Registers: assumptions, hypotheses,
Registers: assumptions, hypotheses, variants, and readiness — It elevates core uncertainties into explicit assumptions so that risks can be managed and progressively tightened by tests and data (LOCK → Derive → Gate). Rather than deleting ambiguity, this register isolates key uncertainties as explicit competing hypotheses.
This register does not hide ambiguity. It elevates core uncertainties into explicit assumptions so that risks can be managed and progressively tightened by tests and data (LOCK → Derive → Gate).
Assumption register (AR)
This register does not hide ambiguity. It elevates core uncertainties into explicit assumptions so that risks can be managed and progressively tightened by tests and data.
ID Assumption Risk / verification (summary) ID Assumption Risk / verification (summary) AR-1 A rapid uplift/expansion on the Pacific side corresponding to Δ R can occur. If Δ R must be unrealistically large, STOP under the Ω-NoGo constraints. AR-2 Tensile stress concentrates near the Atlantic antipode, and weak suture zones act as preferential paths. If the inferred initiation region contradicts the predicted weak paths, C1 is weakened. AR-3 Rupture propagates rapidly in a “zipper-like” manner, creating a long rupture zone within a short time. If the propagation window must be long for feasibility, HOLD. AR-4 The void need not be a perfect vacuum, but an effective pressure deficit Δ P is sustained/repeated long enough to matter. If required Δ P is physically unrealistic, STOP. If the maintenance time τ is too short, HOLD via the impulse constraint. AR-5 Fluid overpressure (P_f) reduces effective stress σ' (hydraulic jacking), and residual resistance is limited by viscous shear so that μ_eff becomes sufficiently small. If required h is unrealistically thin, STOP. If τ≳τ_drain, HOLD/FAIL via AR-10. If lubrication/overpressure signatures are absent, weaken/HOLD. AR-6 Pacific–Atlantic RSL asymmetry can remain as an observable signature of the mechanism. If it does not separate after corrections (GIA / vertical land motion / sedimentation), weaken P3. AR-7 The rupture nucleation point and propagation order can be inferred from geological records (early opening markers). If no propagation order can be inferred at all, AR-3 is weakened/HOLD. AR-8 At least one realistic multiphase-lubrication candidate (F1–F4) can reduce μ_eff sufficiently. If no candidate works, C3 FAIL (friction/heat dissipation No-Go). AR-9 Total energetics and heat dissipation are consistent with independent observables (thermal/fluid/magnetic remnants). If the energy budget is physically excessive or signatures mismatch, HOLD/STOP. AR-10 The event window τ is shorter than the drainage timescale τ_drain (undrained), so overpressure P_f is maintained and σ'=σₙ-α P_f becomes sufficiently small (hydraulic jacking). Estimate τ_drain from k,,Sₛ. If τ≪τ_drain cannot be satisfied, HOLD/FAIL. AR-11 A significant fraction of frictional/dissipated energy can be diverted into phase-change latent heat (m_w L_v) and hydrofracturing, so pervasive melting signatures may not be strictly required. Compare an upper bound on m_w L_v against Q_fric. If melting is absent, hydrofracture/fluid-involvement signatures become mandatory; if neither is present, FAIL. AR-12 The lubricating/overpressure fluid supply may include in-situ generation (e.g., dehydration of hydrous minerals at the crust–mantle boundary) in addition to external seawater injection. Verify against the required fluid volume V_req (fluid-budget accounting) and mineralogical/hydrothermal signatures (P4). If volume is excessive or signatures are absent, HOLD/FAIL. AR-13 (Cross-validation) In dynamical orogeny (P9), a Deborah-number (De) based “fast/slow” regime separation is a first-order guide for crustal deformation mode (brittle vs viscous). If the relaxation time τ_relax is mis-defined, the regime classification becomes unstable. In P9, pre-register multiple τ_relax assumptions (sensitivity) and verify against independent thermochronology/structural signatures. AR-14 (Cross-validation) For Atlantic sediments (P10/P11), a “pure preserved sediment mass” and observed thickness (grid/core) can be compared on the same scale. Shelf/delta sequestration, dissolution/resuspension, and bottom-current redistribution can break a simple budget. In P10, require a three-way comparison: (i) basin mass budget, (ii) core-based sedimentation rates, and (iii) thickness grids; decide HOLD/FAIL accordingly. AR-15 (Cross-validation) In volcanic synchronization (P12), after defining a “major edifice-building phase” (e.g., T₈₀) and controlling preservation bias (erosion/collapse/subsidence), global clustering can be assessed. If selection bias is large, “only young volcanoes remain” can create false spikes. In P12, pre-register selection rules/weights/preservation corrections and cross-validate with negative controls (inactive arcs/hotspots). AR-16 (Cross-validation) The “kinetic-to-potential energy” threshold check in P9 is not evidence against standard orogeny; it is used only as an auxiliary check of the velocity/energy scale implicitly required when claiming an instantaneous (inertial) pile-up. Arbitrary choices (e.g., efficiency η_conv) can inflate or deflate the estimate. In P9, pre-register the allowed range of η_conv together with an alternative long-term work-accumulation model, and interpret only in conjunction with thermal/structural signatures. AR-17 (Operations / reproducibility) External datasets used in P9–P12 (sediment thickness grids, crustal-age grids, volcanic edifice construction ages, etc.) are pinned with version/access-date/CRS in provenance and can be consistently reprojected/mapped. Dataset mismatches (resolution/masks/time scale) can create spurious signals. Pin sources and preprocessing rules in docs/provenance.md, and pre-register QA flags and exclusion rules.AR-18 (Additional evidence) In P14, the kinematic tail form (decay function) and the effective shear-zone thickness h_eff can materially change conclusions. Pre-register (v₀,p,t₀,h_eff) and filter by cross-gates with heat-flow/partial-melt constraints. AR-19 (Hydrology) The P15 approximation Δ SL≈ -Δ V/A_ocean is first-order; regional RSL differs due to geoid/elastic/gravity redistribution. Separate global vs regional sea-level records and decide FAIL/UNLOCK by whether both can be satisfied simultaneously. AR-20 (Climate) P16 freshwater signals require temperature–salinity separation, and chronology alignment (age models) can be uncertain. Require multi-proxy agreement (temperature + salinity + circulation) and include age-model sensitivity analysis. AR-21 (Isotopes) The P18 open-system model is a central vulnerability for “young-event” coupling; if not reproducible in modern samples, the coupling collapses. Place an early gate TEST-ISO1 using modern samples (historical lavas / inherited-zircon cases). AR-22 (Biogeography) P17 divergence times are sensitive to molecular-clock and calibration choices, and sampling bias can be large. Operate P17 primarily as a falsification check: if many candidates consistently show t_div≫ T_event, STOP. AR-23 (Sea-level budget) P19 is sensitive to whether the sea-level budget (altimetry/GRACE/Argo) fails to close within uncertainty. Fix the same time period and baseline, propagate uncertainties, and require source-sensitivity analysis (TEST-SLB1). AR-24 (Basin-volume proxy) Estimating Δ V_proxy depends on ridge cooling subsidence, area definitions, and geoid effects. Use multiple proxies (subsidence rate / crust production rate / area change) and pre-register upper/lower bounds. AR-25 (Geomorphology) The P20 misfit index can be easily confounded by lithology, uplift, glaciation, and human impacts. Require controls for lithology/uplift rate/glacial influence and matched negative controls; otherwise HOLD. AR-26 (Delta chronology) Definitions of “delta initiation” and dating methods (¹⁴C/OSL/stratigraphy) vary, which can create spurious clustering. Fix the onset definition and method, model preservation bias (detection limits), then compute the clustering index (TEST-DELTA1). AR-27 (Kinematics) Comparing geological vs GPS velocities (P21) can be biased by reference frames and averaging windows. Convert to the same reference frame, separate the plate-reorganization alternative (H-DEC), and run a global