Build the autonomous trust index and owned daily game

[dgyy]

Problem Statement

ArcadeBloom currently publishes a catalogue of 2,019 outbound browser-game entries, but most pages lack direct evidence, screenshots, differentiated claims, and current compatibility observations. Search engines have indexed only a small fraction of the site, generic game-directory intent is controlled by hosted portals and authoritative editorial domains, and the present provisional-indexing exception allows weak pages into the sitemap. The owner requires the entire operating loop to run without routine human participation and will intervene only for authorization checkpoints that cannot be delegated.

Solution

Rebuild ArcadeBloom around two connected products. First, turn the third-party catalogue into a browser-game trust index whose public claims, eligibility, discovery cohorts, and status history come from recurring automated browser and source evidence. Second, add one ArcadeBloom-owned daily game with deterministic challenges and repeat-play mechanics as the owned retention and monetization surface. Catalogue membership no longer implies search eligibility: every entry fails closed unless current evidence passes the gate, and every generated discovery page must represent a sufficiently populated measured constraint cohort. The system discovers, captures, validates, publishes, monitors, measures, rolls back, and reallocates effort autonomously.

User Stories

1. As a browser-game explorer, I want to know whether a game was recently observed to load, so that I do not waste time following dead links.
2. As a mobile player, I want device-specific observations, so that a desktop result is not presented as proof of mobile compatibility.
3. As a player with a preferred input, I want measured keyboard, mouse, touch, and gamepad signals, so that I can choose playable games.
4. As a privacy-conscious player, I want to see login requirements, cookies, and contacted third-party hosts, so that I can make an informed choice before leaving ArcadeBloom.
5. As a low-end-device player, I want measured transfer size, request count, and load timing, so that I can find lightweight games.
6. As a player, I want every observation to show when and how it was collected, so that I understand its scope and freshness.
7. As a player, I want broken or uncertain games clearly excluded from recommendations, so that catalogue size does not override trust.
8. As a search visitor, I want discovery pages based on a real measured condition, so that the list answers a concrete compatibility question.
9. As a search visitor, I want undersized or arbitrary filter combinations excluded from search, so that I do not encounter thin pages.
10. As a game creator, I want a free public compatibility report and evidence history, so that I can discover regressions without manual correspondence.
11. As a game creator, I want an embeddable verification badge whose state updates automatically, so that visitors can inspect current evidence.
12. As a paying creator, I want self-serve scheduled monitoring, private-URL checks, change alerts, and machine-readable reports, so that I can use the service without sales or support calls.
13. As a creator, I want payment to remain separate from evidence and editorial rank, so that verification stays credible.
14. As a returning player, I want one new deterministic ArcadeBloom challenge each day, so that I have a reason to return.
15. As a player of the owned daily game, I want a local streak and shareable result that do not require an account, so that replay and sharing remain frictionless.
16. As a player, I want the same dated challenge to be reproducible, so that results are comparable.
17. As the owner, I want evidence capture, eligibility, deployment, monitoring, rollback, and growth allocation to operate unattended, so that the site does not become a job.
18. As the owner, I want failures and quota exhaustion to pause or fail closed, so that automation cannot publish unsupported claims or create unapproved charges.
19. As the owner, I want only genuine authorization checkpoints escalated, so that routine product decisions never wait for me.
20. As the owner, I want search, outbound-click, repeat-use, monitoring activation, and revenue signals to drive automatic stop/continue decisions, so that poor directions do not consume resources indefinitely.
21. As the operator, I want every automated publication to pass a small-batch pull-request gate and post-deployment health check, so that errors have a contained blast radius.
22. As the operator, I want a durable audit trail connecting each public claim to its evidence and workflow run, so that failures can be diagnosed and claims withdrawn automatically.
23. As the operator, I want source access to respect licences, robots directives, published APIs, and rate limits, so that autonomous discovery remains within explicit boundaries.
24. As an advertiser or payment provider, I want ads and paid placements separated from editorial eligibility and play actions, so that monetization does not mislead users.

