Mechanism Lab

Mechanism Lab

Two jobs: certify the validator with positive controls, and establish whether a market phenomenon exists before strategies are built around it. Only after a mechanism survives should TradeRep construct strategies on it. Execution is reboot Phase 3 — every entry below is a register entry, and a study with no committed evidence artifact is shown as such.

DashboardStrategiesResearch History

Positive controls
5
Controls with evidence
0
Validator certified
NO
Mechanism studies
8
Data-blocked studies
2
Strategies tested to date
59
The validator is not certified: no positive control has recorded evidence yet. Until it is, a negative campaign result means “this pipeline did not find an edge” rather than “no edge exists”, and that distinction is the whole reason this lane is first in the reboot’s priority order.

Positive controls

Validate the validator — without lowering any gate
Known strong edgeplannedcontrol-strong-edge
Injects
A large, unambiguous tradable signal in synthetic/controlled data.
Expected behaviour
The injected edge should survive the gauntlet appropriately.
If it fails
If the pipeline cannot recover a strong planted edge, its negative results are not trustworthy evidence of absence.

no evidence artifact

Known weak edgeplannedcontrol-weak-edge
Injects
A small edge whose magnitude is comparable to realistic execution costs.
Expected behaviour
Should weaken or die once costs and slippage are applied.
If it fails
Surviving unchanged would indicate costs are not being applied with real force.

no evidence artifact

Known no-edge nullplannedcontrol-null
Injects
Data with no exploitable structure.
Expected behaviour
Should fail statistically.
If it fails
A surviving null indicates the statistical gates manufacture edge from noise.

no evidence artifact

Known lookahead contaminationplannedcontrol-lookahead
Injects
A rule that is only profitable because it can see information it should not have.
Expected behaviour
Should be detected and rejected by the causality/leakage audit.
If it fails
Undetected contamination invalidates every positive result the pipeline has ever produced.

no evidence artifact

Known execution-fragile edgeplannedcontrol-execution-fragile
Injects
An edge that exists on paper but cannot survive fills, spread and slippage.
Expected behaviour
Should fail execution realism.
If it fails
Surviving would mean execution realism is decorative.

no evidence artifact

Mechanism studies

Does the phenomenon exist, causally, at a magnitude that survives costs?
PhenomenonStatusCausality questionData requiredEvidence
Aggressor imbalance -> future return
mech-aggressor-imbalance
blocked
Canonical dataset is 1-minute OHLC; aggressor side cannot be inferred from it (DATA_BLOCKED).
Is the imbalance measurable strictly before the decision bar?tick/trade data with aggressor side
CVD divergence -> future return
mech-cvd-divergence
blocked
Requires tick/trade source data not present in the canonical dataset.
Same-bar knowability of cumulative delta at the decision timestamp.tick/trade data
Distance from session value -> reversion probability
mech-distance-from-value
plannedSession value must be computed only from bars already closed.1-minute OHLC + volume (available)
Overnight inventory -> RTH direction / reversion
mech-overnight-inventory
plannedOvernight state is fully known at the RTH open.1-minute OHLC across the overnight session (available)
Expansion velocity / acceleration -> continuation vs fade
mech-expansion-velocity
plannedVelocity measured over closed bars only.1-minute OHLC (available)
Session location -> conditional expectancy
mech-session-location
plannedTime-of-day is known without lookahead by construction.1-minute OHLC (available)
Realized-volatility regime -> conditional behaviour
mech-rv-regime
plannedRegime label must use only trailing windows, never full-sample labelling.1-minute OHLC (available)
Trend efficiency / bar overlap -> continuation vs reversion
mech-trend-efficiency
plannedEfficiency ratio over closed bars only.1-minute OHLC (available)

What every study must answer

A study that cannot answer these is not evidence
  • · what phenomenon is being tested?
  • · is it causal?
  • · how much predictive information exists?
  • · on what horizon?
  • · in which regimes?
  • · is it stable chronologically?
  • · does it survive multiple-testing correction?
  • · is the magnitude large enough to survive realistic execution costs?

Only after a mechanism survives should TradeRep construct strategies around it. A PLANNED study is not evidence of anything, and a study in this file never authorises strategy construction by itself.

Read from research-output/registry/mechanism-studies.json · docs/TRADE_REP_MASTER_PLAN_REBOOT.md