Flow Archive

Flow Observatory / Lambda-C Sector Flow Engine

Date: May 2026

Version: Research Memo v1.0

---

1. Executive Summary

Flow Observatory’s ETF-based LRAC engine evolved from V3.1R Momentum Quality into V3.2 Reduced Mode after validation showed that the V3.1R system was disciplined but overly selective.

The main weakness of V3.1R was not the beta guard or the exposure mandate. Those controls worked well. The key issue was the not_enough_filtered_sectors rejection rule. In several months, V3.1R rejected signals because it could not form a full 2-long / 2-short structure, even though the underlying sector-flow information still contained usable relative strength.

V3.2 addresses this by introducing a reduced-gross, reduced-shape mode. When a full 2L/2S signal is unavailable, V3.2 attempts smaller structures such as 1L/1S, 1L/2S, or 2L/1S, while preserving the same beta, dollar, and gross exposure discipline.

In the 2023-01 → 2026-04 validation window, V3.2 improved total return, Sharpe, acceptance coverage, and random benchmark percentile, while only slightly increasing max drawdown.

2. Background: From V3.1R to V3.2

V3.1R introduced three important improvements over the earlier LRAC engine:

• Momentum Quality Filter
  Long candidates needed positive rolling momentum and acceptable current-return range. Short candidates needed negative rolling momentum and acceptable current-return range.
• Mandate Guards
  Signals were rejected if they breached beta exposure limit, dollar exposure limit, or gross exposure tolerance.
• Rejected Signal Audit
  Invalid or unsafe signals were written separately instead of contaminating official performance records.

This made V3.1R safer and cleaner. However, validation revealed that the system often refused to trade not because the signal was dangerous, but because it could not form a full symmetric 2L/2S portfolio.

3. V3.1R Validation Result

Initial V3.1R validation over the 2023-01 → 2026-04 window produced the following results:

V3.1R was therefore highly disciplined, but conservative.

The rejection reasons were:

This indicated that most rejected months were not rejected because of beta exposure breach, but because the system could not form a full 2L/2S structure.

4. Shadow Audit: Were Rejected Months Actually Dangerous?

A shadow audit was performed to test what would have happened if rejected months had been force-traded.

This suggested that rejected months still contained useful sector-flow information. However, raw returns alone were not enough. Some of this performance could have come from market beta.

5. Beta-Adjusted Shadow Audit

To isolate whether the rejected-month performance came from genuine sector-flow information or from market exposure, beta-adjusted shadow returns were calculated:

The key result:

This was the decisive finding. The beta_exposure_breach rejection appeared justified. These months had high beta exposure and negative solver performance.

But the not_enough_filtered_sectors rejection looked too strict. Even after beta adjustment, the rejected months remained strongly positive.

This indicated that V3.1R was likely missing valid sector-flow opportunities.

6. V3.2 Design Decision

V3.2 was created to solve this specific problem.

The goal was not to weaken the risk controls. The goal was to preserve V3.1R’s discipline while allowing the system to express weaker but still valid sector-flow structures.

V3.2 Rule

If full 2L/2S mode is available:

If full 2L/2S mode is unavailable because of not_enough_filtered_sectors:

Reduced mode attempts smaller shapes:

Reduced mode uses lower gross exposure:

The same mandate guards remain active:

Thus, V3.2 does not become aggressive. It becomes more adaptive.

7. V3.2 Candidate Backtest Result

The V3.2 candidate was tested over the same 2023-01 → 2026-04 window.

V3.2 improved return by +2.73 percentage points while increasing max drawdown by only +0.40 percentage points.

This suggests that V3.2 recovered useful opportunities without materially damaging risk control.

8. Reduced Mode Contribution

V3.2 accepted 6 reduced-mode months:

Reduced-only performance:

The beta-adjusted result is especially important. It indicates that reduced-mode contribution was not simply caused by market beta exposure.

9. Reduced Months Detail

Best reduced month:

Worst reduced month:

10. Interpretation

V3.2 appears to improve the LRAC framework in three ways:

1. Better Signal Coverage

V3.2 reduces the rejection rate from 55% to 40%.

This means the system can act in more months while still avoiding unsafe beta-exposure breaches.

2. Better Use of Incomplete Flow Structures

V3.1R required full symmetry. V3.2 recognizes that a market does not always present clean 2L/2S structures.

Sometimes a smaller 1L/1S or 2L/1S relationship carries valid information.

3. Preserved Risk Discipline

V3.2 increases average beta exposure from 0.0070 to only 0.0167, while reduced-mode average beta exposure remains 0.0458.

Both remain comfortably below the 0.10 mandate threshold.

This suggests that V3.2 improves adaptability without becoming uncontrolled.

11. Remaining Risks

V3.2 is not yet a proven production strategy. It is a stronger research candidate.

Key risks remain:

• Small Sample Size
  Only 40 months were tested, with only 6 reduced-mode months.
• Reduced Short-Leg Risk
  Negative reduced months were mostly caused by short legs moving against the system.
• Potential Overfitting
  V3.2 was designed after observing V3.1R rejection behavior. It should be monitored live before further optimization.
• Benchmark Context
  V3.2 should not be compared directly with SPX as a return-maximizing long-only benchmark. Its relevant comparison is against matched long/short random portfolios, simple momentum, and earlier LRAC versions.

12. Decision

V3.2 should become the current primary Flow Observatory ETF research candidate.

It should not yet be marketed or treated as a proven investment strategy. Instead, it should be monitored as an audit-ready sector-flow research engine.

Recommended operational decision:

13. Next Steps

• Continue using V3.2 engine and tracker for monthly live observation.
• Run the V3.2 official tracker after each realized month closes.
• Accumulate at least 3–4 additional live months.
• Review live performance separately from backtest performance.
• Only after additional live data, consider whether a V3.3 reduced-short safety filter is justified.
• Begin designing a Flow Observatory archive/dashboard after the V3.2 evidence base matures.

14. Final Positioning

V3.2 does not prove that Flow Observatory is a finished strategy.

But it does show that the sector-flow hypothesis is becoming testable, versioned, auditable, and empirically meaningful.

The core insight is:

V3.2 captures this insight by allowing reduced-gross, beta-controlled expressions of incomplete flow structures.

This makes V3.2 the strongest LRAC candidate so far.