Scale-in validated is the canonical view for strategy decisions
— multi-tranche groups that pass conviction + fresh-price guards keep their full P&L.
Single-tranche stress is the worst-case bound (strict de-dup clamp).
Live-realistic adds tradability filters
(min-bet, disabled markets, liquidity ceiling). Headline is the raw book.
live
Headline
£0
—
Staked
—
Settled
—
—
—
vs Headline:
—
Scale-in validated
£0
—
Staked
—
Settled
—
—
—
vs Headline:
—
Single-tranche stress
£0
—
Staked
—
Settled
—
—
—
vs Headline:
—
Live-realistic
£0
—
Staked
—
Settled
—
—
—
vs Headline:
—
Assumptions & methodology
Headline
loading…
Scale-in validated
loading…
Single-tranche stress
loading…
Live-realistic
loading…
Pipeline diagnostic
Prediction-to-Bet Sync
Arena runs generate predictions; add_bet() persists paper bets.
Gap = something between them is blocking. Current likely cause: risk gate + stale open bets holding concentration.
live
Last paper bet
—
placed at: —
Sports producing runs
—
of —
Risk gate
—
—
Open bets blocking
—
£0 stake
By arena
Arena
Runs
Last run
Last bet
Gap
Status
loading…
Stale open bets (holding risk gate)
loading…
Backfill accounting (attempted vs inserted vs skipped)
loading…
Source: /api/bets/sync-health · refreshed every 60s · arena runs scanned from
data/arena_runs/<sport>/ · bets from bets.db ·
risk gate from current_risk_status().
Multi-tranche groups that failed the in-play scale-in guards.
Conviction drop = a later tranche's
model probability fell below
prior_max − INPLAY_CONVICTION_DROP_TOLERANCE.
Repeated price = all tranches share
one observed-price snapshot (no fresh refresh between triggers).
Post-guard failures are what matter for tier_1a / live readiness.
Historical guard failures (all / 30d)
include pre-2026-04-26 data when the freshness + conviction guards weren't yet enforced — useful as a warning, not as current enforcement evidence.
historicalshowing all time
📊Post-guard window — these are the failures that matter for live readiness
Multi-tranche groups
—
Conviction drops
—
Repeated prices
—
Either failure
—
P&L on failing groups
—
Filter:live
loading…
Failing P&L on its own does not invalidate scale-in — high conviction can persist on stale prices.
It does mean the validated number is leaning on guards that were not technically met.
Once price_refresher capture rate & conviction monotonicity
run cleanly forward, this list will shrink toward zero on new books.