feedback-betting-discipline

When helping with sports betting, the data-grounded reads held up; the narrative/macro overlays repeatedly misled and cost credibility.

Why: Over a full live session (WNBA + an NBA Game 7), every recommendation that survived came from ground truth — confirmed lineups, last-5 game logs, role/usage logic. Every miss was a confident overlay: "game total Over" on depleted rosters, an injury "catalyst" for a player already out of the rotation, "roster contamination" that was actually my stale training-data knowledge vs correct live data, and labelling a volume-derivative prop the "single best bet." The user caught each one.

How to apply:

• Never trust my own roster/rotation/team knowledge over live system or user-supplied data. Training cutoff makes it stale; live data wins. If they conflict, the live data is right and I say so.
• Lead bet recs with lineup-confirmed + recent-log-validated player props. Explicitly tag any "this whole game will do X" / total / "catalyst" / "best bet" take as low-confidence and make it earn its place.
• Push singles over parlays, every time — quantify the counterfactual (their reads were mostly right; the parlay vehicle, not the picks, lost the night). For a usage/role read, push POINTS/PRA over threes or other volume derivatives.
• Do not reinforce near-miss / "so close to $X" framing — name it as the trap and reframe a multi-leg miss as the expected loss it was.
• [internal detail removed]

Usage-vacuum allocation (added 2026-08-03, graded live): On the 8/02 Hall-out news (IND) I allocated the vacated wing minutes by narrative — bet the sixth-woman Cunningham O 9.5 (+118, scored 3, worst call of the day) and endorsed the books' fade of the replacement starter Hull (over would have paid +126). The actual absorber was THE STAR: Mitchell scored 37 against an UNMOVED −105 line. Rule: when lineup news breaks, usage flows to the best player first, not the next name on the depth chart — and the tell is whose line did NOT move after the news. I even noted "Mitchell marginally better, price unchanged" and failed to act on it. Off-system discretionary overlays keep losing; the priced, system-flagged reads keep holding (Allemand +edge pick won the same night).

Bench-absorber rule (2026-08-06, now 0-2 live): Carrington O 12.5 (8/05, LOSS: 6p in 12 min) repeated the Cunningham error (8/02, 3p) EXACTLY: betting the bench scorer to absorb a vacuum. Vacuum minutes go to the PROMOTED STARTERS, and the promoted-starter overs are the class that pays (Sheldon O 6.5 → 15p same night; Sabally 8/04; backtest 61% at run ≤ 2). Also: the "unmoved line = lag" tell is NOT a rule — it found Mitchell (8/02 win) and trapped Carrington (books were right). Rule: bet the promotion, never the bench absorber; the unmoved line is a prompt to investigate, not a signal to bet. And the news-kill overcorrect: my 3 vacuum-kills on CHI unders went 2-1 WOULD-HAVE-WON — "usage up" does not mean the survivors boom (CHI's survivors scored 15/15/9).