[warnerktsang]

Problem

GET /espn/waivers reports games_next_7d = 0 for every player this season: the league payload carries no proGamesByScoringPeriod block, and the scoreboard fallback in _team_games_next_7d could only anchor period dates off that same block — no block, no anchor, no schedule. The schedule-edge pocket-pick rule has been dormant.

Fix

Derive the schedule from ESPN's public scoreboard (the ScoreboardClient the waivers route already injects):

• get_waivers now tracks a period anchor (period, date): captured when the current period itself is derived from the scoreboard's season start, and otherwise computed from first_regular_season_date when the payload carries no dated schedule at all and the period came straight from status. The anchor lookup is wrapped so a scoreboard outage there degrades to today's behaviour instead of failing the report.
• _team_games_next_7d(..., anchor=...) uses the block's own date map when it publishes one (unchanged precedence — the block stays authoritative), else the supplied anchor, to date each upcoming period for scoreboard.teams_on(day).
• Coverage is real: a period counts as covered only when a source answered for it, so the notes line now reports the true number (games_next_7d covers N of the next 7 days (the rest could not be determined); counts may understate).
• The pocket-pick schedule-edge rule itself is untouched — it activates naturally now that counts flow.

Notes / assumptions

• Team-id spaces line up, no remapping needed: scoreboard competitor ids are ESPN's canonical NBA team ids — the same space as fantasy proTeamId (scoreboard.py's documented contract). Counts are keyed by id end to end; the known 27/28 abbreviation discrepancy is display-only (PRO_TEAM_ABBR untouched).
• Cost: at most ~7 day-fetches per waivers call (plus the season-start month fetch, cached client-side and usually already paid by the period derivation) on top of the existing ~15 — no per-player fetches.
• The existing FakeScoreboard test double gained a teams_on (empty slate) so it models the full injected interface get_waivers now uses.

Tests

Four new tests in test_espn_waivers.py with a programmable FakeScheduleScoreboard (its slate mirrors the fixture pro-games block, so expected counts match the block-based tests):

• full week from the scoreboard with the block absent — free-agent and drop-candidate counts, no shortfall note, schedule-edge pick fires;
• anchor reuse when the period itself was scoreboard-derived;
• partial coverage (3 of 7 days raise) — counts from the 4 fetched days only, note says covers 4 of the next 7 days, report intact;
• total scoreboard outage — all counts 0, covers 0 of the next 7 days note, report intact.

ESPN scope locally: 531 passed (527 baseline + 4 new). Read-only; no new endpoints; no migration.