Play Clock API

Report liveness, the resolved NFL week, and how fresh the ingested data is. ``status`` is ``"degraded"`` when ``meta/freshness`` is empty or any marker exceeds its dataset SLA. The process may be up while its scheduled ingest is stopped, which is still worth an alert before paid answers use stale data. The preseason gap is applied here for the same reason it is applied at the paid gate: a dataset upstream has not published yet is not stale, and reporting it as such means health reads ``degraded`` for weeks over a condition the API itself has decided is not a fault — which trains an operator to ignore the one signal that should wake them. It is *reported* rather than silently dropped, in ``unavailable_datasets``, so the state stays visible without being an alarm. The exemption expires with the gap marker after ``GAP_TRUST_SECONDS``, so a dead ingest still turns this red, and it is believed only for datasets the store can still show a row for -- a cold preseason store reports them missing, which is what they are.

Return every endpoint with its price, schemas, cache policy and payment config. This is the agent-facing discovery document (PRD §4.1). It is assembled from the same :data:`api.x402.ENDPOINT_SPECS` the 402 challenges are built from and the same price table the payment layer charges against, so a price shown here is by construction the price a caller will be quoted.

Return the top movers with identity and trend count only. Adds come first (that is where waiver interest lives); drops top the list up when fewer than ``limit`` adds have been ingested. No engine runs here — the rows are read straight from the ``trending/{add,drop}`` documents the 30-minute poll writes, joined to ``players/`` only for identity fields the poll did not already carry.

Free proof of life from settled receipts without an unbounded hot-path read. Production reads 32 per-network rollup shards after the explicit historical backfill. Before that marker exists, this retains the exact legacy scan: an incomplete migration must never make a public payment total look smaller.

Top trending adds and drops with per-player stat context and a verdict.

Eight to twelve low-rostered starts for the week, with confidence tiers.

Ranked waiver targets with stash/start labels and suggested FAB bids.

Emerging players, injury fallout chains, stock up/down, rookies, streamers.

Last-four-week trends, usage trajectory, schedule outlook and a verdict.

Rank the supplied players best-to-worst start for the week.

Grade a roster, call the lineup, and name the adds available in that league. With a ``sleeper_username`` the roster and the league's free-agent pool are pulled live from Sleeper; with a pasted ``roster`` no league is known, so ``free_agents`` is left **absent** rather than empty — absent means "we do not know who is available", empty would mean "nobody is", and the engine treats those differently.

Positional strength vs. leaguemates, league-available fixes, manager review. Every league-relative number is computed in Python by :func:`api.data.team_analytics.build_team_report_facts` from Sleeper matchup history and passed to the engine as ``team_analytics`` — the engine narrates those numbers and never produces them (tech spec §6). That history covers completed weeks only; see ``history_week`` below.

A tiered draft board with the market's values and reaches called out.

Grade one drafted roster pick by pick against where the market had them.