The paper that originally motivated this whole DSL. In microsatellite-stable (MSS) metastatic colorectal cancer, the paper's central claim is that the LOCATION of an APC mutation — not DNA-damage-response status — predicts survival. Reproduced directly with the study's own APC_GROUPS classification.

• Within MSS mCRC, APC alteration pattern (mutation location) is associated with overall survival, with unaltered APC carrying the worst prognosisN-terminal APC: HR=0.43 (95% CI 0.32-0.58, p=3.5e-8) vs unaltered; C-terminal: HR=0.62 (95% CI 0.46-0.83, p=0.0015) vs unaltered

The paper's central claim — that a cancer-type-specific top-TMB percentile predicts better immunotherapy survival — needed a live percentile cutoff the DSL didn't have when this was first attempted with a hand-run SQL workaround. It does now: percentileOf scopes the threshold to NSCLC alone, so a pan-cancer study's percentile doesn't mix histologies with very different baseline mutation rates.

• Top-20th-percentile tumor mutational burden (within cancer type) predicts significantly better overall survival on immune checkpoint inhibitorsNSCLC, 80th-percentile cutoff (resolved live to 13.70+ mut/Mb): n=69 high vs n=281 low, median OS 22mo vs 10mo, log-rank p=0.000284

Sidedness-associated genomics and BRAF's link to peritoneal spread both reproduced with real p-values landing almost exactly on the paper's own numbers. One of the paper's other claims (KRAS and ovarian metastasis) did not reach significance here — flagged honestly below as a real miss, not papered over.

• BRAF mutation is markedly enriched in right-sided vs left-sided colon cancer21.2% vs 6.8%, p=7.6e-11
• KRAS mutation is enriched right-sided vs left-sided59.8% vs 38.6%, p=1.4e-10
• BRAF-mutant tumors metastasize to the peritoneum more often than BRAF-wildtype23.1% vs 14.0%, p=0.0057
• KRAS mutation is associated with ovarian metastasisdid not reach significance here (p=0.12-0.30 across several codon-specific attempts) — likely needs a finer, adjusted analysis than "any KRAS mutation" captures; a genuine partial miss, not a tool limitation

Every genomic, clinical, and survival association in this paper's EGFR-mutant-vs-wildtype comparison reproduced cleanly — including the never-smoker enrichment at p=7.5e-19. The paper's ctDNA-versus-tissue finding is now reachable too: the paired view compares each patient's plasma and tissue gene by gene, and plasma-only calls are concentrated in PIK3CA, as the paper says, rather than in truncal drivers like EGFR.

• KRAS mutation is depleted in EGFR-mutant NSCLC (the two are largely mutually exclusive)7.9% vs 29.4%, p=1.1e-15
• TP53 co-mutation rate differs between EGFR-mutant and wildtype NSCLC64.7% vs 46.8%, p=1.3e-7
• Never-smoker status is enriched in EGFR-mutant NSCLC68.5% never/light-smoker (EGFR-mut) vs 38.6% (EGFR-wt), p=7.5e-19
• EGFR-mutant NSCLC has better overall survival than EGFR-wildtypemedian 35.4mo vs 23.6mo, log-rank p=4.4e-4
• Alterations in ctDNA not seen in tissue sequencing (25% of patients, time-matched) disproportionately feature subclonal resistance drivers such as PIK3CA and RICTOR682 patients with both tissue and plasma: 39% of PIK3CA plasma calls (9 of 23) are absent from tissue, against 4% for EGFR (7 of 158) and ~11% for KRAS and TP53. Not time-matched (no sample dates here), so tissue-only calls dominate wherever ctDNA was simply undetectable; RICTOR is on too few plasma panels to test (128 patients profiled on both sides, no plasma-only call)

IDH mutation status, one of oncology's strongest prognostic splits, reproduced almost exactly. The 1p/19q co-deletion that used to be out of reach now reproduces too: the genome view reads it straight off the copy-number segments (1p lost in 94%, 19q in 95% of codeleted oligodendrogliomas), and the codeleted tumours live longer. The paper's cell-cycle finding lands in the right direction with a weaker contrast, and its recurrence story (cell-cycle changes first, mismatch-repair hits only at recurrence) is visible but underpowered in 52 paired patients.

