Quant Alpha Validation Framework

Research only. No live trading. No execution. No orders. No account access. Not financial advice. The original alpha failed out of sample.

Most quantitative trading hypotheses fail. The real research skill is not finding profitable backtests—it is designing a process that can invalidate weak hypotheses before they reach production.

This case study shows how a crypto short-alpha hypothesis was proposed, frozen, audited, rejected out of sample, and archived responsibly.

Table of Contents

• Project Summary
• Research-Only Disclaimer
• Research Question
• Hypothesis
• Validation Design
• Discovery Findings
• Frozen OOS Protocol
• Pipeline Integrity Audit
• OOS Results
• Failure Attribution
• Why This Was Not Deployed
• Lessons Learned
• Future Research Directions
• Reproducibility Scope
• Repository Structure
• Limitations
• License And Use

1. Project Summary

This package is a public-safe case study extracted from a larger FDV market-structure research workspace. It documents how a crypto short-alpha hypothesis was formed, instrumented, frozen before forward testing, audited for pipeline defects, rejected out of sample, and archived with explicit failure attribution.

This is not the original FDV repository. Large datasets, runtime state, operational logs, credentials, and trading-capable components are intentionally excluded.

The active public-package folder is quant-alpha-validation-framework/ inside the FDV workspace.

2. Research-Only Disclaimer

This repository is an educational research artifact. It does not connect to exchange accounts, submit orders, manage positions, use leverage, or provide financial advice. The included scores, thresholds, and zones are historical research definitions, not recommendations or decision rules.

3. Research Question

Do repeated liquidity fragility and funding-persistence states in altcoin perpetual markets predict future negative BTC-adjusted returns?

4. Hypothesis

The discovery hypothesis proposed that persistent funding imbalance combined with liquidity thinning could identify structurally fragile altcoins. The strongest discovery diagnostics appeared in high-persistence regions, especially the strict combined Zone C.

5. Validation Design

The research pipeline separated:

1. discovery-period threshold research,
2. a frozen cutoff and immutable zone specification,
3. post-cutoff event production,
4. future-return maturity at 1D/3D/5D/7D,
5. accepted OOS samples,
6. published summaries,
7. rule-based failure attribution.

The implementation uses standard-library Python modules under src/sim_execution/research/. Focused tests use synthetic data and do not require the excluded evidence corpus.

6. Discovery Findings

Discovery diagnostics suggested:

• liquidity fragility and funding positioning imbalance were plausible mechanisms,
• persistence mattered more than isolated events,
• Funding Persistence 51-100 was the strongest diagnostic bucket,
• the effect looked cross-sectional rather than single-asset,
• negative drift appeared to accumulate over several days.

These were discovery-sample findings only.

7. Frozen OOS Protocol

The discovery sample ended at 2026-05-30T14:45:59Z. The frozen research zones were:

• Zone A: Funding Persistence >= 11
• Zone B: Overlap Persistence >= 11
• Zone C: Funding Persistence >= 101 and Overlap Persistence 51-100
• Zone D: Failed Recovery = 1 or 2, watch condition only

The frozen definitions are preserved in data/outputs/out_of_sample_fragility_threshold_spec.json. They were not retuned after forward results arrived.

8. Pipeline Integrity Audit

An apparent contradiction occurred: event-production audits showed thousands of post-cutoff zone hits while the OOS summary remained at zero. The project traced the full funnel:

Frozen Hits -> Candidates -> Mature Samples -> Accepted Samples -> Published Summary

The audit identified stale summary publishing, repaired the refresh path, and confirmed cutoff, threshold, timestamp, and symbol alignment before interpreting OOS results. See:

• reports/fragility_event_production_audit_report.md
• reports/forward_validator_alignment_report.md

9. OOS Results

The original frozen short-fragility result did not survive. The archived post-mortem recorded failure across all four zones. Later reporting continued to show Zone C as failed and the broad threshold system as weakened or failed rather than confirmed.

The exact machine-readable snapshot included here is historical and timestamped. It must not be interpreted as a live signal.

10. Failure Attribution

The rule-based attribution engine classified the failure as:

BROAD_FACTOR_BREAKDOWN with HIGH confidence.

Supporting evidence:

• failure occurred across zones,
• Zone C experienced a sign flip,
• failure contribution was broad across symbols rather than concentrated in a few names,
• no material regime shift was sufficient to explain the result,
• failure days clustered.

11. Why This Was Not Deployed

The alpha failed its forward test. The thresholds were frozen before forward evaluation, and retuning them after failure would invalidate the OOS protocol. Deploying an invalidated strategy would violate the original research protocol. The correct outcome was to archive the thesis as OOS_FAILED_RESEARCH_ARTIFACT, not to convert it into a trading system.

No deployment capability is included in this package.

12. Lessons Learned

• Discovery coherence is not OOS confirmation.
• A zero result must be separated from a stale publishing path.
• Research infrastructure reliability is part of research validity.
• Broad failure cannot be repaired by highlighting a few favorable symbols.
• Frozen thresholds must remain frozen after forward evidence arrives.
• Rejecting an alpha is a valid and valuable research result.

13. Future Research Directions

These are research questions, not strategy promises:

• Longer out-of-sample windows
• Cross-exchange robustness checks
• Alternative fragility definitions
• Regime-aware validation
• Market microstructure extensions

14. Reproducibility Scope

Install the focused test dependency:

python -m pip install -r requirements.txt

Run the included tests:

python -m pytest

Generate the compact case summary from included machine-readable outputs:

python scripts/generate_final_report.py

The complete OOS and pipeline-audit runners require the externally archived evidence datasets. They fail closed with a clear message when those files are absent. See data/manifests/EXTERNAL_DATA_MANIFEST.md.

External historical evidence is archived privately and intentionally excluded from this public portfolio package. Complete end-to-end regeneration is therefore not claimed.

Compact summaries, the generated case summary, and focused synthetic-data tests are reproducible from the included files. Full historical regeneration requires the privately archived evidence. The synthetic samples demonstrate schemas and test behavior; they are not the original OOS evidence.

15. Repository Structure

• src/: selected research-only validation and attribution modules
• scripts/: guarded OOS/audit runners and compact report generation
• tests/: focused synthetic-data tests
• reports/: final validation, audit, attribution, and post-mortem evidence
• charts/: representative exploratory and diagnostic figures
• data/sample/: small anonymized examples
• data/schemas/: field documentation
• data/outputs/: compact historical summaries only
• data/manifests/: external-data and package manifests

16. Limitations

• Large evidence datasets are excluded from the public package.
• Included summaries are historical snapshots, not continuously refreshed state.
• Some representative charts predate the final OOS failure and are labeled as exploratory evidence.
• External historical evidence is archived privately and intentionally excluded from this public portfolio package; complete end-to-end regeneration is not claimed.
• Synthetic samples are not the original OOS evidence.
• The package does not establish a profitable or deployable strategy.

License And Use

Code and documentation are provided under the MIT License. This remains a research-only educational case study, not financial advice and not a trading deployment.