bias test (TEST-DEC1). AR-28 (Oil) P22 competes with the standard explanation that oil distribution is dominated by basin/source-rock/thermal maturity rather than “transport.” Require a null model based on basin distribution and evaluate only incremental association; forbid UNLOCK based on significance alone. AR-29 (Ice / refugia) In P23, separating heat-flow control from climate/precipitation/topography control can be difficult. Without multivariate comparison including climate reconstructions (temperature/precipitation) covariates, HOLD. AR-30 (Endorheic lakes) P24 lake-level transitions can arise naturally from pluvial (glacial–interglacial) cycles. To claim a single-event hypothesis, first show that ages cluster within a narrow window (TEST-LAKE1). AR-31 (Stratigraphy) P26–P28 (Great Unconformity/polystrata/coal) have high risk of global-synchrony ambiguity, preservation bias, and composite local events. If global synchronization is not observed, treat as FAIL/HOLD immediately and separate from core (C1–C3) conclusions. AR-32 (Cross-evidence integration) P19–P29 proxies may not be fully independent, risking overconfidence via “evidence redundancy.” In P29, pre-register proxy_class labeling/weights/representative-value rules and block overconfidence with permutation/negative controls.AR-33 (Event-window alignment) Aligning records with different age models/errors into one event window can create spurious simultaneity. Use a coherence metric (K_joint) that propagates age uncertainty and require age-model ensemble alignment. AR-34 (Confounds / negative controls) P30 and cross-evidence can be over-trusted due to selection bias, omitted covariates, or missing controls. Pre-register controls_registry.csv; without passing permutation/negative-control gates, forbid ERL UNLOCK.AR-35 (Coal rank) Coal rank is strongly controlled by temperature/time/burial depth; subjective “compaction indices” cannot support causality. In P31, control for burial proxies (e.g., Rₒ, TTI) and assess only incremental explanatory power of deformation proxies. AR-36 (Oil budget) With generation/migration/leakage/biodegradation/mixing, basin-scale source-to-reserve budgeting has very large uncertainty. In P32, propagate upper/lower bounds on conversion efficiency, missing source rock, and volume errors; if consistent within uncertainty, HOLD. AR-37 (Oil chronology) Direct chronometers (Re–Os, etc.) can be distorted by mixing/contamination/open-system behavior, but repeated agreement can be a strong falsifier. In P34, if independent chronometers repeatedly agree on deep-time ages, V-REC-family strong claims are HOLD/FAIL. (Current execution: deep-time agreement ⇒ V-REC FAIL.) AR-38 (Oil thermal window) Oil preservation depends on a nonlinear “oil window” in time–temperature and on biodegradation/gasification. In P33, record geothermal/intrusive/maturity indicators and check internal consistency (TEST-TEMP1). AR-39 (Oil discovery bias) Oilfield maps are sets of “discovered” fields and may be biased by exploration intensity, cover (sediment thickness/deltas), geophysical difficulty (salt/gas), and accessibility. In P35, model discovery probability with exploration-maturity covariates; if bias is large, downgrade P22/P32 evidence grade. AR-40 (Oil distribution) Without a sedimentary-basin mask, oil absence in basement/exposure areas is a trivial necessary condition, increasing false-positive risk. In P22, restrict comparisons to samples with basin/source-rock potential; include basin masks and petroleum-system covariates in the null model (TEST-OIL1). AR-41 (Oil biomarkers) Terrestrial biomarkers can enter marine records via normal river/ocean transport; glacial “bulldozing” is not required. In P22, include controls for river input/catchment area/coastal deposition; pre-register biomarker metrics/thresholds/sampling windows (TEST-OIL2).
Competing hypotheses register (Hx)
Rather than deleting ambiguity, this register isolates key uncertainties as explicit competing hypotheses. Each hypothesis must have a testable/falsifiable decision rule, and PASS/FAIL thresholds must be pinned in pre-registration (e.g., constraints.yml) or an equivalent registry file.
