher probability of generating ads because advertiser knows person is interested in the topic
… That is a high-level view
… any questions or comments?
Kleber: thanks, Russell
… my initial reaction to this, which has come up before in this group
… what parts you describe are part of browser v developers who write capabilities that browser provides to build own things on top of
… SUSHI sort of crosses boundaries
… some parts for browser; and does browser provide capabilities for adtech to build what they need
… expressed signals, [@@] may be something browser should let adtech cos do on their own
… Where this ties in is Chrome's Fledge, or PARAKEET proposal
… where there is barrier with ad being shown and person being shown i
… it
… how to 'slice up'
… how to divide this proposal
… here is where we use PARAKEET to show ads
… and here is additional capabilities we need in IG creation to ad to browser platform so that adtech company could do so on top of these creds
Russell: Could be
… these are some ideas to be adopted
… instead of a javascript to set ranking algorithm, may provide adtech with ranking
… algorithm
… could show results with something like PARAKEET
… happy to discuss with you at Google or others on how to utilize aspects on this
Brian: I think this is interesting
… I had similar reaction to Ben's proposal
… @@
… maybe combine advertiser, publisher and user
… I agree with Michael also
… at first not sure where innovation would come in
… but there are opps to determine where to float out
… how to decide what interests to add to browser
… for user
… that is where we can provide innovation
… then SSPs, DSPs, publishers can go about ranking this for a specific opp
… I like it a lot
Russell: thank you
… another aspect, want to report aspects that led to conversions
… understand which ones to bid on in the hierarchy
Jay: Terrific
… awesome, love the presentation
… harkening back to what Alan showed, frequency capping within the CNS
… we separate dividends from royalties
… can contact me X times per month and it's a royalty
… I can see where your SUSHI can work
… set cap and tie into your infrastructure
Russell: Ok
Wendy: so I see you have hosted this as a proposal in the Privacy CG
… should people go to that thread to add comments?
Russell: yes, comment on that thread, or email me directly
Wendy: Thank you
… Sounds like we are seeing interest in the way these proposals combine, highlighting components and who implements
… hope this feedbacks into our discussion
… of next phases and what progresses toward web like standardization
… If we have no one else on queue
… Angelina had raised some questions about FLoC and topic identifiers
… that make a good follow up from your presentation, Russell
… We have a few minutes to address
FloC ID vs. Topic-Based?
Angelina: I am following up following up article in DigiDay
… where Josh Carlin commented good idea to move away from FLoC IDs and move to tagging in sites
… is this being considered?
… If so, when would decision be made?
… when could we trial again?
… And are there learnings on the FLoC trials, anything further on topical?
Kleber: I can answer that
… so the DigiDay article you are referring to is an article about talk that Josh Carlin gave to @ inside IETF
… that talk was intended to be a summary of what the nature of the first FLoC origin trial was
… stuff people here are already familiar with
<wseltzer> https://
Kleber: and a highlevel overview of comments we received on first origin tiral
… including ways to improve FLoC
… You specifically brought up topics
… as part of the input to FLoC
<wseltzer> https://
Kleber: people in this meeting will be familiar with
… three meetings suggested
… use of topics instead of cohorts
… idea of discussion was to talk about the range of things being suggested
… none was an announcement
… just listening to feedback and how to incorporate with revised plan
… once we have revised plan, we will come back to this group
… nothing I have right now
Angelina: yes, that article might have given people some optimism for a more topical path
… so it seems it was a bit misleading
Kleber: if you think topical approach is a good thing
… that is helpful feedback and we are happy to hear this feedback in this forum or on Twitter or elsewhere
Angelina: based on conversations with members
… going topical route has given people...more hope...or rather more desire for more topical
… ways to label content with numerous topics
… mahybe 3-5 topics per page or per domain, people are excited to hear about
Kleber: yes, thanks for comments
… Topics showed up in Josh's talk because of this type of feedback
Wendy: Like W3C, IETF takes minutes, so I included a link to their minutes
… further conversation here is welcome
Kleber: and I think there is also a video on YouTube
… if you prefer that to minutes
Wendy: We are at the end of time here, so send your agenda requests
… Thank you Russell, Jay and Alan for sharing presentations
<wseltzer> [adjourned]
<kleber> YouTube video of Josh Karlin's PEARG presentation on FLoC: https://
<kleber> (Starts at 1hr32min into the 2hr video)