Version 3.1 · Reviewed 9 August 2026 · Current operating model
01 / Abstract
The Qadam Experiment
Large quantitative funds traditionally combine specialised researchers, engineers, data infrastructure, traders, risk controls and substantial compute. Qadam asks whether one operator can now assemble a smaller, inspectable version of that capability from tools available to the public.
Can a small artificial hedge fund team, built from consumer-accessible AI, software, data and quantum tools, discover a genuine, explainable trading edge?
Consumer AI execution, quantum pattern recognition and Akber's investment filter.
Different technologies perform the roles their capabilities and limitations suit.
Every idea moves through evidence gates rather than a forced trade conveyor belt.
Untouched data, costs, forward observation, paper outcomes and attribution determine belief.
Qadam's real product is not a stream of trade ideas. It is a disciplined operating system that can reject an idea, continue observing it, or run a small governed paper experiment. A paper experiment gathers evidence; it does not retroactively prove the idea was an edge.
02 / Origin
Why Qadam Exists
An institutional-scale challenge became buildable
Consumer AI coding tools made it possible for one product builder to attempt software that previously needed a sizeable engineering team: scheduled ingestion, model orchestration, historical replay, quantitative research, governance, broker-paper integration and a public explanation layer.
Disconnected APIs might contain connected information
Product work across geospatial, climate, automotive, financial and other API ecosystems had revealed how many useful facts sit in separate systems. Qadam tests whether comparing them point in time exposes relationships the market has not fully absorbed.
Trader judgment could become auditable
Friends who had succeeded in finance contributed practical ways to distinguish an interesting story from a tradeable setup. Their questions about context, catalyst, confirmation, risk, execution and learning became Akber's 6-Stage Filter.
Quantum curiosity needed a real test
Qadam provides a measurable use case for quantum computing: find nonlinear structure that a fairly matched classical method misses, then prove whether the difference survives unseen evidence and realistic costs.
The challenge is to compare previously disconnected evidence, map repeatable relationships to disciplined strategies, and let the system improve only when evidence earns that right.
03 / Research agenda
Three Foundational Hypotheses
These are falsifiable research questions, not proven claims. Each one uses the same scientific structure so engineering progress cannot be confused with market proof.
Consumer AI execution
Can advances in consumer-accessible AI execution technology enable the discovery of a genuine trading edge?
Models, open APIs, affordable compute and broker-paper environments let one operator build and inspect a sophisticated research operation.
An unattended Python control plane schedules ingestion, research, decisions, lifecycle, learning and visibility while bounded models work through one provenance-linked lifecycle.
Reliable unattended operation, reproducible research, safe rejection and an edge that survives unseen data, costs, forward observation and paper testing.
The system works technically but produces no durable advantage, cannot reproduce its research, leaks future information or costs more than it contributes.
Quantum pattern recognition
Can quantum computers recognise a genuine trading pattern and form a successful trading strategy from backtested data where a matched classical approach cannot?
Markets can contain interactions, regimes and path dependence that simple linear rules represent poorly. Quantum methods may encode some structure differently.
Classical and quantum lanes receive the same frozen evidence, labels, horizons, train-test boundaries and economic evaluation.
Out-of-sample improvement over the strongest fair classical baseline, reproducible on hardware and economically useful after spread, slippage, delay and turnover.
The advantage disappears on untouched evidence, cannot be reproduced, fails to beat classical methods or is too small to survive costs.
Akber's investment filter
Can hedge fund trader Akber's evaluation expertise - captured in his six-stage investment filter - be operationalised successfully through automated systems?
Experienced traders reject ideas for current-market reasons a historical relationship alone cannot capture: stale catalysts, weak confirmation, poor asymmetry or unsuitable execution.
Every evidence-classified setup receives an auditable pass, hold or veto across Context, Catalyst, Confirmation, Risk, Execution suitability and Postmortem learning.
Frozen strategies with and without the filter must show whether passes improve expectancy, vetoes avoid losses and holds improve timing after costs. Discovery outcomes remain separately labelled.
