Sharp Tightness — American-odds gap by mode

How sharp tightness works

The sharp-tightness gate blocks bets when the non-placement sharp books disagree among themselves on price by more than a configurable threshold (in American-odds cents). It is a consensus-coherence check, not a stale-line detector.

The check, end to end

1. Build a sharp-odds snapshot from the books in sharpBooks for this market.
2. Drop the candidate (placement) book from the snapshot — controlled by filters.sharpTightnessExcludeCandidateBook, hardcoded to true (marked required).
3. If fewer than 2 sharps remain, skip the check (not enough signal).
4. Compute best = highest American odds, worst = lowest, in the remaining sharps.
5. gap = americanOddsGap(best, worst).
6. Resolve the threshold: scoped match by book/league/market/mode/flow first, falling back to filters.sharpTightnessMaxOddsDiffLive (live) or filters.sharpTightnessMaxOddsDiffPregame (pregame). High-EV tail guard can tighten further.
7. If gap > threshold → FAIL. Otherwise PASS.

Worked example

Live MLB Cubs Over 13.5 placed at DraftKings −350. sharpBooks = [DraftKings, Pinnacle, FanDuel], with prices DK −350, Pinnacle −1389, FanDuel −1400. After excluding the placement book (DK):

• Remaining snapshot: { Pinnacle: −1389, FanDuel: −1400 }
• best = −1389, worst = −1400, gap = 11¢
• Threshold (live, no scoped match): 50. 11 ≤ 50 → PASS.

Note: the 1039¢ gap between DK (−350) and the consensus (~−1395) is invisible to this gate. That gap is the signal of a stale-line trap — but it is a candidate-vs-consensus distance, which this filter does not measure.

What this gate catches vs. doesn't

• Catches: markets where the (non-placement) sharps cannot agree among themselves on a price — a sign that fairOdds is being computed from noisy, contested, or stale inputs and is unreliable.
• Does not catch: markets where sharps tightly agree but the candidate book is a wild outlier vs. the consensus (the classic stale-line trap). For that pattern, use a separate candidate-vs-fair distance filter (not yet implemented).
• Skipped entirely for: player props (isPlayerPropDom returns true), markets with fewer than 2 remaining sharps, and any DOM lacking a current mode.

Threshold tuning

• Lower threshold = stricter consensus required. Cuts volume but raises ROI on what passes (the 14d sweep showed an empirical optimum around 28–30¢ on live; we run live at 50 as a compromise between coverage and tightness).
• Higher threshold = more tolerant. Lets through markets where sharps disagree more — fine if you trust the median sharp price, costly if disagreement signals stale data.
• Scoped overrides (in filters.scopedSharpTightnessMaxOddsDiff) match by combinations of book, league, market, mode, flow, and minSharpCount. The most specific match wins; ties go to the lower (stricter) value. Use these to keep liquid markets loose and illiquid markets tight.
• High-EV tail guard can tighten the live and pregame thresholds independently when the recent tail-health diagnostic flags as unhealthy and EV ≥ HIGH_EV_TAIL_GUARD.evCutoff. Effective threshold becomes min(base, tightened).

Implementation: shouldSkipForSharpTightness, buildSharpTightnessSnapshot, getSharpTightnessMaxDiff in osd-auto-edge-bb.js. Gate call sites: the EV/arb evaluator (badge pipeline), the high-EV tail diagnostic, and the watch-list near-miss path.

Data

Controls

Summary

Charts

Distribution