The settlement layer for predictions and skill on free-to-play.
You keep your games, users, brands, funds and KYC. PLLAY verifies the outcome from gameplay video, settles it and returns a settlement_id with a receipt. You pay a clearance fee only on pools that settle.
Recorded video first, benchmarked on 14 titles. Every live result has a human reviewer. Now scoping launch partners.
- Propose
- Vision frames
- Confidence vs threshold
- Receipt
The result is recorded.
- settlement_id
- 7f3ac91e-2b4d-4e60-8a15-c0d9e3f1a762
- confidence
- 0.987 · threshold 0.900
- evidence
- 00:41.200 → 00:44.800 · 8 frames
- model_version
- vision-vod-2026-09
Recorded case: Resolved automatically. State settled.
- What it verifies
Supported gameplay outcomes, read from recorded video. Live results need a reviewer.
See supported titles → - Why it matters
A settled result carries its evidence and a settlement_id. Uncertain evidence is held for review.
See the contract → - Start now
Recorded-video verification is built and benchmarked. Pulse, Duels and Arena are scoped with our team through enterprise access.
See the records → - Evaluate or buy
Run the sandbox, then scope an engagement with sales.
Open the sandbox →
From evidence to a verified result.
Pick an example and follow it from the challenge rule to the receipt. Verified means the result can enter a reward workflow. Held means a person reviews it first.
Sandbox demonstration
Sandbox fixtures, not production output. Every field is a settlement-contract field; no footage is shown.
The customer can use this result in its eligible reward workflow.
A result, not a payment. Funds and payouts stay with the partner.
Decision. The score is at or above the 0.90 threshold, so recorded video settles without a person. A model score for the outcome. It is not the probability that the result is correct.
A receipt records a decision. It does not move money.
Technical details
- Event definition
- type=Head-to-Head · path=vod · spec Sandbox Fixture (sandbox) · threshold 0.9
- Evidence reference
- https://resolve.pllay.io/e/5d0e8a14-3b7c-4f29-a6d1-9c2e47b0f385
- Decision state
- settled
- Score
- 0.987 against a threshold of 0.9. A model score for the outcome. It is not the probability that the result is correct.
- Model version
- vision-vod-2026-09
- Review status
- Not needed: recorded video at or above the threshold. reviewer_id is null.
- Settlement identifier
- 7f3ac91e-2b4d-4e60-8a15-c0d9e3f1a762
Start from what you run.
- Platforms and publishersOutcomes and evidence arrive through fragmented systems. Verify them one way, with review paths and auditable receipts.See the platform journey→
- Agencies and creator organizationsSponsored gameplay challenges, verified on evidence and delivered as a managed engagement.See the agency journey→
- Operators and event platformsContested outcomes need precise rules and traceable evidence inside your own resolution workflow.See the operator journey→
Why an Oracle?
A score tells you the result. A verification record shows how the result was established.
- Consistent rules
- Each event states its rule and threshold up front, and every result is judged against them.
- Traceable evidence
- A result points to the evidence it was based on, so a reviewer can look at what the decision rested on.
- Explicit handling of uncertainty
- Uncertain evidence is held for review. A hold is a recorded state that waits for review.
- Dispute and review records
- A reviewer’s decision and the reversal of a contested result are recorded with a reason.
- Structured outputs for applications and agents
- Results come back in a fixed structure through the API and webhooks, so an application or an agent can act on them without parsing text.
Where reliable, authorized game data exists, it can be a better evidence source than video analysis. Game telemetry is a planned evidence adapter and is not built today: every title is video-only.
Oracle verifies what happened in gameplay. By itself it does not prove advertiser ROI, purchases, installs or payment eligibility.
- OracleBuilt · enterprise accessVerification and settlement infrastructure. Checks evidence against supported rules, manages the decision state and returns an auditable result.
- Vision AIBenchmarked offline · recorded videoSupported visual evidence extraction from gameplay footage. It supplies the evidence; Oracle decides.
31 clips auto-settled, 0 wrong. 38 of 38 matched a human check across 14 titles.
Offline benchmark on recorded clips we chose; not a production test. Certification is underway on Fortnite, League of Legends and Valorant.
Listing a title, building its adapter, benchmarking it offline and validating it in production are four different things. They are counted separately here.
- 25Titles listedNamed on /coverage, including overlay-only titles with no Vision path.≠
- 14Vision adapters builtAn adapter exists in the codebase. Nothing is implied about accuracy.≠
- 14Benchmarked offlineRun on recorded clips with the results published. Not a production test.≠
- 0Production-validatedAn authenticated production test is on record in /status.
Offline benchmark · recorded video only · not a production test · as of 2026-08-10 · 38 clips across 14 titles
- Auto-settled: 31 of 38The model settled it at or above the threshold, with no person.
- Held as expected: 4 of 38Held below the threshold, and a hold was the expected result.
- Sent for confirmation: 1 of 38Sent to a person to confirm before it resolved.
- Voided: 2 of 38No usable result on the clip.
This is the offline benchmark. It does not establish production reliability, and no authenticated production test is on record.
- Recorded video only. There is no live-video result in this set.
- Public clips chosen by the team, 2 to 5 per title. Not a random sample and not production traffic.
- It shows the pipeline on these clips. It is not an accuracy rate for any title and not a production test.
- Recorded video
- Built and benchmarked; switched on per partner when their plan is signed.
- Live outcomes
- A reviewer is required, and live results are not benchmarked. Certified automated live is not available (Phase 2).
- Production validation
- Authenticated production tests: none on record yet.
Records and methodology → · Raw benchmark file → · Coverage by title →
From clip to receipt.
- 1Submit a clip
Send a recorded clip with its game adapter and the market.
Switched on per partner when their plan is signed. - 2Open the event
Create the event with its rule and threshold, and lock it while Vision AI reads the clip.
- 3Oracle decides
It settles, holds or voids against the rule and the threshold.
- 4Receive the result
A webhook carries the result, the evidence reference and, when settled, the settlement_id.
You keep funds, KYC and distribution. Oracle returns a result; it does not move money.
From verified outcomes to repeatable monetization.
Autopilot is the planned expansion from verifying outcomes to orchestrating campaigns and monetization on top of Oracle.
Roadmap preview. Not available today, and no autonomous campaign actions run for partners.
See the roadmap preview →Start with one recorded-video title.
Run the sandbox, then scope an engagement with the PLLAY team.