The filter only suppresses activity, relies on hindsight, cannot be applied consistently or fails to improve risk-adjusted outcomes.
04 / Operating system
The Artificial Hedge Fund Team
The important story is not that Qadam uses several technologies. Their different capabilities and limitations determine their jobs.
Runs the unattended 18-service loop, preserves the temporal evidence graph, enforces gates, reconciles state and controls the only guarded paper route.
Boundary: Cannot invent missing evidence or bypass a failed gate.
Processes high-volume information locally and turns observations into structured research questions.
Boundary: Proposes research; cannot establish proof, approve risk or reach the broker.
Challenges narratives, considers alternatives and tests whether a pattern has a plausible economic mechanism.
Boundary: Persuasive reasoning is not statistical validation or trade authority.
Tests linear, nonlinear, regime-dependent and quantum-assisted relationships against fair baselines.
Boundary: Compute cannot manufacture information or declare an edge without untouched evidence.
Defines constitutional boundaries, reviews major changes and judges whether the experiment has earned greater trust.
Boundary: Cannot retroactively alter evidence or turn a dashboard action into an order.
Python supplies continuity and control; the local model supplies scalable attention; the frontier model supplies broad reasoning and challenge; the quant layer supplies measurement; and the Fund Manager supplies purpose and accountability. Models never receive broker credentials, and the public dashboard remains read-only.
05 / Research terrain
The Evidence Universe
Qadam's frozen baseline spans 41 registered sources and 19 watched instruments. These are coverage boundaries, not a claim that every source is fresh, equally trusted or historically complete.
Six evidence families
- Geopolitical and security events
- Physical-world and geospatial signals
- Macroeconomic and trade data
- Markets and technical evidence
- Corporate, political and regulatory disclosures
- Narrative and probability signals
Markets and paper expressions
Five configured macro strategy families operate inside the watched universe. New relationships can form separately governed emerging strategies, including the current power scarcity and congestion research sleeve. Qadam separates an economic market from the liquid paper expression and records proxy basis risk.
Open live Trading UniverseA rise in verified conflict activity may matter more when shipping disruption also increases, prediction markets remain complacent and oil prices have not yet repriced. The possible edge may live between systems, not inside one feed.
Qadam stores provider observations, entities, instruments, patterns, experiments, strategy versions, decisions and outcomes in an append-only temporal evidence graph. This lets a new cycle retrieve prior tests and rejections, detect duplicate hypotheses and trace a result back to the evidence available at the time. Graph links remain research leads: they cannot satisfy source quorum, validate an edge, approve a trade or reach the broker.
The live Data Sources page owns the exact provider inventory, connection state, freshness, licence posture and historical availability. The live Trading Universe owns the current markets and instruments. Only the provider-backed, timely subset allowed by a strategy-specific profile can affect a current setup. Every comparison preserves when information became available; missing history is labelled, not fabricated.
06 / Experimental method
The 10-Stage Experimental Method
The lifecycle is a sequence of evidence gates, not a conveyor belt that must produce a trade.
Understand the world
Observe and qualify evidence.
Observe the World
In: Source and market changes.
Work: Capture provenance and availability time.
Out: Observation record.
Stop: Stale, unsafe or unsupported input.
Lead: Research Analyst and COO.
Tests: Consumer AI execution.
Qualify the Evidence
In: Observation records.
Work: Check trust, relevance and point-in-time safety.
Out: Qualified or rejected evidence.
Stop: Leakage, duplication or weak provenance.
Lead: COO.
Tests: Consumer AI execution.
Search for an edge
Discover patterns, form strategies and validate them.
Discover Patterns
In: Qualified source-price evidence and prior experiment memory.
Work: Search the graph and test linear, analogue, state, nonlinear and quantum relationships.
Out: Ranked finding, preregistered question or rejection.
Stop: Duplication, instability, false discovery or no mechanism.
Lead: Head of Quant.
Tests: AI and quantum hypotheses.