ID Competing hypothesis (summary) Key decision / falsifier (summary) ID Competing hypothesis (summary) Key decision / falsifier (summary) H-STD Standard baseline: long-term ridge spreading plus cumulative margin subduction/collision forms the Atlantic. If P1–P5 mostly FAIL while the baseline explains with fewer assumptions, H-STD dominates. H0 Magnetic striping as chronology (H0): stripes record geomagnetic reversals (GPTS) in time. If P8 shows a strong H0 fit (TEST-M1), H2 is FAIL or downgraded. H2 Magnetic striping as resonance (H2): catastrophic/rapid cooling freezes resonance/standing-wave patterns in space. If H0 systematically fails while H2 predicts better and matches independent signatures, H2 becomes an UNLOCK candidate. H-E1 Energy source (H-E1): an internally stored trigger (jamming–unjamming; materials/VP frame) produces Δ R. If the energy budget (AR-9) and accompanying signatures (Appendix C) are absent, HOLD/FAIL. H-E2 Energy source (H-E2): an external trigger (impact/influx, etc.) produces Δ R. If impact/injection signatures are absent and required energy is excessive, FAIL. H-E3 Energy source (H-E3): electromagnetic residue/impulse remnants constrain or explain event scale. If W_total does not match independent observations (magnetic/thermal), FAIL. H-S1 Fluid source (H-S1): external injection of seawater/brine along rupture/fault zones. If P4 fluid-involvement signatures and the fluid budget are inconsistent, FAIL. H-S2 Fluid source (H-S2): in-situ generation driven by dehydration reactions. If mineralogical/hydrothermal signatures and V_req mismatch, FAIL. H-S3 Fluid source (H-S3): multiphase transitions (supercritical/bubbles/fluidization) maintain/distribute fluids. If phase-change/multiphase signatures are absent and Q_fric cannot be “hidden,” FAIL. H-ORO (Cross-validation) Orogeny: long-term viscoelastic accumulation (De≪1) vs rapid dynamical pile-up (De≫1). Decide using P9 via thermal–structural disequilibrium and fast (or slow) P–T–t paths. H-SED (Cross-validation) Atlantic sediments: long-term accumulation (tens to hundreds of Myr) vs short accumulation time (young opening). Decide using P10 (thickness–age) + P11 (Mn nodules / low-sedimentation indicators). H-VOLC (Cross-validation) Volcanism: long-term random/continuous generation vs a global pulse (narrow-window clustering). Decide using P12 via clustering metrics after preservation-bias control. H-FRIC (Additional evidence) Plate friction/driving: weak-boundary continuous drift vs threshold-crossing stick-slip/runaway slip. Decide using P13 via R_τ and thermal/stress upper bounds (threshold necessity vs non-necessity). H-TAIL (Additional evidence) Velocity/thermal “tail”: steady mantle background vs a recent decaying tail plus frictional heating. Decide using P14 by jointly satisfying v(t) fit and heat-budget gates. H-DRAIN (Additional evidence) Canyons / great drainage: long-term sea-level cycles + turbidite accumulation vs rapid drainage from basin-volume increase. Decide using P15 via Δ SL gates and canyon timing clustering. H-FW (Additional evidence) Freshwater shock: standard ice-lake outbursts/ice-sheet retreat vs basin-opening sink coupling. Decide using P16 via multi-proxy agreement and spatial simultaneity. H-BIO (Exploratory) Trans-Atlantic divergence: deep-time (Myr) divergence vs very recent (kyr) divergence. Use P17 divergence-time compilation for immediate FAIL/HOLD/STOP decisions. H-ISO (External coupling) Chronology/isotopes: a closed-system “clock” vs an open-system “thermal–diffusion recorder.” Decide using P18 and P34 based on modern/petroleum-system reproducibility (TEST-ISO1, TEST-PETAGE1). H-SLB (Cross-evidence) Sea-level budget: the budget closes within uncertainty; residuals are measurement/correction/model mismatch. If TEST-SLB1 yields R_SL≈ 0, H-SLB dominates. H-RIV (Cross-evidence) Misfit rivers: valley scale can be inflated by glacial cycles, uplift, or river capture without requiring an event. If misfit vanishes after P20 controls, H-RIV dominates. H-DEC (Cross-evidence) Velocity differences: long-term vs present differences are due to plate reorganization or reference-frame differences. If TEST-DEC1 shows no global bias, H-DEC dominates. H-SYNC (Cross-evidence) Apparent simultaneity: “simultaneous” proxies can reflect age-model bias, selection bias, or common-mode climate response. If P29 fails permutation/controls, H-SYNC dominates. H-SHELF (Cross-evidence) Shelf asymmetry: explainable by margin type (passive/active) and long-term sedimentation/subsidence. If P25 finds no