Implementation Decisions

• Third-party games remain outbound-only. Only ArcadeBloom-authored games may be hosted by ArcadeBloom.
• The frozen 2,019-entry manifest becomes a historical cohort marker, not an indexing entitlement. Registry state and current valid evidence are authoritative.
• Index eligibility is a single fail-closed policy module consumed by robots metadata, sitemap generation, discovery cohorts, feeds, cards, and publication checks.
• Evidence records are versioned, immutable observations. A new run appends an observation and may change derived status; it does not rewrite historical evidence.
• Browser evidence records distinguish attempted, observed, inferred, unsupported, and unknown states. Missing evidence never becomes a positive claim.
• Initial measurements cover final URL and HTTP outcome, rendered state by viewport, screenshot artifact, console/page errors, request count, transferred bytes, timing, interaction attempts, login signals, storage/cookie signals, third-party hosts, source identity, and licence evidence.
• A normalized trust projection derives current public facts from the newest valid observations while preserving history.
• Constraint cohorts are allowlisted measured predicates with minimum membership and freshness thresholds. Arbitrary combinations are not published.
• Evidence artifacts and registry updates use one canonical directory and artifact contract across capture, assessment, publication, and CI workflows.
• The capture step must create real screenshot artifacts and verify their hashes; metadata that merely names a screenshot is invalid.
• Assessment can summarize observed evidence but cannot override deterministic failures such as no render, missing source identity, missing licence status, missing screenshot, expired observation, or integrity mismatch.
• Publication creates small pull requests, uses required checks, auto-merges only successful batches, deploys only from the protected production branch, and automatically reverts a failed post-deployment batch.
• Recurring monitoring prioritizes previously index-eligible games, then high-potential candidates. Quota exhaustion pauses lower-priority work without weakening thresholds or enabling paid overage.
• The owned daily game uses a versioned deterministic seed derived from date and ruleset version. The result can be reproduced without storing personal identifiers.
• Owned-game analytics and outbound-click analytics remain first-party and aggregate; no account or cross-site profile is required.
• The first monetization surfaces are the owned game and evidence-rich public reports. Developer monitoring is self-serve; manual sponsorship and bespoke work are excluded.
• Paid monitoring and paid placement are separate products. Payment cannot make a game index-eligible or improve organic ordering.
• Automated stop/continue rules are evaluated after a stable observation window and recorded as machine-readable decisions.
• Authorization checkpoints are limited to identity, tax/payment onboarding, accepting account terms, new spend, secrets unavailable to the agent, and destructive production operations with material consequences.

Testing Decisions

• The primary end-to-end seam is one complete trust-index batch: candidate input must result either in an evidence-backed, index-eligible page included in the sitemap or a fail-closed page with noindex,follow excluded from the sitemap.
• Tests assert externally visible policy outcomes rather than internal helper calls.
• Registry precedence tests prove that explicit ineligibility and expired/invalid evidence override historical manifest membership.
• Evidence validation fixtures cover successful observations, metadata-only records, missing screenshot files, hash mismatch, load failure, partial viewport results, stale evidence, redirects, and network timeouts.
• Artifact-contract tests pass the output of capture directly into assessment and publication using the same layout used by CI downloads.
• Static-site smoke tests assert robots metadata, sitemap membership, evidence display, outbound-link safety, content without JavaScript, and absence of unsupported claims.
• Constraint-cohort tests prove minimum membership, freshness, predicate correctness, canonical URLs, and suppression of arbitrary combinations.
• The daily-game seam uses fixed clocks and seed vectors to prove same-date reproducibility, date rollover, ruleset versioning, scoring determinism, and local streak behavior.
• Privacy tests assert that owned-game and outbound metrics do not set identity cookies or transmit prohibited personal fields.
• Workflow tests or static validators assert correct branch propagation, artifact paths, required checks, batch limits, concurrency, free-tier boundaries, and rollback triggers.
• Existing Playwright smoke tests, evidence-validator fixtures, indexability tests, registry validation, collection validation, and model-adapter tests provide prior art and should be extended rather than duplicated.

Out of Scope

• Hosting, mirroring, or embedding third-party game builds.
• Building a generic Poki/CrazyGames-style portal or creator marketplace.
• Mass-producing owned games.
• Human-written routine reviews, manual game selection, manual screenshot QA, cold outreach, bespoke sales, or manual content calendars.
• Author-supplied compatibility claims being accepted as verified observations.
• Arbitrary programmatic SEO pages and generic “best/free/no-download browser games” expansion.
• Payment influencing evidence, eligibility, organic rank, or trust labels.
• Unapproved paid infrastructure or silent overage.
• Circumventing source robots rules, access controls, platform terms, or rate limits.

Further Notes

• ADR-0010 records the product transition and supersedes ADR-0006's indexing entitlement.
• The complete competitor research is stored in the repository under the dated competitor-opportunity report.
• Initial delivery should first repair the eligibility and evidence artifact seams, because every later page, cohort, report, and monetization feature depends on those facts being trustworthy.
• After the trust pipeline is sound, deliver one thin vertical public report, then constraint cohorts, then the owned daily game, and only then self-serve paid monitoring.