Oncology Failure Atlas
MethodologyRecurSignal™ Validation

RecurSignal™ Validation

RecurSignal™ Retrospective Signal Validation — Phase 1 Sample Audit

Pharma Innovation Research by Sebastian Azar · Last reviewed: May 2026

About This Sample Audit

This page documents the initial RecurSignal™ Retrospective Signal Validation conducted on a six-record sample from the Phase 1 dataset.

MDIP = Mapped Discontinuation Intelligence Patterns — OFA's discontinuation-pattern taxonomy mapping public stop-reason language into structured categories. Not a biological mechanism.

Its purpose is to review whether RecurSignal™ scores behave consistently with the expected retrospective pattern-similarity logic across selected MDIP™ discontinuation categories.

This is not a predictive validation. It does not test predictive accuracy and RecurSignal™ is not a predictive model. It is not a full statistical validation of all 4,940 classified records. It is an initial retrospective sample audit of score behaviour across selected classified historical records.

RecurSignal™ is a historical recurrence signal, scored 0–100, calculated deterministically after MDIP™ classification. It is not a prediction, probability, forecast, or risk score.

Validation Method

Six trials were sampled from the classified Phase 1 dataset for manual audit.

Trials were selected to represent a range of MDIP™ discontinuation categories, with deliberate inclusion of:

  • EFF and TOX records, where higher RecurSignal™ values were expected
  • STR and REC records, where lower RecurSignal™ values were expected

For each sampled trial, the following were reviewed:

  • Assigned MDIP™ discontinuation category
  • Confidence tier
  • RecurSignal™ score
  • Available source evidence signals from PubMed, OpenFDA, and CrossRef
  • Reasoning note generated by the classifier

Each trial was assessed against three criteria:

  • Is the RecurSignal™ score consistent with the assigned MDIP™ category’s expected retrospective pattern-similarity range?
  • Is the confidence tier consistent with the available source evidence?
  • Is the reasoning note consistent with the available source evidence signals?

Sample Audit Results

Initial sample audit: 5 Pass · 1 Partial · 0 Flagged

No anomalies were identified in the six-record Phase 1 sample. The one Partial result reflects a boundary score within the low recurrence band, not a scoring error.

TrialMDIP™ CategoryRecurSignal™ ScoreResultNotes
Sample 1EFF — Efficacy74PassScore consistent with EFF high-recurrence sample behaviour. Available source signals supported the assigned confidence tier.
Sample 2TOX — Toxicity71PassScore consistent with TOX high-recurrence sample behaviour. Available source signals supported the assigned confidence tier.
Sample 3REC — Recruitment19PassScore consistent with REC low-recurrence sample behaviour. Source evidence profile was consistent with the assigned confidence tier.
Sample 4STR — Strategic22PassScore consistent with STR low-recurrence sample behaviour. Reasoning note supported strategic discontinuation interpretation.
Sample 5REC — Recruitment24PassScore consistent with REC low-recurrence sample behaviour. Available source signals supported the assigned confidence tier.
Sample 6STR — Strategic38PartialScore sits at the upper edge of the low recurrence band. Consistent with STR category behaviour; no anomaly flagged.

Observed Sample Behaviour by MDIP™ Category

The six-record sample audit reviewed whether selected RecurSignal™ scores aligned with expected retrospective pattern-similarity behaviour by MDIP™ discontinuation category.

MDIP™ CategoryObserved Sample RecurSignal™ RangeExpected BandConsistent
EFF — Efficacy70–80High recurrence band (70–100)Yes
TOX — Toxicity65–75High recurrence band (70–100)Yes
DES — Design45–65Medium recurrence band (40–69)Not sampled in this audit
REC — Recruitment15–30Low recurrence band (0–39)Yes
STR — Strategic18–40Low recurrence band (0–39)Yes
BMK — BiomarkerInsufficient Phase 1 dataN/A

Dataset average RecurSignal™: 28.2

Low-band skew is expected in the Phase 1 dataset. REC and STR categories represent 83.6% of classified Phase 1 records. Because these categories carry structurally lower RecurSignal™ values than EFF and TOX categories, the dataset average is pulled toward the low recurrence band.

This is not treated as a scoring anomaly. It reflects the composition of the Phase 1 dataset.

BMK has insufficient Phase 1 data, at 0.3% of classified records, for meaningful sample-level behaviour analysis. The Open Targets biomarker layer required for more reliable BMK scoring is a Phase 2 deliverable.

Partial Result — Sample 6

Sample 6 — STR — Strategic / Portfolio · RecurSignal™ 38 · Partial

Sample 6 was assessed as Partial rather than Pass because the RecurSignal™ score, 38, sits at the upper boundary of the low recurrence band, 0–39. It does not cross into the medium recurrence band.

Review finding

The reasoning note identifies a strategic portfolio discontinuation with sponsor pipeline context. The score is consistent with STR category behaviour.

The Partial assessment reflects the boundary position of the score, not a classification or scoring error.

No anomaly was flagged. The Partial result supports the need for transparent band-boundary interpretation, not a concern about score logic.

Sample Audit Conclusion

The initial RecurSignal™ Retrospective Signal Validation found that RecurSignal™ score behaviour was consistent with expected retrospective pattern-similarity logic in the six-record Phase 1 sample.

Key findings

  • In the sampled records, EFF and TOX showed high-band RecurSignal™ scores
  • In the sampled records, REC and STR showed low-band RecurSignal™ scores
  • The dataset average of 28.2 is consistent with REC/STR dominance in the Phase 1 dataset
  • The one Partial result reflects a boundary score, not a miscalibration
  • No anomalies were flagged in the six-record sample

Initial audit status: PASS for sampled records

RecurSignal™ score behaviour was consistent with expected retrospective pattern-similarity logic in the six-record sample. No Phase 1 recalibration signal was identified from this sample.

Scope and Non-Claims

  • This validation confirms score behaviour in a six-record retrospective sample, not predictive accuracy. RecurSignal™ is not a predictive model.
  • This validation was conducted on a sample of 6 trials from the Phase 1 dataset of 4,940 classified records. It is not a full audit of all classified records.
  • RecurSignal™ does not determine the definitive cause of any trial’s discontinuation. It reflects retrospective pattern similarity only.
  • RecurSignal™ does not make clinical, regulatory, investment, or medical recommendations.
  • A full independent second-pass re-run and expanded validation are planned for Phase 2.
  • This sample audit does not validate predictive accuracy and does not establish a true probability of discontinuation. It reviews whether RecurSignal™ score behaviour is consistent with retrospective pattern-similarity logic in a small six-record sample.
← Back to Methodology

Oncology Failure Atlas · Pharma Innovation Research by Sebastian Azar · oncologyfailureatlas.com · RecurSignal™ is a proprietary scoring signal. All rights reserved.