Finding Alpha

A field journal for the hard part of systematic trading.

Finding Alpha follows the live work of turning model signals into better allocation decisions: what to trade, which strategy to trust, how much conviction to size, and how to prove whether the change actually added value.

Signals are not enough Attribution beats storytelling Risk and sizing decide whether alpha survives
Researcher reviewing a trading signal
Podcast premise The pursuit matters because alpha is fragile.

The series is not a victory lap. It is a build log about finding where edge appears, where it disappears, and what needs to be measured before capital moves.

The central question

What does alpha mean here?

Alpha means improvement beyond a fair comparison. It is not just a positive return; it is evidence that a decision added value versus the baseline, benchmark, or risk profile it should be judged against.

In AlphaWeave Capital, alpha is decomposed by layer: strategy selection, asset selection, AI/PSM decisions, sizing, duration, and final portfolio events. The point of the podcast is to show which layer helped, which layer did not, and whether the report evidence supports the claim.

PSM means position size multiplier. When AI/PSM is discussed, the question is whether AI-assisted action, duration, or size-multiplier changes improved the executed portfolio path after risk and attribution are accounted for.

Benchmark context

Does finding alpha require market neutrality?

No. Market neutrality is one way to define and constrain alpha, but it is not required. The broader requirement is a fair comparison: the system has to show that a decision improved returns after accounting for the risk it took.

For a directional sleeve, the natural benchmark is often a market index, sector proxy, or beta-adjusted comparison. If the strategy made money only because the market went up, that is exposure. If it improved return beyond the relevant benchmark or improved risk-adjusted participation, that is closer to alpha.

For a market-neutral sleeve, alpha is cleaner to describe because the goal is to reduce broad market exposure and earn from relative value: long leg versus short leg, spread behavior, hedge quality, and net/gross exposure discipline. But market neutrality is a design choice, not a universal requirement.

Report-review format

The lead episodes review real run evidence.

Each alpha-case episode starts from a dated market window, runs the system, and then walks through the generated evidence: staged performance, detail rows, diagnosis, policy context, risk summaries, PSM ledger, and decision audit. The discussion is anchored to what the reports prove: strategy pickup, asset pickup, AI/PSM effect, total improvement, drawdown, and whether the allocation path beat the static buy-and-hold comparison.

Dated alpha case

Dated alpha case: report stack and execution path.

These dated cases use generated capital reports and event evidence from actual sampled runs. The point is to show the evidence trail, not to hide the result inside one headline number.

Alpha caseReport review

Actual Run Forensics: September 12-18, 2024.

The case starts with a static COIN/NVDA buy-and-hold book, then compares strategy review, asset switching, and AI/PSM overlay behavior. The strategy selector evaluated macd crossover as a shadow candidate, but the active strategy remained buy-and-hold; the meaningful changes came from asset selection and the executed overlay path.

  • Static baseline: buy-and-hold COIN/NVDA
  • Baseline strategy: buy-and-hold
  • Strategy candidate before asset switching: macd crossover, not activated
  • Asset switching: selected a broader asset book
  • AI/PSM overlay: changed the executed participation path
  • Report evidence: P and L dashboard, risk summaries, PSM ledger, and decision audit
Read case script View case manifest ReportAgent event rows PSM decision ledger
Alpha caseCapital reports

January 8-15, 2025 Capital Forensics Report.

This case is focused on capital reports only: staged P and L, position risk summary, decision risk summary, and the AI/PSM capital evidence. It starts with a static COIN/NVDA buy-and-hold book and shows how asset selection protected capital while the AI/PSM overlay remained conservative.

  • Starting capital: $1.0M
  • Static baseline: buy-and-hold COIN/NVDA
  • Strategy stage: no active strategy switch
  • Asset stage: main capital protection
  • AI/PSM overlay: reviewed the path with no executable resize
  • Report evidence: staged P and L, position risk, decision risk, and PSM ledger
Read case script View case manifest ReportAgent event rows PSM decision ledger

Presentation walkthrough

AlphaWeave Capital presentation walkthrough.

This video is a deck review, not a dated alpha case. It walks through the current presentation section by section and explains the measured evidence, compact AI/PSM path, policy governance, simulator feedback loop, and next-stage roadmap.

PresentationDeck walkthrough

AlphaWeave Capital Presentation Walkthrough.

This episode walks through the current presentation: competitive placement, single-bar decision flow, monthly evidence, policy governance, compact AI/PSM, TradeCrew, asset and strategy selection, simulator feedback, granularity, market extensions, and next-stage enhancements.

  • Presentation version: August 16, 2026
  • Compact AI/PSM is the preferred measured LLM path
  • Policy-only overlay and compact AI/PSM are separated
  • Performance claims are tied to sampled windows or labeled estimates
  • Report evidence and simulator replay anchor the improvement loop
Open presentation HTML Podcast HTML description Read script View walkthrough manifest

Background

Start here: the concepts behind the event studies.

These videos are the background layer for the alpha cases above: how to define alpha, choose the fair comparison, separate attribution layers, use AI/PSM, and trust the report evidence.

Episode 01Background

Why Finding Alpha Is Harder Than Generating Signals

Signals are only the beginning. This episode frames the real allocation problem: deciding whether a signal deserves capital, how it should be sized, and how to prove the decision added value.

  • Signal versus capital decision
  • Baseline, strategy, asset, and AI attribution
  • Why positive return alone is not proof of alpha
Read script
Episode 02Background

Alpha Needs a Benchmark, Not Always a Hedge

Market neutrality is useful, but not mandatory. This episode explains how directional strategies can still be judged honestly using a baseline, index, sector proxy, or beta-aware comparison.

  • Directional versus market-neutral evidence
  • Benchmarking instead of storytelling
  • Spread, hedge, net, gross, and beta diagnostics
Read script
Episode 03Background

The Four Layers of Allocation Alpha

This episode introduces the staged framework: static baseline, strategy selection, asset selection, and AI-assisted review with position size multiplier behavior.

  • What changed in the run?
  • Which layer actually helped?
  • How staged attribution prevents false credit
Read script
Episode 04Background

When AI Should Stay Out of the Trade

AI should not override clean strategy signals with weak confidence. This episode explains the do-no-harm rule: preserve exposure when the current model response is not strong enough to improve it.

  • Weak AI should not erase strong strategy evidence
  • PSM as the capital dial
  • Final portfolio events as the source of truth
Read script
Episode 05Background

The Honest Backtest Is a Flight Recorder

A performance number is not enough. This episode focuses on audit trails, cache validation, final portfolio events, and the evidence needed to trust what a run claims.

  • Decision records and run evidence
  • Cache hits, fallbacks, and model calls
  • Why attribution needs inspectable state transitions
Read script

Next step

Turn research progress into public proof.

Finding Alpha can become the narrative layer for AlphaWeave Capital: a steady record of experiments, failures, improvements, and measured attribution.

A Walkthrough