• IDH-mutant glioma has dramatically better overall survival than IDH-wildtypemedian OS 207mo vs 24.2mo, log-rank chi2=249.0, p=4.3e-56
• IDH-mutant patients are younger at diagnosismean age 37.6 vs 57.2, Welch's t p=5.0e-86
• 1p/19q co-deletion separates IDH-mutant glioma into oligodendroglioma (codeleted) and astrocytoma, and the two differ in outcomegenome view on the segments: 1p lost in 94% and 19q in 95% of the 119 segmented codeleted oligodendrogliomas vs ~5% and 22% of IDH-mutant astrocytomas (and IDH-wildtype glioblastoma shows the classic +7 in 77%, -10 in 76-82%); median OS 248mo vs 136mo, log-rank p=6.6e-4
• In astrocytomas, cell-cycle alterations arise preferentially in enhancing tumours (abstract: 44% vs 8%, P < 0.001)OncoKB-driver alterations in CDKN2A/B, CDKN2C, CDK4/6, CCND1/2 or RB1: 46.0% of enhancing vs 19.7% of non-enhancing astrocytomas (grade 2-3, IDH-mutant or wildtype; q=2.2e-5) — the enhancing rate matches, the non-enhancing one is higher than the paper's 8%; the paper's exact gene list and its per-sample pairing are not reproducible from the data as loaded
• Cell-cycle alterations likely preceded the acquisition of alkylating-therapy-associated hypermutationin the 52 patients with a primary and a recurrence sample: IDH1/TP53/ATRX concordant (88-92%), CDKN2A mostly already present at the primary (15 both, 6 recurrence-only), while MSH6/MSH2/PMS2/MLH1 alterations appear only at recurrence (11 recurrence-only, 0 primary-only) — the paper's order, but no single gene is significant at this size

The four-way receptor-subtype genomic comparison reproduced cleanly, and the k-group Welch's-ANOVA/Kruskal-Wallis test this project built specifically closes a gap the first reproduction attempt hit (numeric attributes used to be refused outright above 2 cohorts). The paper's headline — ESR1 and MAPK-pathway alterations enriched after endocrine therapy — was out of reach while no view could separate samples by when they were taken; this study still records no treatment exposure, but its sample sites do say which samples came before any treatment, and against that proxy ESR1, NF1 and ERBB2 all line up with the paper.

• The four receptor subtypes differ genomically (TP53, PIK3CA, CDH1, GATA3, ESR1)TP53 p=1.6e-67, PIK3CA p=7.6e-10, CDH1 p=7.3e-8, GATA3 p=4.6e-7, ESR1 p=3.8e-5 — all match the paper's directions
• Survival differs by receptor subtype, worst in triple-negativemedian OS 176mo (HR+/HER2-) / 46mo (TNBC) / 262mo (HR+/HER2+) / 114mo (HR-/HER2+), k-group log-rank chi2=145.1, p=3.0e-31
• ESR1 ligand-binding-domain mutations are present in 18% of HR+/HER2- tumours collected after endocrine therapy, and rare before itOncoKB-driver ESR1 in 0.7% of HR+/HER2- patients with only treatment-naive primary samples vs 14.3% of those with a sample taken after treatment (q=3.7e-20); within the 42 patients with both, 7 have ESR1 only in the later sample and none only in the naive primary. The proxy is sampling site, not recorded endocrine exposure, so it mixes in tumours treated otherwise
• Activating ERBB2 mutations and NF1 loss-of-function mutations are more than twice as common in endocrine-resistant tumoursany alteration, naive primary only vs sampled after treatment: NF1 2.0% vs 7.4% (q=1.5e-5), ERBB2 2.9% vs 8.2% (q=1.1e-4) — both more than twice as common; MYC also enriched (5.9% vs 10.4%), as the paper reports

The unstratified race comparison reproduced cleanly, matching the paper's FOXA1 and TP53 findings closely. The paper's more careful analysis stratifies by disease stage first — this DSL has no first-class way to hold a second categorical attribute constant while comparing a third.

