[kartojal]

Market makers can merge ABOVE on a lower threshold with BELOW on a higher threshold to receive pUSD and the middle-bucket YES positions, then reverse it with the same two IDs. Ports the current position-ID workflow from TypeScript PR #428.

pairs = client.find_directional_pairs()
if pairs:
    selected = pairs[0]
    transaction = client.merge_directional(
        selected.above_position_id,
        selected.below_position_id,
        amount=selected.amount,
    )
    transaction.wait()

• Sync SecureClient is the default; AsyncSecureClient offers the same methods with native async transport. Both accept unordered wing IDs, keyword-only base-unit amounts, and a default of "max". Splits check pUSD and every consumed bucket.
• Python prepare_* methods return the existing immutable TransactionCall for execute_transaction or custom signing, rather than copying the TypeScript signing generator. Models use frozen dataclasses, output enums, and plain-string input vocabularies.
• Include pair discovery with fresh on-chain balances, local ID decoding/derivation, bucket/Void and whole-event operations, threshold compose/decompose, and condition-based split/merge/redeem without a market listing. Pair candidates can overlap and are not an execution plan.
• Add the four dedicated activity models, directional resolution kinds, a usage guide and an offline example. Existing approval and EOA / Deposit Wallet / Safe / Proxy execution paths are reused. Collateral-return service planning remains deferred.

Validation

uv run ruff check ., uv run ruff format --check ., uv run pyright, uv run pytest -m "not integration", uv build, and the Sphinx reference build pass. 3,573 tests pass; focused coverage pins Router selectors against the TypeScript ABI and exercises unordered pairs, balance limits, exact Void-inclusive baskets, all four wallet paths, paginated discovery, activity economics, and local directional redemption.

The offline directional example runs. Added an opt-in metered staging round trip using shared integration fixtures; it skips without metered opt-in and credentials. Live Python staging execution remains unverified. The test seeds inventory using line splits and does not establish trading/fill settlement.