BEFORE THE DESERT: IMMUNE REPERTOIRE ARCHITECTURE, STROMAL PERMISSIVENESS, AND THE SELECTION OF LONG-TERM SURVIVORS IN PANCREATIC DUCTAL ADENOCARCINOMA The Stroma-First vs Repertoire-First Named Binary: Why Two Competing Immunotherapy Paradigms Both

Structured Abstract Background Pancreatic ductal adenocarcinoma (PDAC) has a five-year survival rate of approximately 12%. The field is divided between two immunotherapy paradigms that each address a genuine biological mechanism but share one prediction failure: neither reliably identifies, before treatment, which patients will achieve long-term survival. The Stroma-First paradigm holds that desmoplastic stroma physically excludes T cells — the solution is stroma remodelling (TGF-β inhibitors, hyaluronidase, FAK inhibitors). The T-cell Activation paradigm holds that checkpoint suppression inactivates tumour-specific T cells — the solution is checkpoint blockade (anti-PD-1, anti-CTLA-4). Neither paradigm achieves response rates above 10% in PDAC, and neither can predict pre-treatment which patients respond to its own intervention. Gap The two competing paradigms share a structural silence: neither asks what property of the patient's pre-treatment immune architecture determines the ceiling of what either intervention can achieve — a property encoded in the naive and memory TCR repertoire before any tumour cell has been recognised. The oncologist at the moment of diagnosis has no patient-side architectural biomarker independent of tumour characteristics. PDAC survival prediction is entirely tumour-side and microenvironment-side. No patient-side architectural variable with a pre-registered head-to-head test against the best combined Stroma-First + T-cell Activation baseline has ever been proposed. Approach We introduce the Stroma-First vs Repertoire-First Named Binary and the IRITR-PDAC framework, which identifies the Effective Tumour-Surveillance Diversity Index (TSRDI_eff = TSRDI × H_TME) — a dual-architecture product combining the patient-side Unexhausted Structural Reserve Capacity (USRC, operationalised as TSRDI) with the tumour-side infiltration permissiveness factor (H_TME, using the two-component additive formula: α × ChemoScore + (1−α) × (1−BarrierScore)). Three anomalies unresolvable by either existing paradigm are identified. A novel TSRDI × H_TME interaction prediction is derived: high TSRDI actively elevates H_TME through peri-tumoral CXCL9/10 induction in tumour-draining lymph nodes, partially overcoming the stromal barrier. A pre-registerable CCS is specified using three published cohorts with true pre-treatment peripheral blood TCR sequencing. Results The Stroma-First vs Repertoire-First framing reveals that the two competing paradigms have been explaining the wrong level of the problem: both address what happens after the surveillance attempt fails; neither addresses what architectural property determines whether the surveillance attempt can succeed. IRITR-PDAC resolves three anomalies (long-term survivor paradox, neoadjuvant immune priming paradox, stroma paradox) that are unresolvable by either paradigm. The TSRDI × H_TME interaction term is the article's most distinctive empirical prediction: it is not testable within either existing paradigm but is directly testable in TCGA PAAD RNA-seq + TRUST4 data and in the three primary CCS cohorts. Implications If confirmed, IRITR-PDAC provides the first pre-treatment patient-side biomarker for PDAC survival prediction independent of tumour characteristics. The clinical sequencing implication is a concurrent strategy: FOLFIRINOX at standard of care with STING agonist + IL-7 + shared KRAS neoantigen vaccine during recovery windows, followed by checkpoint inhibition. This addresses the repertoire insufficiency that neither Stroma-First nor T-cell Activation alone can overcome. Weil Protocol: REQUIRED — L3. Keywords: Stroma-First, Repertoire-First, IRITR-PDAC, TSRDI, H_TME, USRC, pancreatic cancer, PDAC, desmoplastic stroma, checkpoint inhibitor, long-term survival, KRAS, CXCL9/10, neoadjuvant, TCGA PAAD, TRUST4, ARI principle, ALGUILAS-AI