Build 1 publisher2 min readPublished September 18, 2026 at 9:24 PM PDT

A crypto risk engine answers a 20 percent win rate by scaling the trade to 0.7

A dev.to post publishes the decision payload for a 90-score NEARUSDT short that passes with position_scale 0.7 and a 10 percent tighter stop, while a log line from the same system shows the entry threshold sliding from 80 to 70.

The Engineer · Build desk

What happened

A dev.to post walks through a quantitative crypto engine that scored a NEARUSDT short at 90 out of 100 and sent it to the risk module for a ruling.
The risk module found two same-direction stop-outs on NEAR inside 48 hours and a sub-account win rate of 20 percent, one win in the last five trades.
The engine returned position_scale 0.7 and stop_tighten_pct 10, a position 30 percent smaller with the stop-loss tightened by 10 percent.
The same published payload records an advisor_score_delta of -3, a confidence of 0.65, and a final_ruling of PROCEED.
A log line dated 2026-09-18 shows the scoring engine dropping its elastic threshold from 80 to 70 after a counter named consecutive_veto reached 58.

Compiled by The EngineerSomething wrong?How this is made

Why it matters

cost Every gray-scale pass gives up 30 percent of the size on a signal the model rated 90, and that cost falls hardest on the trades the signal got right.
constraint Scaling substitutes for a veto only where exposure is continuous and loss is proportional to size; a gate in front of a schema migration or a physical interlock has no 0.7 setting.
capability The ruling carries its reasoning and its scale in the same record, so a review can ask why the discount was 0.7 and not 0.5.

Cutting size to 0.7 shrinks wins and losses in equal measure. Tightening the stop by 10 percent only shrinks the loss. Multiply the two and the worst case lands at 0.63 of what the full-size trade with the original stop would have lost, while 0.70 of the upside stays on the book ^[16]. A binary gate would have taken both to zero ^[3].

For 0.63 to be the real number, the tighter stop has to be hit no more often than the loose one. A stop 10 percent closer to entry is easier to reach, and the post itself describes crypto markets as "highly noisy and non-stationary" ^[10]. Anyone copying the 0.7 would need stop-out rates at both distances, measured on their own fills, before the 63 percent transfers.

The score does not pass through untouched. An advisor_score_delta of -3 takes the 90 to 87 ^[17]. If the entry threshold in the logs is the same 80, the penalty does not move the signal to the other side of it ^[18].

What makes the discount defensible here is what the gate was measuring. It fired on two stop-outs and one win in five, an outcome count over five trades ^[2]. The post's position is that such a run "might just be short-term volatility, liquidity hunts, or minor market regime shifts" and not a breakdown of core alpha ^[13].

One line of the published log records original=80 and floor=60 alongside the drop ^[9]. I would keep that floor. A threshold that decays every time a signal is blocked and has no lower bound reaches zero. Two controls in this system now loosen in the same direction: risk converts a veto into a smaller position ^[5], and scoring lowers the bar a signal has to clear before risk ever sees it ^[9]. The post presents one sanitized payload and two log lines, with no backtest, so the joint effect of the two is unmeasured ^[15]. The authors' own framing is that they are open-sourcing a thought process ^[14].

A risk module that records why it let a trade through leaves better notes than most deploy gates. The post's translation of the reason field reads, in part, "proceeding with 'position reduction + slight stop-loss tightening', not vetoing high-score system signal due to single trade results, nor is there a better SWITCH target" ^[8]. The alternatives array is empty ^[7].

What to watch

A backtest comparing the veto path and the gray-scale path over the same signal set would test the expected-value claim the post asserts.
Whether the floor of 60 holds under a longer veto streak, or whether the floor itself becomes adaptive.
Whether code accompanies the thought process the authors say they are open-sourcing, including the scaling function and the rule that increments consecutive_veto.

Claim ledger

Ranked by verification strength, evidence, and original report placement.

