docs: add Seal as an RL reward

[abrichr]

Adds docs/commercial/seal-reward.md, "Seal as an RL reward", to the Commercial section beside The Seal and the Execute page, with back-links from The Seal and Effect verification.

What the page states

• The rule: reward 1 only on a tier-2 or tier-3 verified; screen-only evidence is never certified; a reward receipt is not an Execute Seal; unscored episodes are dropped, never paid 0.
• What runs where: organization worker, OpenAdapt control service, trainer node.
• The outcome-to-scalar table and the certified rule, including calibration_scope (synthetic today; production needs the unpublished Phase-1 calibration). No epsilon or delta value appears on the page.
• Wiring: the OpenAI grader route, TRL CertifiedRewardFunction, verl CertifiedRewardManager, and the openadapt-mockmed-extradup verifiers environment.
• The certificate fields, expiry in policy updates, re-certification, and a link to the public preregistration.
• The MockMed run and the evals proof command.

Where every command comes from

Item	Source
RewardContractV1, RewardCertificateV1, RewardOutcomeV1, RewardEvidenceReceiptV1, score, the outcome table	openadapt-types#37 (merged, 0.16.0)
calibration_scope, issuer, production_certified	openadapt-types#38 (merged, 0.17.0)
HttpRewardEndpoint, CertifiedRewardFunction, metadata_columns, as_async, CertifiedRewardManager, register_with_verl, python -m openadapt_evals.reward.proof	openadapt-evals#326 (merged)
environments/openadapt_mockmed_extradup, vf-eval, scripted_policy.py serve, check_fails_closed.py	openadapt-evals#324 (merged); the git-subdirectory install was run locally on 2026-09-02
openadapt-flow serve-reward, --contract, --port, --seed-mockmed, the routes, the envelope fields, the MockMed descriptors	openadapt-flow#452 (merged, f90f723b)

Checks run locally

uv run mkdocs build --strict, uv run python scripts/validate_docs.py, and tests/test_validate_docs.py pass.

Opened by an agent session, not the founder.

🤖 Generated with Claude Code