Benchmark / Model training / Research roadmap

BiasForge Beta / measured in public

Track the edge, not just the idea.

The benchmark stack is live in beta. Preset models, webhook learning, replay imports, and style tracking are running now, while verified public result boards remain intentionally TBA until the data is clean enough to publish.

View TradingView Tools See News Roadmap

Benchmark home

Rank what actually works, not one blurred result.

Public build

Tracked models

Preset buckets reporting now

Trained models

Models with learned history

Stored trades

003

Live alerts + imports

Current mode

Beta

Measured in public

Spotlight board

TBA

Execution Map Bundle

Balanced / Daily to 1H / TVC:GOLD

Public build

Net

--.--

Trades

003

Win rate

--%

Model board preview

placeholders

Sweep + Structure Bundle

Active / Sweep board

Collecting

Net

--.--

Trades

---

Win rate

--%

Execution Map Bundle

Balanced / Daily -> 1H

First board

Net

--.--

Trades

---

Win rate

--%

Core Workflow Bundle

Balanced / Bias board

Collecting

Net

--.--

Trades

---

Win rate

--%

Full BiasForge Suite

Balanced / Preset compare

Queue

Net

--.--

Trades

---

Win rate

--%

XAU starter workflow

Start the first model with one clean XAU lane.

This is the first practical benchmark setup: one preset, one market, one repeatable workflow. Keep it research-only and use closed-trade labels to teach the model what actually holds up.

Regime

Keep the higher-timeframe bias and range context stable before you score anything lower.

Structure

Map session direction, swing structure, and where the market is leaning next.

15m

Setup

Use this to confirm the actual setup type and whether the idea is worth logging.

Trigger

Make this the first live training timeframe so closed trades are labeled consistently.

Refine

Use this only for optional entry refinement at first, not as a separate learner.

Training rules

Start with one preset only: XAU Core Replay.

Keep 4H and 1H as context lanes, then log the actual setup from 15m and 5m.

Train from closed trades, not open ideas, so the labels stay clean.

Use X as a fast context lane for macro and gold headlines, not the final authority for actual prints.

Stay research-only for now. BiasForge benchmarking tracks setups and outcomes, not broker execution.

Only the closed-trade payload trains the model. Context payloads are for storing the higher-timeframe map so you can review the setup later without contaminating the learner.

Context payload

Use this from 4H and 1H to log the regime and structure without training the model yet.

no label

{
  "preset": "XAU Core Replay",
  "event": "context",
  "ticker": "OANDA:XAUUSD",
  "timeframe": "240",
  "side": "long",
  "long_score": 13,
  "short_score": 5,
  "threshold": 10,
  "trend_strength": 7,
  "entry_stretch": 2,
  "cooldown_active": 0,
  "session_allowed": 1
}

Closed-trade payload

Send this JSON after the 5m trade closes. This is the first payload that should teach the learner.

research only

{
  "preset": "XAU Core Replay",
  "event": "closed",
  "ticker": "TVC:GOLD",
  "timeframe": "5",
  "entry_mode": "sweep-reclaim",
  "pair_mode": "solo",
  "side": "long",
  "label": "win",
  "pnl": 2.4,
  "long_score": 15,
  "short_score": 4,
  "threshold": 10,
  "trend_strength": 8,
  "entry_stretch": 3,
  "cooldown_active": 0,
  "session_allowed": 1
}

Webhook route:
/benchmark-api/v1/webhooks/tradingview/benchmark?token=YOUR_TOKEN

X starter lane

Keep X as a fast context layer for gold, dollar, and rates flow. Use official handles plus one fast headline account.

Tracked accounts

federalreserveecbbankofenglandustreasurybls_govfinancialjuice

Tracked keywords

xauusdgolddxyusdjpyfomccpinfppowellbojtreasuryyields

Current stage

Public build

Preset models, webhook learning, replay import, and style tracking are already live, but public benchmark boards are still being verified.

Goal

Rank what actually works

The target is to compare presets, sides, and styles by market and timeframe instead of judging the whole stack as one vague result.

Where we are now

Collecting and cleaning

Right now the benchmark layer is about gathering cleaner trade history, keeping models separate, and only publishing boards once the data is solid.

Current focus

Preset separation

Keep preset models separate instead of blending workflows together.

Better samples

Collect more closed trades with both wins and losses.

Board verification

Clean imported history before turning it into public benchmark boards.

Private research bridge

Connect the benchmark layer to the future private access and dashboard flow.

Public Results

Benchmark boards are getting a real home.

These cards stay placeholder by design. Public result boards only go live after the runs are verified, grouped cleanly, and easy to compare.

TBA

Execution Map Bundle

Balanced / TVC:GOLD / Daily to 1H

Reserved for the first verified public board after shared-range comparison is locked.

TBA

Net

--.--

Trades

---

Win rate

--%

TBA

Net

--.--

Trades

---

Win rate

--%

TBA

Net

--.--

Trades

---

Win rate

--%

TBA

Full BiasForge Suite

Balanced / Multi-market / Preset comparison

Reserved for preset, side, and style comparisons once the benchmark samples are broad enough.

Preset

TBA

Net

--.--

Trades

---

Win rate

--%

Side

TBA

Net

--.--

Trades

---

Win rate

--%

Style

TBA

Net

--.--

Trades

---

Win rate

--%

Live Now

The first benchmark loop is already running.

The current focus is straightforward: collect benchmark trades, keep training by preset, and make style performance easier to inspect.

Preset models

Sweep + Structure, Execution Map, Core Workflow, and Full BiasForge Suite can train as separate styles.

Automatic learning

TradingView benchmark alerts can keep feeding the model while charts and alerts run overnight.

Replay and import

Stored events can be retrained, and historical benchmark trades can be imported to build a stronger base faster.

Style tracking

The goal is simple: learn which preset, entry mode, pair mode, and side keep performing best.

TBA

The next benchmark layer.

These are the next public pieces: more data coverage, company pages, and a private interface that turns raw benchmark output into usable research.

Broker and venue coverage

TBA: collect benchmark context from more brokers, venues, and feeds instead of one chart at a time.

Company intelligence

TBA: pages for brokers, prop firms, and companies with status, key updates, and linked news flow.

Private dashboard

TBA: a login-protected dashboard for styles, benchmark stats, imported trade history, and research views.

Get Access