Detailed critique — strengths, gaps, and contradictions

• Strength — accurate problem framing: the review correctly catalogs the real-world fragmentation of PD-L1 assays (22C3/28-8/SP263/SP142), the different scoring systems (TPS/CPS/IC) and their regulatory/companion roles, which is supported by comparative studies and FDA listings.
  Zdrenka et al. diagnostic overview
• Strength — multi-modal prospects: the paper sensibly reviews orthogonal approaches (mIF, QIF, PCR/NGS, FISH for copy number, sPD-L1 ELISA, ctDNA/CTCs, imaging SPECT/PET, iFRET, AI). These are the right directions for reducing false negatives and capturing dynamics.
  MITRE mIF reproducibility
• Key gap — lack of prospective meta-analysis / trial roadmap: the review catalogues methods but does not provide a prioritized, tumor‑specific validation roadmap (e.g., which cancers should test deglycosylation + mIF + ctDNA in a randomized prospective design). That gap matters because policy and reimbursement require prospective validation.
  ESMO MSI testing recommendations
• Counterpoint / blindspot — clinical predictive value not equal to analytical concordance: the review sometimes conflates assay concordance with predictive validity (i.e., similar staining ≠ same association with response). The CLOVER study and others show PCR/RT-PCR can concordantly call negatives but are poor at identifying positives that predict response — this distinction should be emphasized.
  CLOVER PD-L1 concordance
• Methodological concern — statements on assay interchangeability should be quantified: the review cites interchangeability (e.g., 22C3, 28-8, SP263 similar) but does not present concordance statistics (kappa, sensitivity/specificity) or indicate where thresholds cause major reclassification; readers need those metrics to estimate clinical impact of assay switching.
  Blueprint Phase 2 comparability

Concrete, evidence‑based recommendations (actionable)

1. Adopt tumor‑type specific pipelines. For NSCLC, use 22C3/28-8/SP263 interchangeably only after validating local kappa at clinical cutoffs; for HNSCC/TNBC use CPS-specific validation cohorts; for rare cancers use NGS-first if available. (See ESMO consensus for MSI as template.)
   Zdrenka recommendation for subtype pipelines
2. Prospectively evaluate deglycosylation pre‑treatment of FFPE slides in a randomized diagnostic sub‑study. Protocol: randomized split-sample IHC vs deglycosylation + IHC, correlate with response in pembrolizumab-treated cohort; primary endpoint: increase in sensitivity for responders without unacceptable false positives.
   Deglycosylation protocol paper
3. Mandate multi‑marker reporting in trials. Clinical trials using ICIs should report PD-L1 (standardized clone), TMB (panel & cutoff), MSI/MMR status, and baseline ctDNA — permit retrospective multi-parameter modelling and independent validation (example: TMB cutoffs vary; report exact platform and mutation counts).
   TMB-pembrolizumab pan-tumor analysis
4. Use digital pathology/AI as QC, not sole decision-maker yet. AI-assisted TPS/CPS scoring can reduce interobserver variability, but needs prospective clinical outcome calibration and regulatory validation before replacing pathologist judgement.
   Deep learning PD-L1 from H&E
5. Prioritize prospective, randomized diagnostic‑therapeutic trials. Only RCTs that randomize diagnostic pathway (e.g., standard IHC vs multi‑modal test-guided therapy) with clinical endpoints (OS/PFS/ORR) will show whether improved PD-L1 assessment changes outcomes — narrative reviews cannot substitute for these.

Where the paper performed well vs where it should be improved

Essential citations used in this critique (selective, evidence-focused)

• Zdrenka et al. PD-L1 review
• Blueprint phase 2
• CLOVER PD-L1/PCR analysis
• MITRE mIF reproducibility
• Deglycosylation protocol (PNGaseF)
• TMB-pembrolizumab pan-tumor

Confidence, limitations, and falsifiability

Confidence in the review's descriptive parts (assay lists, alternatives, known heterogeneity) is high because they accurately compile many primary studies; confidence that the proposed solutions (deglycosylation, mIF, AI, ctDNA) will improve patient selection is moderate — each requires prospective validation. The null/falsifying test: a randomized trial showing that a refined pipeline (e.g., deglycosylation + mIF + ctDNA) does not improve response or harms specificity would overturn the main applied recommendations.

Immediate, practical next steps (for researchers & diagnostics labs)

1. Run a pilot split‑sample study: standard IHC vs deglycosylation+IHC, compare classification changes and correlate with retrospective response data where available.
2. Standardize reporting in trials: always report PD-L1 clone, platform, TPS/CPS calculation rules, TMB platform & numeric count, MSI method, and ctDNA assay used.
3. Create multi‑center prospective diagnostic‑therapeutic trials randomizing therapy assignment by standard vs multi‑marker test to measure clinical benefit.