{
 "what": "The readers' experiment of the Markets page: group 0 sees the AIs' call before its first call of the session, group 1 only after it; both are scored on the same outcome as the AIs. This file is the scheme of how a reader gets a group, what is stored and for how long, what counts as a bot, and the unit of analysis. Read from the running votes worker at every build.",
 "unit": {
  "dependent_unit": "the reader (the anonymous voter id)",
  "id": "a random id the reader's browser makes and keeps (localStorage); no cookie, no account, no IP, nothing personal",
  "id_pattern": "/^[a-z0-9]{16,40}$/",
  "clusters": [
   "reader (voter id) — several calls of one reader are not independent",
   "address of the day (salted hash of IP + day) — readers behind one address"
  ],
  "analysis": "intervals clustered by reader (and, as a check, by the address of the day); a reader's first call of a session is the only call that counts"
 },
 "group": {
  "rule_now": "the sum of the character codes of the anonymous voter id, mod 2 (the page and the worker compute the same; the id is made by the reader's browser) — predictable: a reader who picks the id picks the group",
  "rule_planned": "group = the first byte of sha256(SALT_G ‖ voter) mod 2 — a reader cannot pick the group",
  "salt_sha256": "f8879b67fc4ed3ab975173cfbb842df6676d29e34dd9de1a37906589f9a6cd91",
  "salt_note": "the sha256 of the salt is printed here before the worker uses it; the salt is published when the rule ends, so every group can be recomputed",
  "status": "in force",
  "from": null
 },
 "retention": [
  {
   "table": "votes",
   "fields": [
    "id",
    "voter",
    "side",
    "ts"
   ],
   "kept": "for the tally of “how other readers answered”; no IP"
  },
  {
   "table": "mktx",
   "fields": [
    "id",
    "voter",
    "day",
    "h",
    "g",
    "side",
    "ms",
    "c",
    "ts",
    "pk"
   ],
   "kept": "the first call of a session, its group, time to decide and confidence — for the experiment; h is the salted hash of IP + day (the IP itself is never stored)"
  },
  {
   "table": "ipday",
   "fields": [
    "h",
    "day",
    "n"
   ],
   "kept": "a count per salted hash of IP + day, for the daily caps"
  },
  {
   "table": "viewseen",
   "fields": [
    "h",
    "day"
   ],
   "kept": "a salted hash of IP + day + id, so one address counts one view a day"
  }
 ],
 "retention_rule": {
  "now": "address hashes older than a day are deleted daily",
  "planned": "delete h (mktx), ipday and viewseen rows older than one day, every day; the experiment's rows keep the call, the group and the day, without h",
  "status": "in force"
 },
 "bots": {
  "now": [
   "only calls from the page count (they carry the time from opening the page, 0 to 24 h)",
   "at most 2 first calls of one address a session",
   "at most 200 votes of one address a day",
   "the voter id must match its pattern"
  ],
  "planned": "a call without the page's token is kept with bot_suspect = 1 (never dropped); the main analysis without the mark, a variant with every call",
  "status": "in force"
 },
 "published": "totals by event and group (council-experiment-funnel.json and the book's “experiment” lines) and, from №592, one anonymous row per first call (council-reader-first-calls.json): a pseudonymous reader key made in the worker with a secret of its own that is never published (not the voter id, not the group salt, not an address hash; the same reader keeps the key while its fingerprint is unchanged; deletion on request), the market, group, side, confidence, time to decide and the final call; no voter id, no address, no exact time stamp",
 "shown": "a methods file; not on the first screen of the site"
}