ROUND22 actualpacket.21/30complete. ### R22-MR-01 — Measure Invitation-Caused Play, Not Views Use an **eligible-recipient denominator** defined before results: people who could validly receive the approved invitation, had not already entered that specific branch/encounter, and received it through the tested sharing path. For a bounded experiment, compare voluntary invitations against a matched condition where comparable eligible recipients encounter the same underlying playable opportunity through ordinary discovery **without the personal story invitation**. Do not withhold already-promised access or force anyone to share. Measure separately: **Delivery → invitation opened → entry preview inspected → branch voluntarily started → first meaningful action performed → bounded encounter completed → later voluntary continuation.** A view alone counts only as viewing. **Attribution** asks: *Did this recipient arrive through this invitation?* **Incrementality** asks: *Would comparable recipients have meaningfully played without it?* The first can be measured from the tested entry path; the second requires comparison and remains probabilistic. Personal provenance, story quality, sender-recipient familiarity, placement, prior game interest, and checkpoint friction are confounds and should be recorded where available without invasive relationship profiling. Privacy boundary: sender sees **no recipient-level open, play, completion, or decline receipt** unless the recipient separately chooses to return something under R07/R08. Experiment analysis should use privacy-preserving aggregate results with minimum reporting thresholds where needed; analytics permission itself must be disclosed appropriately. Track incremental meaningful starts/completions alongside generation, storage, moderation/review, checkpoint-support, and measurement costs. **Strongest failure:** invitations generate high opens because personal stories are emotionally compelling but produce little additional meaningful play—or analytics quietly turn sharing into social surveillance. **Falsifiable test:** if invitation-exposed recipients show no credible lift in meaningful starts/completions versus comparable ordinary discovery after accounting for major confounds, narrow the growth claim. Even strong lift establishes this invitation mechanism under tested conditions—not guaranteed virality or a universal growth loop. **Godz2:R22-01 — Test additional play without turning invitations into surveillance** **Proposed comparison:** after a player independently chooses to share, offer an approved story invitation with supported entry or a plain, honestly described invitation to the same supported branch. Randomize the offered format only within a consented study where both are acceptable; allow withdrawal. Keep recipient access, price, and checkpoint conditions equivalent. This estimates the story format’s added value over a plain invitation—not sharing’s total effect versus no invitation. Use assigned willing sharers as the primary denominator, including those who ultimately send nothing. Separately report delivered invitations only where delivery is actually known. Clicks cannot establish how many humans saw an externally copied file. Distinguish consenting recipients’ first arrival, meaningful start, and satisfying completion; define meaningful play beforehand as performing a consequential supported interaction, rather than merely loading the scene. An invitation reference can associate arrival with a route. It cannot establish that the person would otherwise never have played. Prior familiarity, multiple invitations, existing intent, recipient selection, and timing remain confounds. Ask minimal optional questions about prior play and intended participation; do not construct social graphs. Measure only with recipient consent. Keep study records separate from game histories, minimize retention, and publish aggregates that cannot reveal individual activity. Senders receive no usage receipt, completion notice, or individual conversion dashboard. Declining measurement preserves access; report missing observations rather than treating them as non-play. **Strongest failure:** attractive clips increase curiosity clicks while disappointing recipients or making creators feel responsible for recruitment. Record generation, delivery, support, and study costs alongside additional meaningful starts and satisfaction. No rewards for views or compulsory sharing. **Falsifiable check:** if the story condition increases clicks but produces no credible improvement in meaningful play—or increases mistaken entry expectations—revise the invitation. A small inconclusive result cannot establish virality or rule out useful private sharing. # Round22 preparation — not performed F-R22-01: growth goal is incremental valued play, not rawclipviews or invites. Eventlinked share permits private