Merge pull request #12 from MagniteEngineering/TURTLEDOVe

Writing up TURTLEDOVe style workflow

TURTLEDOVE-style.md

@@ -0,0 +1,83 @@
+# Proprietary Cohorts provided via Two (or more) Uncorrelated Requests Then Locally Executed Decision of Victory Everybody!
+
+We often have the opportunity to trade some functionality or convenience for increased privacy. The workflow descibed in the TURTLEDOVe proposal offers a method which could be applied to the concept of Proprietary Cohorts to increase privacy.
+
+We will outline how this could work, and discuss the tradeoffs of this method.
+
+## How it would work
+
+A website (news-publisher.example) who chooses to utilize proprietary cohorts adds a new html tag to the page header:
+
+```
+<link rel="cohort-provider" href="https://cohort-provider.example/.well-known/cohort-assignment.js" reader="first-ad-newtork.com">
+```
+
+A supporting browser reads this tag and executes a request to https://cohort-provider.example/register?cohortRegistrationId=12345&domain=news-publisher.example where 12345 is a unique identifier generated by the browser included on all cohort registration requests across domains to this provider.
+
+The cohort provider records this cohortRegistrationId and domain to build a proprietary graph of browserIds and behaviors and returns a cohort ID to the browser.
+
+```
+{
+    cohortID: abcd
+}
+```
+
+The browser stores this cohortId but does not make it directly available to webpage.
+
+The reader attribute indicates the ad-network that should get access to the cohort id.
+
+From here, we borrow some of the TURTLEDOVe workflow:
+
+Since a relevant ad network is used by a web site I visit regularly, a while later my browser contacts first-ad-network.com and requests ads targeted at this interest group:
+
+```
+GET https://first-ad-network.com/.well-known/fetch-ads?cohort=cohort-provider.example_abcd
+```
+
+The response points at a [Web Bundle](https://wicg.github.io/webpackage/draft-yasskin-wpack-bundled-exchanges.html) holding an ad, along with some browser-intelligible metadata about when that ad can appear, and some ad network logic and data for later decision-making.
+
+```
+[{'cohort-owner': 'cohort-provider.example',
+  'cohort-id': 'abcd',
+  'ad-cbor-url': 'https://first-ad-network.com/ads/ad-123456789.wbn',
+  'cache-lifetime-secs': 21600,
+  'max-times-per-minute': 1,
+  'max-times-per-hour': 6,
+  'auction-signals': {'atf_value': 250, 'btf_value': 25},
+  'decision-logic-url': 'https://first-ad-network.com/js/on-device-bid.js'}]
+```
+
+The browser fetches the on-device bidding JS and the Web Bundle.  The bidding JS is very simple:
+
+```
+function(adSignals, contextualSignals) {
+  return contextualSignals.is_above_the_fold ? 
+           adSignals.atf_value : adSignals.btf_value;
+}
+```
+
+A while later I visit news-publisher.example again (or perhaps another publisher using cohort-provider.example).  An ad network script on the page requests ads from first-ad-network.com.  The response requests an in-browser auction to pick between a contextually-targeted ad and the previously-fetched proprietary-cohort-targeted ad:
+
+```
+const contextualBid = 107;
+const metadata = {'network': 'first-ad-network.com',
+                  'auction-signals': {'is_above_the_fold': true}};
+const rendered = navigator.renderInterestGroupAd(contextualBid, metadata);
+if (!rendered) {
+  ...render the contextually-targeted ad...
+}
+```
+
+The ad network's on-device auction logic runs, and since the network believed the ad would render above the fold, it returns the atf-value 250.  The browser compares that to the contextual ad's bid of 107, finds that the contextual bid is lower, and renders the previously-downloaded web bundle.  The news-publisher.example page shows an ad for the targeted cohort.
+
+## Tradeoffs
+
+As compared to the initial proposal for Proprietary Cohorts, the TURTLEDOVe style workflow offer the following additional benefits:
+* The users cohortID can not be associated with any first party data and cannot be persisted after it has been cleared/changed.
+    * This adds a couple of TURTLEDOVes goal outcomes to Proprietary Cohorts. namely:
+        * People who wish to sever their association with the interest group can do so, and can expect to stop seeing ads targeting the group.
+        * Web sites cannot learn the interest groups of any specific people who visit them.
+
+These benefits come with the following costs:
+* Marketers cannot quickly adjust campaign goals as cached client side ads must have reasonably long ttls to have a change to serve
+* Additional bandwith and memory is needed on the client over-use of which could lead to degraded user experience, particularly on older devices.