event-like truncation/megaflood deposits, H-SHELF dominates. H-OIL (Exploratory) Oil: distribution is dominated by basin/source rock/thermal maturity; “glacial bulldozing” is unnecessary. If P22 shows no association against the null model, H-OIL dominates. H-DISC (Observation) Oil distribution can be distorted by discovery probability (exploration maturity, cover/deltas, geophysical difficulty, accessibility). If the effect is large in P35, downgrade P22/P32 to HOLD; if bias is small or opposite-signed, H-DISC weakens. H-REF (Exploratory) Refugia: ice absence reflects climate (dryness/precipitation), elevation, and atmospheric circulation. If P23 shows no heat-flow association, H-REF dominates. H-LAKE (Exploratory) Endorheic lakes: outcomes of pluvial climate cycles. If P24 shows no age clustering, H-LAKE dominates. H-UNCON (Exploratory) Great Unconformity: long-term erosion and composites of multiple events. If global synchronization fails in P26, H-UNCON dominates. H-POLY (Exploratory) Polystrata: repeated local rapid-burial events; no global event is required. If P27 environments are locally restricted, H-POLY dominates. H-COAL (Exploratory) Coal: mainly in-situ peat-swamp origin; marine fossils arise from transgression/reworking. If P28 shows root/soil indicators are common, H-COAL dominates. H-GLAC (Confound) Normal glacial/postglacial processes: meltwater/climate/base-level changes explain oversized valleys/misfit. If effects vanish after glacial covariate controls in P20, H-GLAC dominates. H-CLIM (Confound) Climate-regime shifts: precipitation extremes/seasonality increased Q_peak. If effects vanish after including rainfall/flood proxies in P20, H-CLIM dominates. H-ANTH (Confound) Anthropogenic impacts: dams/channelization/land use modified recent forms. If effects vanish after removing anthropogenically affected samples in P20, H-ANTH dominates. H-RANK (Resource) Coal rank: temperature (burial) and time (thermal maturity) dominate; deformation is an indirect proxy. If burial proxies explain rank in P31, H-RANK dominates. H-BUDG (Resource) Oil budget: including basin-scale source/migration/leakage, a standard model can close within uncertainty. If P32 shows closure within uncertainty, H-BUDG dominates. H-CONF (Method) Cross-evidence: selection bias/multiple testing/omitted covariates can create spurious coherence. If P30 fails negative controls/permutation gates, H-CONF dominates.
Variant registry — locking “additional claims” as options
The core of this white paper (C1–C3) does not conclude an absolute chronology. However, if a reader wishes to make additional claims (e.g., “very young opening” or “global pulses”), those claims must be locked by mandatory gates, not by narrative plausibility. Since v1.20, time-scale/scope variants are separated as an explicit registry.
ID Variant (summary) Mandatory gates (enabled) FAIL/STOP rule (summary) V-BASE Baseline: episodic component + (optionally) subsequent gradual opening (incl. geologic time) P1–P6 + Ω-NoGo STOP if Ω is violated; if P4 FAIL then C3 FAIL V-REC Very young opening (kyr–0.1 Myr) claim variant V-BASE + P10+P11 mandatory (P12 recommended) If P10 shows core-based sediment age (or basement/crust age) exceeding config/p10_sed_prereg.yml:definitions.young_max_age_Ma, V-REC FAILV-PULSE Global pulse (volcanic synchrony centered) extension V-REC or independent + P12 mandatory (P9 recommended) If C_volc remains large after bias control (weak clustering), FAIL/HOLD V-SLIP Threshold slip emphasis (stick-slip / tail) V-BASE + P13+P14 If P13/P14 gate FAILs, V-SLIP FAIL/HOLD V-HOLO Holocene coupling (Great Drainage + freshwater shock) V-REC + P15+P16 If sea-level gates or multi-proxy agreement FAIL, V-HOLO FAIL V-COUPLED Pacific V2 coupling (open-system + global-pulse link) V-PULSE + P18 If TEST-ISO1 FAILs, V-COUPLED FAIL/HOLD V-HOLOX Cross-evidence package (sea-level buffering + misfit rivers + deceleration + shelf asymmetry) V-HOLO + P19+P20+P21 + P25 If any of P19/P20/P21/P25 FAILs, V-HOLOX FAIL/HOLD V-EVID Evidence-grade integration (V-HOLOX + Pacific coupling + event-window coherence) V-HOLOX + P12 + P18 + P29 If P18 or P29 FAILs, V-EVID FAIL/HOLD V-STRATA Stratigraphy–rapidness (unconformity/polystrata/coal mix) extension V-HOLOX + P26 (+P27/P28 recommended) If the core P26 signature FAILs, V-STRATA FAIL/HOLD V-RES Resource cross-evidence (oil/coal/heat/chronology) V-HOLOX + P22 + P31 + P32 (+P34 recommended) If P31 or P32 FAILs, V-RES FAIL/HOLD; if P34 is deep-time, V-REC-family claims are HOLD/FAIL
Operating rules (important).