draft, chosenaudience, honestentryrequirements and decline without creatorusageleak. No forcedposting to finish, unlockbasicexport, earnlove or receivefundedaccess. Recipient can view without paidinference or entering life. Measurement distinctions: eligiblehumanview, optionalclick, supportedlanding, acceptedentry, firstmeaningfulaction, satisfyingcompletion, voluntaryreturn. Botpreview/linkscanner and creator's ownchecks don't becomeplayers. Track share/acquisition linkage only with applicable dataauthority, shortlived minimal IDs, aggregated cohorts; do not infer hidden socialgraph or report recipientactivity tosender. Measurementconsent distinct from storypublication. Comparison: matched eligibleplayers randomly offered optional sharetool vs ordinaryprivatecompletion, no sharingpressure; compare actualadditional eligiblehumanstarts/valuedplay under equallyclearlanding. Or staggered rollout with labeled confounds. Moreviews don't proveincrementality when same people wouldplayanyway. Availability/content differences must not be credited toinvitation format. Cost per incrementalvaluedplayer includes mediaattempts, distribution, bandwidth, support and acquisitionops, with uncertainty. No causalclaim from attributedlastclickalone. Paidpromotion separatefromorganic/referral; giftedaccess funding separatefromrecipientinterest. Test dead/deniedcheckpoint, mobileentry, sharedpreviewbot, repeatopens, deniedtracking, crossdevice consent. Ensure plainview works, truthfulfallback no baitandswitch, no surprisecharge and no privatepayload inURL. If sharing risesbut recipients don't reachmeaningfulplay, fix promise/entry or stop media spend. Networkeffect requires later evidence others' participation improves value, notsharecount. Godz4 initialresponse [internal record ID omitted] ### Round 22 — Measure Meaningful Entry, Not Sharing Volume **Godz4:R22-01 — Separate attribution from incrementality.** Define the denominator before launch: **eligible recipients who actually receive an approved invitation**, deduplicated as humans where reasonably possible. Track separately: invitation delivered/opened → clip viewed sufficiently to understand the premise → entry offer inspected → supported branch accepted → **first meaningful player-controlled action completed** → voluntary continuation/ Actual requested missingtail [internal record ID omitted] continuation at a later supported point. Views alone are not success. **Concrete comparison:** attribution asks, “Did this entrant arrive through this invitation?” Incrementality asks, “Would comparable recipients have played anyway?” Use fresh, comparable eligible cohorts where feasible: approved story invitation versus a neutral/direct invitation to the same supported starting opportunity. Keep access, cost, audience eligibility, and timing comparable. Do not withdraw promised access or manufacture scarcity. **Privacy:** the sender receives no recipient-level open, play, branch, continuation, or identity receipt. Measurement uses the minimum event data necessary for aggregate experiment analysis, with applicable consent/retention boundaries. Report only sufficiently aggregated outcomes. A recipient may play silently. **Costs:** separately record clip preparation/generation, moderation/permission review, delivery, checkpoint/runtime processing, support, and measurement overhead. Incremental play that costs more to acquire/service than its demonstrated value is not automatically a growth engine. **Strongest failure:** highly emotional invitations produce excellent attributed conversion because they target already-interested friends, then get mislabeled as viral acquisition. **Falsifying test:** if invitations generate many views/opens but no credible lift in first meaningful actions or voluntary continuation versus the comparison condition—or produce misunderstanding/pressure—the viral-growth hypothesis weakens. Positive attribution without incremental evidence remains distribution evidence, not proven growth. Godz3 actualpeercritique F-R22-02: above draft now actualcontribution. Prefer Godz2 randomized offeredformat among consentedwilling sharers with intent-to-treat denominator; estimates storyformatincrement overplaininvitation not totalsharingeffect. GodzI ordinarydiscovery contrast asks largerquestion but recipientselection/exposure mismatches confound. Godz4 directinvitationcontrol similarly cannot calldifference totalviralimpact. Deliveredrecipient count unknown for externalfiles; don't fingerprint for perfectdedupe. Track consenting observations/missingness, not nonconsent=zero. Meaningfulplay doesn't require irreversibleconsequence; deliberate supported observation/refusal can matter. Sufficientaggregation thresholds must consider tinygroups/reidentification. Returnrecipientoptionalnote not covertconversiondashboard. Paid/unpaid acquisitioncost compared to measuredvalue not assume allvalue monetary.