Has RegimeSignal been independently audited?

Yes. A PhD-level independent replication and robustness audit reproduced all four locked signals (MBS T1 and MBS T2 bit-exact), found no feature-level label leakage, and confirmed every signal sits above the permutation null. Determination: Validated with Qualifications.

How does the audit differ from the earlier PhD validation?

The audit is a deliberately stricter, independent replication and robustness audit conducted by a second PhD auditor. It goes beyond confusion-matrix reproduction to add permutation-null checks, alternative-classifier comparisons, and a forward-shifted out-of-sample window.

Are the audit findings unconditional?

No. The determination is 'Validated with Qualifications.' Event counts are small, the RRS operating point is precision-first and exploratory, and every qualification is disclosed in full on this page.

← Report 1 · PhD Validation

Independent PhD Audit · Report 2 of 2

Independently audited — all four signals across the full S&P 500 market cycle

Determination · Validated with Qualifications

~86%

bear-market precision · flagship BRS · reproduced exact

✓ Genuine signal, confirmed. Across 1,000 shuffled-label trials, all four signals scored far above random chance — zero matched the real model.

A deliberately stricter, independent PhD replication and robustness audit of RegimeSignal™'s four S&P 500 prediction signals — the full market cycle — designed not to duplicate the validation. Auditor: Dr Jamal Hossain, PhD · July 8, 2026.

PhD Independent Auditor

Dr Jamal Hossain, PhD

Lecturer in Applied Health Statistics · University of Southampton

PhD · MSc, Statistics (with distinction) · BSc (Hons), Statistics · FHEA · Member, Statistical Sciences Research Institute (S3RI)

PhD in Social Statistics & Demography (University of Southampton, 2023), completed as a Commonwealth Scholar. Peer-reviewed publications in journals including JAMA Network Open and PLoS ONE, with research spanning statistical modelling, multilevel & longitudinal analysis, classification metrics, reproducibility, and machine-learning methods. ORCID: 0000-0002-2728-1055.

Independence: No role in the model's construction and no financial relationship with Cronus Market Intelligence beyond the audit fee. The engagement was limited to statistical audit, replication, implementation-integrity, and robustness checks — not an assessment of market or investment merit.

Four signals across the full cycle

BRS — Bear market (−20%)

−20% · audit-clean

MBS T1 — Pullback (−5%)

−5% · audit-clean

MBS T2 — Correction (−10%)

−10% · audit-clean

RRS — Recovery (+10%)

+10% · exploratory*

Signal	Precision	AUC	Repro
BRS — Bear market (−20%)	85.7%	0.91	Exact
MBS T1 — Pullback (−5%)	83.3%	0.88	Bit-exact
MBS T2 — Correction (−10%)	84.4%	0.92	Bit-exact
RRS — Recovery (+10%)	81.8%*	0.76	Exact

What the audit confirmed

✓Every signal sits far above the permutation-null baseline — genuine signal, not artifact.
✓Benchmarked against three independent alternative learners on identical data — all within a tight AUC band; one scored marginally higher. Results are not algorithm-picked.
✓Three of four signals (BRS, MBS T1, MBS T2) reproduced audit-clean and well-supported on the out-of-sample record.
✓All 35 locked files checksum-verified and matched; four signals reproduced with no divergence.
✓MBS T1 and MBS T2 reproduced bit-exact (~1×10⁻⁸; every one of 154 fire/no-fire decisions).
✓No feature-level label leakage — the predicted month is excluded and features are lagged one month.
✓The Market Break signals degrade gracefully under a 12-month cutoff shift.

⚠ Qualifications — disclosed in full

RRS — real signal, conservative claim. RRS cleared the same chance test, so the signal is genuine. The one caveat is narrow: its 82% was measured on the window used to tune it, so we publish it as a historical upper bound and treat ~69% as the honest prospective estimate. Precision-first · recall ~39% by design.

Small event counts across all signals (67 / 46 / 33 / 23) — bands are wide. No purge/embargo for the 4-month label horizon; BRS reproduced on a locked Layer-1 composite.

* RRS 82% = 9 of 11 fires correct at the locked high-conviction threshold, reproduced exactly. Because the threshold was chosen on the reported window, it is in-sample-optimistic; prospective estimate ~69% (Wilson 95% CI ≈ [0.52, 0.95]). Presented as exploratory pending out-of-sample data postdating the lock.

Full report available to qualified reviewers under NDA on written request to info@cronusmarketintelligence.com.

Independent review of the documented walk-forward methodology and reproducibility of the headline confusion matrices from locked artifacts as of June 3, 2026. Not an opinion on future results; does not validate production deployment; not investment advice. Past statistical results are not indicative of future results.