Research note · Workbench operations

Research throughput and evidence at scale

A systematic research platform must support more than one promising
backtest. It must process multiple instruments and candidate
configurations while preserving provenance, separating raw results
from approved candidates and retaining evidence of failure as well
as success.

View the Workbench evidence

Follow one candidate through the pipeline

Depth and breadth are different engineering problems

The T0094/P95 case study demonstrates depth: one EUR/USD candidate
can be traced from discovery through validation, shadow observation
and controlled live micro-lot operation. The wider research estate
must also support breadth—processing many configurations, batches
and instruments through a common evidence workflow.

VERTICAL DEPTH

Following one candidate

A candidate retains its identity as it advances through discovery,
replay, out-of-sample review, promotion decisions, shadow operation
and live telemetry.

Read Beyond the Backtest

HORIZONTAL BREADTH

Managing a research estate

The same infrastructure receives results from multiple searches,
instruments and historical windows. Each batch must remain
identifiable without treating every apparently attractive result
as a viable strategy.

Observed multi-instrument research throughput

The Workbench view below records 478 raw Optuna runs in the selected
14-day window. The visible batches span EUR/USD, EUR/JPY and USD/CHF,
with individual batches containing as many as 134 candidate
configurations and tens of thousands of associated trade rows.

Lucitech Alpha Workbench showing 478 raw Optuna runs across EUR/USD, EUR/JPY and USD/CHF during a selected 14-day period, together with batch sizes and trade-row counts. — Figure 1 — Multi-instrument research intake over a selected
14-day window.
The Workbench records 478 raw Optuna runs across EUR/USD, EUR/JPY
and USD/CHF batches, together with trade-row counts, aggregate
outcomes and batch provenance. These are intake records, not
approved strategy candidates.

What this figure demonstrates

Breadth: several currency pairs and research
batches use the same intake workflow.
Throughput: 478 configurations and their
associated trade-level evidence were recorded in the selected
14-day view.
Traceability: batches retain their instrument,
simulation method, run count, trade-row count and completion
timestamps.
Evidence retention: positive and negative batch
outcomes remain visible at intake.
Separation of duties: raw optimisation output
remains distinct from registered and promoted candidates.

The displayed totals describe a selected historical research
window. They are not performance claims, capacity guarantees or
evidence that every raw configuration was independently viable.

Throughput is not evidence of viability

Producing more backtests increases the need for evidence discipline.
A large search can discover interesting configurations, but it also
creates more opportunities to select an attractive result by chance.

01 · INTAKE

Record the raw result

Results enter the Workbench with their batch, instrument,
simulation method and completion information. Intake preserves
what happened; it does not imply approval.

02 · SCREEN

Challenge the headline metric

Profit factor, average trade and net performance must be read
alongside trade density, temporal distribution and the conditions
under which the result was produced.

03 · GOVERN

Register selectively

Only selected configurations become governed candidates.
Registration creates a traceable research object for additional
validation; it is not an automatic deployment decision.

Scale increases the burden of proof.
The purpose of the Workbench is not to maximise the number of
persuasive backtests. It is to retain enough structured evidence to
determine which results deserve further investigation—and which
should stop.

Why high profit factor requires context

The screenshot deliberately retains the Workbench’s raw
Best batch PF field. This is an intake statistic, not a claim
about expected future performance.

Low trade density can exaggerate a metric

Profit factor can appear unusually high when it is calculated
from a small number of trades. A favourable short sequence may
dominate the result without demonstrating repeatability across
time or changing market conditions.

Filtering belongs in the evidence process

Subsequent screening can require minimum trade density and examine
stability across periods before a result is considered for
registration. The raw intake view remains useful because it
preserves the original result rather than rewriting the research
record.

Interpretation boundary:
No individual profit factor, net-pip total or average-trade figure
shown on this page should be interpreted as proof of a deployable
strategy. The evidence value lies in the governed handling of the
research population.

What has—and has not—been demonstrated

Supported by this evidence

Multi-instrument research intake
Batch-level provenance and timestamps
Hundreds of configurations in the selected window
Substantial associated trade-row populations
Retention of both favourable and adverse results
Separation between raw runs and registered candidates

Not established by this evidence

Unlimited or linear computational scalability
A specific maximum number of parallel workers
Guaranteed processing or database-ingestion rates
The viability of every configuration shown
Future profitability or execution performance
Equivalence with a large public cloud research platform

A formal capacity study would measure wall-clock processing time,
trials per hour, worker concurrency, resource consumption and database
behaviour as workloads increase. The present evidence supports the
narrower claim that Lucitech has demonstrated repeatable,
multi-instrument research throughput inside a common evidence system.

How this fits the Lucitech proposition

The commercial value is not simply the ability to run an optimisation
sweep. It is the integration of broker-aligned historical testing,
structured intake, candidate registration, validation evidence,
controlled promotion and operating telemetry.

01
Discover
Run bounded searches across instruments and historical windows.

02
Record
Preserve batch provenance, outcomes and trade-level evidence.

03
Challenge
Test density, stability, robustness and out-of-sample behaviour.

04
Promote
Advance selected candidates through governed evidence gates.

05
Observe
Compare expectations with shadow and controlled live telemetry.

The research estate is part of the product.
Lucitech is being designed to make systematic research reproducible,
inspectable and operationally governable—not merely to generate
attractive retrospective results.

Continue through the evidence pipeline

Read the developing T0094/P95 case study to see how one candidate
moved from research output into validation, demo-shadow observation
and controlled live micro-lot operation.