PRD: Verified Polymarket + Chainlink baseline research path

[proerror77]

Problem

The BTC 5-minute Polymarket research E2E is blocked because the currently selected immutable cohort has complete Polymarket event evidence and authenticated Chainlink reference prices but does not contain Binance aggregate trades or a compatible full Binance market tape. Requiring those unavailable surfaces prevents a ResearchSnapshot from being built, while fabricating them would corrupt provenance and execution claims.

Solution

Add one explicit, fail-closed Polymarket + Chainlink baseline profile. It consumes only externally anchored, verified immutable Polymarket evidence; uses authenticated Chainlink observations as the reference-price timeline; preserves each market as an independent episode with independent Up and Down token books; records Binance feature surfaces as intentionally omitted; and produces the existing ResearchSnapshot format for the existing evaluator and shared MCTS pipeline.

The full-surface Polymarket + Binance path remains unchanged and strict.

User stories

1. As a researcher, I can explicitly request the Polymarket + Chainlink baseline rather than having the system infer it from missing data.
2. As a researcher, I can build a snapshot from verified Polymarket evidence without inventing Binance trades, order-book updates, source clocks, or fills.
3. As a researcher, I receive one episode per market ID, with the Up and Down token order books kept separate.
4. As a researcher, I can evaluate final settlement probability from authenticated Chainlink reference observations and event-local Polymarket evidence.
5. As a researcher, I can evaluate Up-token and Down-token repricing separately from the same event snapshot.
6. As a reviewer, I can distinguish this baseline from a full-surface snapshot in immutable manifest provenance and quality metadata.
7. As a reviewer, I can prove that all Binance freshness indicators are false in the baseline.
8. As a reviewer, I can prove that omitting a required Polymarket token book, settlement, or usable Chainlink observation fails closed.
9. As an operator, I can still run the existing full-surface path, which continues to reject missing Binance inputs.
10. As an operator, I can pass the baseline snapshot into the existing evaluator and shared MCTS interfaces without introducing a second research engine.

Implementation decisions

• The public seam is verified immutable Polymarket evidence in and the existing ResearchSnapshot out.
• Baseline selection is explicit; missing Binance data never silently changes profiles.
• Authenticated Chainlink observations provide the baseline reference-price observation timeline.
• Derived feature rows may reuse authenticated Chainlink values, but they must never be represented as Binance observations.
• All Binance freshness and availability surfaces are marked absent or false.
• Polymarket raw evidence, token-book separation, settlement, trust anchors, and immutable readback remain fully verified.
• The existing full-surface validation contract is not weakened.
• The snapshot records an explicit baseline profile marker and an explicit intentional-Binance-omission quality marker.
• The existing evaluator and shared MCTS remain the only downstream research path.

Testing decisions

• Test the public CLI/profile selection and ResearchSnapshot boundary.
• Prove that the explicit baseline accepts anchored Polymarket evidence without Binance arguments.
• Prove that baseline mode rejects supplied Binance segments rather than ambiguously mixing profiles.
• Prove that the ordinary full-surface mode still rejects missing Binance inputs.
• Prove that baseline observations retain market and token identity and set every Binance freshness flag false.
• Add targeted counterexamples for missing one token book, missing settlement, and unavailable, future, or out-of-window Chainlink reference evidence.
• Validate immutable snapshot serialization and digest readback before running the evaluator.
• Run focused tests first, then the affected package suite and repository-prescribed checks.

Out of scope

• Collector code, deployment, or cutover.
• Synthesizing or backfilling Binance aggregate trades, LOB sequence IDs, receive clocks, or fills.
• Changing the full-surface verifier or weakening its required surfaces.
• A new ResearchSnapshot schema or a second evaluator/MCTS implementation.
• Paper or Live execution, OMS, RiskGate, promotion policy, or profitability claims.
• 15-minute and 1-hour products.
• Optional historical Binance feature enrichment.
• Retirement of duplicate loop state, which remains owned by Cut over prediction research and retire duplicate loop state #189 after parity evidence.

Notes

This PRD unblocks the existing E2E with the data that is actually present. It is a settlement and token-repricing research baseline, not a claim of Binance-informed execution fidelity.

[proerror77]

Implementation clarification: the verified evidence currently exposes each event-local Chainlink opening anchor, while the independent Polymarket Up/Down book availability clocks drive baseline observation times. The baseline does not invent a continuous in-event Chainlink price path and does not label the derived feature clock as Binance. Any later continuous Chainlink enrichment is out of scope and requires its own verified input contract.