Form Strategy Hypotheses
In: Supported pattern, mechanism and prior-attempt record.
Work: Define a reversible version with expression, horizon, invalidation, costs and risk concept.
Out: Versioned strategy hypothesis.
Stop: Weak causality, duplicate failure or unavailable instrument.
Lead: Strategy Lead.
Tests: Consumer AI execution.
Validate the Edge
In: Frozen strategy hypothesis.
Work: Backtest, walk forward, hold out, stress and shadow.
Out: Validated edge, bounded discovery eligibility, more research or rejection.
Stop: Leakage, overfit, costs, failed holdout or incomplete current evidence.
Lead: Head of Quant and Strategy Lead.
Tests: All three hypotheses.
Decide whether it is tradeable
Apply practical judgment and portfolio governance.
Akber's 6-Stage Filter
In: Evidence-classified setup and current evidence.
Work: Audit Context, Catalyst, Confirmation, Risk, Execution suitability and Postmortem learning under the declared profile.
Out: Pass, hold or veto.
Stop: Missing or adverse evidence.
Lead: Akber's filter.
Tests: Akber's investment filter.
Govern the Decision
In: Filter result and portfolio state.
Work: Reconcile risk, concentration, duplication, freshness and route safety.
Out: One governed decision.
Stop: Risk breach, stale evidence or duplicate intent.
Lead: COO and Fund Manager governance.
Tests: AI and Akber hypotheses.
Test it in reality
Use the guarded paper route and reconcile every state.
Execute and Monitor
In: Clean paper-review decision with a risk tier.
Work: During the real market session, refresh Akber, shadow, risk and Router in one generation, then submit only through PaperOps to Alpaca Paper.
Out: Attributable paper lifecycle or safe no-order result.
Stop: Broker mismatch, stale route, liquidity, duplication or idempotency.
Lead: COO and PaperOps.
Tests: AI and Akber hypotheses.
Compound knowledge
Learn cautiously, test changes and return only approved versions.
Learn From the Outcome
In: Matured research, shadow, hold, veto, paper or system outcome.
Work: Compare expectation with reality, attribute contribution and write the result to persistent experiment memory.
Out: Supported, rejected or unmeasurable lesson.
Stop: Missing lineage or immature horizon.
Lead: Research Analyst and Strategy Lead.
Tests: All three hypotheses.
Improve and Re-enter
In: Supported lesson and measurable proposal.
Work: Freeze a challenger, test, observe, review, version, monitor and make it reversible.
Out: Approved version, rejection or more testing.
Stop: No improvement, unsafe change or no rollback.
Lead: COO and Fund Manager.
Tests: AI and Akber hypotheses.
The validated-strategy lane keeps the full edge standard. The discovery lane may run a complete but under-evidenced setup as a small, explicitly labelled paper experiment. The frozen ladder is US$500 for a first discovery experiment, up to US$2,000 after independent repeat confirmation, and an absolute US$5,000 ceiling only for a validated paper setup. One score or one winning trade cannot advance a tier.
07 / Evidence standard
How Qadam Establishes Proof
Proof is a ladder. A claim cannot skip a rung because an earlier result looks impressive.
- Point-in-time datasetReconstruct only what was knowable, including publication delays, revisions and market calendars.
- Predefined hypothesisFreeze the mechanism, market, horizon, expected effect, costs, benchmark and failure condition.
- Historical backtestMeasure recurrence across independent events and regimes after realistic friction and proxy basis risk.
- Walk-forward and untouched holdoutDesign on earlier data; judge on later information that did not influence the rules.
- Multiple-testing controlAccount for how many relationships, instruments, horizons and model variants were searched.
- Matched specialist comparisonGive nonlinear and quantum methods the same frozen evidence as the strongest reasonable classical baseline.
- Forward observationWatch new information arrive over real market time without changing the strategy.
- Guarded paper tradingTest timing, slippage, lifecycle and portfolio behaviour through Alpaca Paper without live capital. A discovery trade gathers evidence; it does not create proof.