• FOXA1 alteration rate differs significantly by self-reported raceWhite 14.9%, Black 16.4%, Asian 30.2%, p=0.0044 — highest here in the Asian cohort, with Black still above White
• TP53 alteration rate also differs by raceWhite 29.7%, Black 19.4%, Asian 34.9%, p=0.0119
• The race association should be checked WITHIN disease stage, since stage itself varies by race and could confound the resultno first-class way to stratify by a second categorical attribute — only hand-crossing every stage×race combination into its own cohort, which doesn't scale past two dimensions

The genomic subtype comparison (built from the study's own POLE/MSI/copy-number attributes) reproduced textbook-correctly. The paper's prognostic claim was originally entirely unreachable — this study's outcome data lives under a bespoke column-name pair — and naming those columns directly, a feature this reproduction effort motivated, does close THAT gap; but re-run today, the MSI-H-vs-MSS survival split itself no longer reaches significance, a real reminder that a live warehouse can drift a finding out from under a note that quoted it once.

• Tumor mutational burden falls in a steep gradient across molecular subtypes: POLE (ultramutated) >> MSI (hypermutated) >> copy-number-high/lowPOLE mean 293.1, MSI mean 38.5, CN-high mean 4.6, CN-low mean 5.6 mut/Mb, Welch's ANOVA p=0.0043
• ARID1A and TP53 mutation rates are subtype-specific (ARID1A in POLE/MSI, TP53 in copy-number-high)ARID1A 100%/96.3%/5.5%/27.5%, p=6.8e-16; TP53 0%/14.8%/89.1%/46.3%, p=1.8e-10
• MSI-H endometrial cancer has different progression-free survival (from first-line treatment) than MSSlog-rank chi2=0.39, p=0.53 on PFS_MONTHS_FIRST_LINE_TX/PROGRESSION_STATUS as of this run — NOT significant, though this same query reportedly gave p=0.044 in an earlier check; the warehouse has genuinely moved this result, not just its rounding, which this reproduction states plainly rather than quoting the older number

TP53/RB1 rates by histology matched the paper exactly — but the paper's single most iconic finding, MDM2 amplification in well-differentiated liposarcoma, initially read 0% altered because this study has zero copy-number data at all, and the engine reported that as a real negative instead of refusing. Fixed this session: the same query now discloses the missing profile type rather than reporting a false zero.

• TP53 mutation is markedly enriched in leiomyosarcoma vs well-differentiated liposarcoma57.6% vs 3.8%, p=1.4e-97
• RB1 mutation is also enriched in leiomyosarcoma25.0% vs 2.0%, p=1.1e-32
• MDM2 amplification is well-differentiated liposarcoma's near-universal defining alterationsarcoma_msk_2022 has zero samples with any CNA data — the paper's central finding cannot be checked in THIS study at all, and the view now discloses that explicitly instead of reporting a false 0%

Mutation burden's effect on platinum-based chemotherapy response is at least testable via Cox now, with real per-patient outcome data. The paper's actual headline (TMB and survival specifically on immunotherapy) remains blocked: the relevant outcome data exists for only 40 patients, coded in a way this engine's event-detection convention doesn't parse correctly.

• Mutation burden is associated with platinum-based chemotherapy responseCox: HR=1.02 per mutation/Mb on platinum-PFS, p=0.11 on 187 patients — a real, honestly-reported non-significant result, not the paper's own MSI-based comparison (this release has no precomputed MSI category, only a continuous MSIsensor score too sparse above any standard threshold to split usefully here)
• Tumor mutational burden predicts survival specifically on immunotherapy (the paper's headline claim)the relevant IO-outcome columns exist for only 40 of 345 patients, and their status values don't follow the "1:"-prefixed event convention this engine's survival parser expects — unreachable with usable statistical power even with custom-endpoint-naming

A genuinely federated reproduction — comparing UTUC against a completely separate bladder-cancer study — and it comes back with no gaps at all: every reported direction and magnitude matches known UTUC-vs-bladder biology, and the cross-study caveat wording (fixed this session) is now precise about when a stratified test would and wouldn't apply.

• FGFR3 alterations are more common in UTUC than in bladder cancer46.2% vs 18.7%, p=6.7e-9
• TP53 alterations are less common in UTUC than in bladder cancer21.8% vs 50.4%, p=1.9e-8
• RB1 alterations are markedly less common in UTUC than in bladder cancer1.7% vs 25.3%, p=7.5e-11
• CDKN2A alterations are less common in UTUC than in bladder cancer15.1% vs 38.7%, p=8.7e-7