[1]
A dev.to post describes a quantitative crypto engine producing a 90/100 confidence score for a SHORT setup on NEARUSDT.
ReportedSupportedView cited source
[2]
The risk management module flagged that the sub-account had suffered two consecutive stop-losses in the same direction on NEAR over the past 48 hours, with a recent win rate of 20 percent (1 win in the last 5 trades).
ReportedSupportedView cited source
[3]
The post says that in a naive, rule-based trading system this condition immediately triggers a hard VETO and the trade is killed.
ReportedSupportedView cited source
[5]
Instead of a veto the engine output position_scale=0.7, reducing position size by 30 percent, and stop_tighten_pct=10, tightening the stop-loss by 10 percent.
ReportedSupportedView cited source
[6]
The published payload for the NEARUSDT setup records advisor_score_delta of -3, confidence of 0.65, and final_ruling of PROCEED.
ReportedSupportedView cited source
[7]
The payload's alternatives field is an empty array.
ReportedSupportedView cited source
[8]
The post's translation of the payload's reason field reads in part: "proceeding with 'position reduction + slight stop-loss tightening', not vetoing high-score system signal due to single trade results, nor is there a better SWITCH target".
ReportedSupportedView cited source
[9]
A system log line timestamped 2026-09-18 01:15:37,107 from scoring_engine reads: "F-229/F-230: Elastic threshold: 80 -> 70 (consecutive_veto=58, original=80, floor=60)".
ReportedSupportedView cited source
[10]
The post describes crypto markets as "highly noisy and non-stationary".
ReportedSupportedView cited source
[11]
The post characterises early-stage algorithmic risk management as a boolean logic gate (Signal > Threshold = Buy; Risk OK = Trade; Risk Bad = VETO) that provides "a false sense of security".
ReportedSupportedView cited source
[13]
The post says a string of losses "might just be short-term volatility, liquidity hunts, or minor market regime shifts" and not a complete breakdown of core alpha.
ReportedSupportedView cited source
[14]
The post says the authors are open-sourcing their thought process on transitioning from binary risk management to a dynamic gray-scale scaling approach.
ReportedSupportedView cited source
[15]
The post's published evidence is one sanitized decision payload plus two log lines; it contains no backtest and no measured expected-value comparison between the veto and scaling paths.
ReportedSupportedView cited source
[16]
A position scaled to 0.7 with a stop 10 percent tighter caps the loss at 63 percent of a full-size position stopped at the original distance, while retaining 70 percent of the upside.
DerivedSupported1 source— sign in to open
[17]
The advisor_score_delta of -3 applied to the 90 score leaves 87.
DerivedSupported1 source— sign in to open
[18]
A score of 87 exceeds both the original threshold of 80 and the reduced threshold of 70 recorded in the log line.
DerivedSupported1 source— sign in to open
[4]
The post argues a binary veto destroys long-term expected value because it fails to differentiate between a broken model and a temporarily adverse environment.
ReportedInsufficientView cited source

Sources

Publishers with included, body-backed reporting in this cluster.

dev.to
1 article · September 18, 2026
Gray-Scale Risk Control: Dynamic Scaling Instead of Hard Veto for High-Score Signals

Ask Clarity

Have a question this piece left open? The answer comes from its reporting and claim ledger — nothing else.

Try

Clarity analysis

How the coverage leans and what the record actually supports. Read the methodology

Perspective Coverage

How much of the coverage speaks to builders, operators and investors. It measures emphasis, not whether the story is right.

One sanitized decision record and a short block of log lines are all that is published, which is enough to copy the logic and not enough to know it pays. The engine takes a 90-score short on NEARUSDT, docks three points for two same-direction stops inside 48 hours, and lets it through at 0.7 size with the stop pulled in 10 percent. Three points leaves 87, still above the original 80 entry threshold, so the penalty never came close to blocking the trade. At those settings a loss costs about 63 percent of a full-size stop-out while 70 percent of the upside stays on, which is the trade-off the post is really describing. It answers a real design question with a worked example and no test of the alternative.

Reality

Evidence, adoption, hype gap, incentives and confidence, each read independently from the current record on the scale shown beneath it. How these are measured.

Evidence22

0100

Thin and self-reported

Everything checkable comes out of one system's own output: a single sanitized decision record, a handful of log lines, and a win rate the author reports on their own sub-account. The internal numbers are consistent, 90 less 3 is 87, and 0.7 size against a 10 percent tighter stop puts risk near 63 percent of a full position. Whether the scaled short made or lost money is not in the post, and no historical test compares scaling against the veto it replaced.

Adoption12

0100

One operator, its own logs

The logs are the only sign that anything is running, and they come from the same account that wrote the post. No other desk, exchange, library, or user is shown using this approach, and the post publishes a thought process rather than code anyone else could pick up.

Hype gap

Topics and entities

Follow any of these and your For You feed starts watching them — no settings page required.

Topics

Entities

build
Three manual interventions in a month, and every guard was working as designed
1 publisher
build
Six MariaDB versions, one real difference: the only reason to leave 10.6 is the July 2026 clock
1 publisher
build
A fixture-digest lockfile moves the merge gate out of CI and into commit identity
1 publisher
build
Seven of sixteen compacted agent sessions answered from the summary instead of re-reading the file
1 publisher

What happened

Why it matters

What to watch

Claim ledger

Sources

Let Clarity write the post for you.

Topics and entities

Topics

Entities

Related stories

Claim ledger

Let Clarity write the post for you.

Topics and entities

Topics

Entities

Perspective Coverage

Reality

Related stories

Sources