- A Variant may FAIL without automatically failing C1–C3. Only the additional claim fails (e.g., “very young opening”).
- If a reader asserts a Variant while disabling its mandatory gates (P10/P11/P12, etc.), that Variant is immediately treated as HOLD/FAIL.
- The Variant choice and its mandatory gates are pre-registered via
config/constraints.ymlusingscenario.variant_id.
Evidence Readiness Level (ERL) — not “listing cases,” but “prereg + reproducibility + falsification”
In this document, “evidence grade” does not mean simply collecting many examples. ERL increases only when pre-registration + negative controls + a reproducibility bundle are satisfied.
ERL Requirements (summary) 0 Idea/narrative stage: data, thresholds, and alternative comparisons are not pinned (UNLOCK forbidden). 1 Single-module PASS: at least one prediction (P) passes pre-registered thresholds (alternative comparison may be limited). 2 Cross-module PASS: independent PASS across distinct data families (e.g., hydrology + kinematics + stratigraphy) and negative-control/randomization tests. 3 External replication: third-party replication of ERL-2 results using independent data and parameter lock (no post-hoc tuning).
V-EVID targets at least ERL-2. Its key strictness devices are P29 (event-window coherence) and P30 (negative controls/confounder controls). If P29 or P30 FAILs, “cross-evidence” is treated as parallel anecdotal listing and is downgraded to ERL-0/1.
ERL scoreboard (bundle basis; v1.31).
The table below summarizes whether each module is ready to be used as evidence in the current bundle. (Actual ERL-2/3 achievement is decided only after data are filled and PASS/FAIL is locked.)
Module Target ERL Readiness (summary) P19 2 prereg + code stub; requires residual/sign checks and negative controls (cross-family). P20 2 prereg + code stub; misfit + clustering with randomization/null design. P21 2 prereg + code stub (velocity comparison) + reconstruction/reference-frame controls. P18 1–2 prereg + code (open-system toy) + modern anchor casebook (seed). Before data fill, forbidden as event-window evidence. P22–P24 0–1 exploratory: UNLOCK forbidden without fixed data/controls definitions. P25–P28 0–1 stratigraphy/geomorph cross-signals: high confound risk → requires P30 controls. P29 2 prereg + reproducible engine (C18): range null (not value-permutation) + look-elsewhere + N_eff + jackknife; verdict HOLD pending prereg CSV. P30 1–2 prereg + code stub (control hard gate). Needs a module-level summary to reach ERL-2. P31 1–2 prereg + code stub + pilot example. Causal claims forbidden without burial-proxy controls and negative controls. P32 1 prereg + code stub (budget bounds). HOLD until uncertainty propagation and sensitivity analysis. P33 0–1 prereg + stub (thermal window/intrusions). Used as constraints/falsifiers rather than “evidence.” P34 2 prereg + stub (chronometer). If repeated deep-time agreement occurs, strong falsifier for V-REC/V-HOLO family. P35 1–2 prereg + code stub (discovery-bias regression). Must be executed before interpreting P22/P32.
Minimum sample-size checklist for ERL-2 promotion (recommended).
ERL-2 means the presence of pre-registered definitions, negative controls, and sample-size adequacy, not merely “interesting observations.” The table below provides recommended minimum sample sizes per module (internal project guideline; editable before release if needed).
Module Recommended minimum Notes (controls/filters) P10 N_core≥ 50 Ridge-distance stratification + Indian/South Atlantic negative controls. P11 N_nodule≥ 30 Low-sedimentation zones only; control for winnowing/benthic disturbance. P12 N_volc≥ 100 Fix “main cone/building phase” definition; separate hotspots vs arcs. P19 N_RSL≥ 30 Multiple correction scenarios (ice/geoid/uplift); residual sign checks. P20 N_river≥ 30 Climate–topography stratification + randomization null; Indian Ocean controls. P21 N_plate≥ 10 Compare MORVEL/NNR/ITRF frames; GNSS vs long-term. P22 N_basin≥ 20 Include non-glaciated giant fields (e.g., Middle East); requires P35 first. P24 N_lake≥ 20 Separate endorheic vs coastal; control evaporation/inflow. P31 N_coal≥ 50 Rank vs deformation proxies + burial/heat-flow (P33) controls.