- Attribution and postmortemLink outcomes to exact evidence, strategy version, filter decision, governance and execution.
- Controlled improvementApply only an approved version after separate testing, monitoring and rollback design.
A repeatable source-backed relationship with a defensible mechanism that remains useful on untouched evidence, after realistic costs and against a relevant benchmark. It needs enough independent observations, tolerable risk, a liquid paper expression and complete lineage.
A research score is not a probability of profit. A backtest is not forward proof. A successful quantum job is not quantum advantage. One winning paper trade is not a strategy. The paper proof ledger preserves those distinctions.
08 / Current evidence
Current Findings
This section records durable conclusions from this edition. Mutable counts, freshness and readiness belong on the live dashboard.
The experiment can be operated
- One operator can run the multi-role research architecture.
- An unattended 18-service control plane, ten-stage lifecycle and guarded Alpaca Paper route are implemented.
- A rebuildable temporal evidence graph preserves research lineage, duplicate attempts and negative results.
- A bounded discovery lane can gather small paper outcomes without mislabelling them as validated edges.
- Provider-backed history is retained where permitted; unavailable history is labelled.
- IBM Quantum hardware research has run through an auditable hybrid path.
Market advantage has to be earned
- No static document may claim a durable edge; the current registry belongs on the dashboard.
- Quantum processing has not established useful out-of-sample advantage over fair classical baselines.
- Akber's filter needs more independent forward and paper outcomes to prove incremental value.
- Open-market conversion is structurally ready but still needs its real market-day canary and multi-session soak.
- The clean US$100,000 Alpaca Paper baseline is an experiment, not proof of performance.
- The graph-assisted conversion trial requires five real market days and cannot be backfilled; its completion would not by itself prove an edge or profit.
Read mutable truth from its operating surface
The dashboard separates what is designed, what is implemented, and what current operating evidence permits now. Archived testing returns are not current performance or proof.
09 / Constitutional boundary
Governance And Limits
Qadam is deliberately asymmetric: research can be imaginative, but claims and actions must be conservative.
The governed route targets Alpaca Paper. Live capital is outside the current authority contract.
The frozen ladder starts at US$500, requires independent evidence before US$2,000, and reserves the US$5,000 ceiling for validated paper setups.
Models, Telegram and the public dashboard cannot access broker credentials or call the broker directly.
The dashboard explains state but cannot create research, approve risk, submit orders, change code or award proof.
Cadence and trial windows cannot override evidence, risk, liquidity, duplication, freshness or safety.
Forward evidence and the 30-day paper growth trial use real elapsed calendar time and are never backfilled.
Learning produces proposals. Changes require tests, review, an approved version, monitoring and rollback.
A provider, model, quantum run, Akber pass, Router state, PaperOps handoff or profitable trade does not inherit proof.
Rejections, no-trades, failed experiments and classical-preferred results remain part of the record.
Three internal terms, defined once +
These names describe controls. They do not create authority.
10 / Success criteria
What Would Make Qadam Successful?
Consumer AI execution succeeds if
One operator can maintain a reliable, auditable system that discovers at least one repeatable edge, rejects false discoveries, survives real operations and produces attributable outcomes. It fails as an investment hypothesis if it remains an elaborate research factory that cannot outperform simple baselines after costs.
Quantum pattern recognition succeeds if
A reproducible IBM Quantum or quantum-assisted method improves untouched prediction beyond the strongest fairly matched classical method and survives costs in a useful strategy. Hardware access or circuit execution alone is not success.
Akber's investment filter succeeds if
Its passes, holds and vetoes improve net expectancy, timing, drawdown or avoided losses relative to the same frozen strategies without the filter. It fails if it only suppresses activity, relies on hindsight or cannot be applied consistently.
The experiment may prove all three hypotheses, some of them, or none. That uncertainty is not a weakness to conceal; it is the reason Qadam exists. Its most important discipline is to remain capable of saying that the available evidence contains no durable edge.
No edge, no trade. No proof, no claim.