BEMXC Credibility Scoring: Elo-Based Trust Rating & R:R Verification Model

The flaw of raw win-rate

A 90% win rate with negative risk-to-reward is not excellence. It is a martingale in slow motion. Expectancy per decision is:

E[R] = p_win × avg_win − p_loss × avg_loss
     = p_win × (R:R) × risk − (1 − p_win) × risk

Let risk be 1 unit. If the provider’s typical winner is +0.4R and the typical loser is −1.8R, then at p_win = 0.90:

E[R] = 0.90 × 0.4 − 0.10 × 1.8 = 0.36 − 0.18 = +0.18

That case is still barely positive. Shift the payoff to the shape actually common in fake “90% accurate” FX rooms — winners of +8 pips against stops of 80 pips, i.e. realized R:R ≈ 1:0.10:

E[R] = 0.90 × 0.10 − 0.10 × 1.00 = 0.09 − 0.10 = −0.01

Negative expectancy at 90% hit rate. The arithmetic does not care about testimonials. One stopped-out loser cancels nine scalps, and the tenth scalp is profit only in the marketing copy. BEMXC therefore refuses to rank on p_win. The protocol ranks on whether the published payoff, once settled, can support a positive edge after a hard R:R floor.

Two further corruptions of raw win rate are sample censorship and payoff hiding. If losers can be deleted, p_win inflates. If stops can be widened after the fact, avg_loss shrinks in the brochure and not in the account. Both are structurally impossible once entry, TP, and SL are hashed onto the read-only log described in What is BEMXC.

BEMXC Elo rating mechanics

Every provider identity starts at Trust Score R = 1500. Each settled decision is an Elo match against a market opponent whose rating R_m is the rolling mean of the tape for that currency pair class (majors, crosses, gold). The expected score is the standard logistic:

E = 1 / (1 + 10^((R_m − R_p) / 400))

Observed score S is 1 on a qualifying win and 0 on a loss. Voids do not produce an Elo match; they increment the void counter used in demotion. The rating update is:

R_p ← R_p + K_eff × (S − E) × π_RR

K_eff = K_base × φ_freq × φ_dd
K_base = 32                    if n < 80 settled
        = 24                    if 80 ≤ n < 200
        = 12                    if n ≥ 200

Realized R:R modifier π_RR

Planned R:R is |TP − entry| / |entry − SL| at commit. Realized R:R is the signed excursion at terminal state, still measured against the committed SL distance as the risk unit. π_RR scales the Elo exchange so that a skinny winner is not treated as a full point.

Signal frequency φ_freq

Skill that cannot be repeated is not skill the protocol can rank. Let λ be settled decisions per rolling 30 days.

• λ < 4: φ_freq = 0.70 on wins, 1.00 on losses (sparse luck is discounted).
• 4 ≤ λ ≤ 40: φ_freq = 1.00 (healthy desk cadence).
• λ > 40: φ_freq = 0.85 (spam and correlated micro-scalps are damped).

Frequency is not a virtue by itself. The cap exists because a desk that publishes 200 EUR/USD micro-calls with near-identical hashes is farming sample size, not demonstrating independent skill.

Drawdown modifier φ_dd

Let DD be peak-to-trough sum of settled R-multiples over a 90-day window. φ_dd compresses win credit as pain accumulates and expands loss debit, so a desk in a violent equity hole cannot climb VIP on a short bounce.

φ_dd = 1.00                     if DD ≤ 6R
      = 0.75                     if 6R < DD ≤ 12R
      = 0.50 on wins / 1.15 on losses   if DD > 12R

Worked example

A PRO provider at R_p = 1680 publishes EUR/USD BUY, planned R:R 1:2.4, SL inside the 4× TP1 bound. Market opponent R_m = 1500. E ≈ 0.74. The call hits TP. π_RR = 1.00, n ≥ 80 so K_base = 24, cadence is normal (φ_freq = 1), drawdown is 4R (φ_dd = 1).

ΔR = 24 × (1 − 0.74) × 1.00 = +6.24
R_p ← 1686

If the same payload had planned R:R 1:1.1, ΔR on the win would be 0 and the R:R violation counter would increment. If SL had been 5× TP1, the win would still pay 0 Elo and raise an integrity flag.

Automated tier progression

Tiers are computed; they are not sold. A payment cannot purchase PRO or VIP. Marks are evaluated on every settlement and on a daily batch. Falling below any hard gate drops the mark immediately. The three provider tiers are ROOKIE, PRO, and VIP.

ROOKIE is not a punishment. It is the honest state of an unproven desk. PRO means the sample is large enough and the payoff structure is not a scalp-and-hope. VIP means the desk has survived a longer clock with contained drawdown. None of the three is a prediction that the next EUR/USD call will win.

Demotion and anti-gaming penalties

Demotion is public and automatic. BEMXC does not quietly delist a desk to protect a brand, and a desk cannot reset identity to escape a score. Linked accounts inherit the lower Trust Score.

SL wider than 4× TP1

Let d_tp = |TP1 − entry| and d_sl = |SL − entry|. If d_sl > 4 × d_tp, planned R:R is worse than 1:0.25. That structure is the textbook 90%-win / negative-expectancy product. The protocol flags the call at commit, pays no Elo on a win, applies an aggravated 1.25× loss debit, and increments the structural-violation counter. Five such calls in a 30-day window strip PRO/VIP immediately (−80 Elo).

Signal deletion penalties

Committed logs cannot be deleted. An API or UI attempt to delete, hide, or “unpublish” a hashed call is recorded as an integrity fault: −40 Elo per attempt, PRO/VIP stripped, and a 14-day publish cooldown after the third attempt. The original object remains in the win-rate denominator.

Inactivity decay

Skill that is not re-demonstrated mean-reverts. If settled decisions in 90 days fall below 20, Trust Score decays 8 Elo per week toward 1500, and VIP cannot be held. Decay stops when cadence resumes. Decay is not a ban.

A high Trust Score is not a forecast of profit. Elo on BEMXC measures consistency of verified, adequately paid-off FX decisions against the published tape. It does not measure your lot size, your broker’s spread, or your temperament. Those remain yours. See the execution disclaimer.

FAQ