Pegasus - Glavior's First Production Decision Engine
The full technical dossier on Pegasus - model code PEG, generation v1.0. 65 factors, 4 protocols, 3 strategies, four decision layers, the anti-chase safety contract, and a public append-only ledger.
Pegasus - Glavior's First Production Decision Engine
Pegasus is the first production-grade decision engine to graduate from Glavior's research lane. It turns raw market structure into entry, target, and stop - and commits every output to an immutable public ledger the moment it fires.
This is the full technical dossier: architecture, decision layers, risk constraints, resolution protocol, and supported markets.
Quick Specification
- **Model Code:** PEG
- **Generation:** v1.0 - first production release
- **Status:** Live · Production
- **Asset Class:** Crypto Spot + Perpetuals
- **Decision Cadence:** On-event (event-driven, never time-boxed)
- **Outputs:** Deterministic · Append-only ledger · Per-model attribution
Mission - Decisions Over Dashboards
Pegasus is not a charting toy and not a feed of opinions. It is a closed-loop decision engine: every analysis it emits carries a directional thesis, a planned entry, a take-profit, a stop-loss, and a model attribution. There is no neutral "FYI" output - an analysis either commits to a trade or it stays silent.
That commitment is what makes the ledger meaningful. Without skin-in-the-game outputs, a track record is just curated screenshots.
The Engine - 65 Factors, 4 Protocols, 3 Strategies
Pegasus evaluates **65 distinct factors** per analysis - 54 base structural / momentum / microstructure factors plus 11 Smart-Money (SMC) factors. These factors are routed through **4 protocols** (Market Context, Structural Confirmation, Microstructure Validation, Execution Safety) and resolved into **3 strategy variants** (Trend-Continuation, Reversal-at-Liquidity, Range-Mean-Reversion) selected automatically by the live market regime.
A trade only exists if all four protocols agree and a strategy variant validates the setup.
Four Decision Layers - Evaluated In Strict Order
Each layer is a hard gate. If a layer rejects, the lower layers never run. That order is what prevents Pegasus from rationalising a bad trade with surface-level momentum signals.
L0 - Market Context (the King override)
BTC correlation, BTC dominance regime, and macro tape direction. A long signal on an alt while BTC is breaking down is blocked here - before the asset is even scanned.
- A short on an alt while BTC is pumping > +0.5% in 1h is a hard block.
- A long on an alt while BTC is dumping is a hard block.
- Rising BTC.D reduces position confidence on altcoin longs; falling BTC.D adds caution to altcoin shorts.
L1 - Structure
Higher-timeframe bias, break-of-structure, key levels, and VWAP relationship. The trade has to make sense on the chart at the macro scale before microstructure gets a vote.
- HH/HL (bullish) or LH/LL (bearish). Mixed structure is not a trade.
- Break of Structure (BOS) must confirm the intended direction.
- Higher-timeframe bias (weekly/daily) must align or be neutral.
- Position relative to VWAP and key institutional levels.
L2 - Microstructure
CVD, order flow imbalance, liquidity sweeps, relative volume, open interest, and funding. This layer answers a single question: is the move being paid for in real flow, or is it a wick?
- Cumulative Volume Delta must confirm the move.
- Relative Volume (RVOL) > 1.0.
- Open interest must be rising into the move.
- Funding rate must not be punishing the direction.
- Liquidity sweeps and reclaims are preferred entry conditions.
L3 - Momentum & Risk
EMA stack, RSI context, MACD, ADX, ATR, and regime classification. Used to size confidence and to fail trades that look right on structure but are arriving exhausted.
- EMA200 trend filter, EMA stack state.
- RSI context with divergence detection.
- MACD momentum check, ADX > 20 for trend trades.
- ATR-normalized volatility for sizing decisions.
- Regime classification (trending vs ranging) selects the strategy variant.
The Anti-Chase Contract - Execution Safety
Most retail systems die the same way - they chase. Pegasus has three hard refusals built into the execution layer itself, not as soft warnings.
Minimum 2.0 R:R
Any setup that does not offer at least 2R after spread is dropped, no matter how clean it looks. A model that takes 1.2 R:R trades to "stay active" is a model that bleeds out slowly.
Extension Block
If price is more than ~3% extended from VWAP in the trade direction, entry is invalidated - wait for pullback. The system's literal status output in that case is: *"Entry Invalid - Extended. Wait for pullback."*
Quality Threshold (TQS < 75 = no trade)
Every signal computes a composite Trade Quality Score across all 65 factors. Below 75, Pegasus stays flat. This is why the live feed will sometimes be quiet - silence is a feature, not a bug.
Resolution - Only Two Ways An Analysis Can Close
Pegasus follows the platform-wide no-expiry rule. An analysis is born with a target and a stop - those are the only valid terminal states. There are no silent timeouts, no quiet deletions, no "marked as invalid".
1. **Take-Profit Hit** - Price physically touches the published target. The trade is recorded as a win with its exact close timestamp and realised return. 2. **Stop-Loss Hit** - Price physically touches the published stop. The trade is recorded as a loss. It is never hidden, edited, or re-attributed to a different model.
If conditions change before either level is hit, the system may rebase the entry to the new market, but the signal does not vanish.
The Public Ledger Guarantee - What You Get
Every Pegasus signal is written to a public, append-only ledger the moment it fires:
- Entry price, stop, target, and confidence at the moment of firing.
- Final outcome (TP / SL) and realised R-multiple.
- Regime and BTC context snapshot at trigger time.
- Permanent **PEG** attribution - future models start at zero and cannot inherit Pegasus's wins or losses.
The Asymmetry
- **Deterministic outputs.** Same inputs produce the same decision. Pegasus does not "feel" a trade - every emission is reproducible from the recorded market state.
- **Immutable record.** The moment a Pegasus analysis fires, it lands in a database with edit-protection at the trigger level. We cannot retouch its history if we wanted to.
- **Per-model attribution.** Every output is forever tagged PEG. When future models ship, their record starts at zero.
What Pegasus Is Not
To set expectations honestly:
- Pegasus is not a high-frequency scalper. It will skip days.
- Pegasus is not a "100% win rate" product. The math of a 2.0 R:R minimum makes a 40–55% win rate already extremely profitable.
- Pegasus is not a black box that adapts opaquely. The 65 factors, 4 protocols, 3 strategies, four decision layers, and the safety contract are the entire surface.
The Ledger Is The Proof
Open the live ledger to inspect entry, target, stop, and outcome for every analysis Pegasus has ever emitted. The numbers on this page are computed from that ledger - not the other way around.
If a competitor wants to claim a better track record, they have to ship the same ledger. Most won't, because they can't.
[INSIGHT] Open the live Pegasus ledger and equity curve on the public Track Record page. The numbers are not the point - the auditability is.