83 hypotheses formulated, tested and documented
| ID | Status | Hypothesis | Finding |
|---|---|---|---|
| H1 | Discarded | 180-day seasonal bias in curtailment | Active pipeline, no identifiable bias |
| H3 | Confirmed | Aug/2023 blackout as curtailment inflection point | NE saturation ~13 GW, non-linear relationship. Post-blackout: ONS conservatism |
| H4 | Inverted | Forecast error causes curtailment (under-forecasting) | Over-forecasting causes 2.67x more curtailment than under-forecasting |
| H5 | Partial | DESSEM input diverges from actual output | Input identical to output. Conservatism is post-DESSEM, not in the model |
| H6 | Confirmed | Correlation between NE curtailment and SE CMO | Spearman 0.387, both high during the dry season |
| H7 | Confirmed | Physical NE-SE bottleneck limits power flow | Observed maximum 8.3 GW of 26 GW nominal (32%). Effective 4,800-5,200 MW |
| H8 | Refuted | SE thermal is a direct driver of NE curtailment | Correlation rho=-0.086. Thermal operates independently of curtailment |
| H9 | Partial | MMGD contributes to curtailment | Aggravating factor +35%/year, but not the main driver. ~400 MW invisible at peak |
| H10 | Partial | NE-SE interchange saturated at specific hours | Daytime saturation confirmed, nighttime partial |
| H11 | Partial | SE hydro inversely correlated with NE curtailment | rho=-0.512. ONS already reduces hydro when renewables are high, but insufficient |
| H12 | Refuted | Curtailment concentrated in few clusters | HHI dropped 41%, monotonic deconcentration. Problem is systemic |
| H13 | Partial | Double waste: simultaneous thermal + curtailment | 73-83% theoretically avoidable. Thermal ramp-down limited by must-run |
| H14 | Confirmed | NE Canyon Curve: negative net load 24h | H2 2025: ALL 24 hours negative. 125% penetration |
| H15 | Inverted | Solar forecasting worsening (invisible MMGD) | Solar improving, wind worsening. MMGD is +35%/year aggravating factor |
| H16 | Confirmed | Hydro spillage correlated with curtailment | 59.8 TWh spilled. But independent (rho=0.059) |
| H17 | Confirmed | Growing NE afternoon ramp | Ramp grows with solar. Duck curve deepening |
| H18 | Confirmed | 6 constraints cover 95% of curtailment | FNESE, Rio Novo do Sul, FPOTPPA, FNEN, Itabira, Jaguara |
| H20 | Confirmed | ENE grows faster than CNF and REL | ENE: 45% to 73-79%. ONS projection: 96% in 3 years |
| H21 | Confirmed | CNF/ENE classification follows hourly pattern | Switch at 11h BRT. 232 of 236 clusters switch simultaneously |
| H22 | Confirmed | ENE without traceability (generic description) | 100% of ENE records: 'SIN frequency control.' No specific constraint |
| H23 | Confirmed | CNF/ENE predictor based on hour and regime | 87.5% accuracy with 2 rules. Naive baseline: 54.7% |
| H24 | Partial | Law 15,269 compensation becomes residual | If ENE reaches 96%, REL+CNF compensation will be negligible |
| H25 | Confirmed | Renewable ROI degraded by curtailment | R$ 5.155B wasted in 2025. 662 plants revoked (27.8 GW) |
| H26 | Partial | Retroactive compensation quantifiable | R$ 1.766B calculated record by record. Semi-hourly CMO |
| H27 | Partial | LRCAP makes BESS economically viable | Payback 1.6 years with LRCAP vs 17.8 without. IRR 100% dependent on the auction |
| H28 | Confirmed | Financial impact of ENE vs compensable | R$ 3.8-4.1B non-compensable out of R$ 5.2B total in 2025 |
| H29 | Confirmed | Radialization as a curtailment driver | Rio Novo do Sul: 4,162 GWh CNF. Largest individual constraint |
| H30 | Partial | Campina Grande III as emerging constraint | Emerged Sep/2025, #1 in Feb/2026. Accelerating growth |
| H31 | Confirmed | NE-SE reverse flow occurs frequently | 1,399 periods (5.2% of total). Minimum -1,916 MW |
| H32 | Confirmed | Data centers can absorb NE surplus | 26.2 GW in requests. 7,298 MW approved in the 1st Redata batch |
| H33 | Confirmed | BESS optimal duration varies seasonally | 4h: Q1=38%, Q3=44%. 6h sweet spot year-round. 8h: +R$ 168M CapEx |
| H34 | Confirmed | NE load grew despite curtailment | +19% YoY. Not a lack of demand -- it is a grid bottleneck |
| H35 | Confirmed | BG load forecast outperforms ONS | Stacking beat ONS in 3/4 subsystems. NE -8.6% MAE |
| H36 | Refuted | NE thermal contradictory with curtailment | Not a contradiction: internal NE constraints force thermal for stability |
| H37 | Partial | MMGD should share curtailment responsibility | Agnes Costa (ANEEL) proposed it. +35%/year aggravating factor. Solar DG pays less than centralized |
| H38 | Confirmed | BA surpassed RN as #1 state | Jan/2026: BA 1,266 TWh vs RN 0.975. REL 190x increase |
| H39 | Partial | BA REL concentrated in FNESE | FNESE 70%, Rio Novo 15%, FPOTPPA 12%. SGI Janauba 3 = 85% of FNESE |
| H40 | Confirmed | Janauba compensator as dominant event | 471 GWh in 116 days. Return Apr/2026. R$ 115M opportunity cost |
| H41 | Partial | MG solar has its own bottleneck ecosystem | 3 independent bottlenecks: Itabira, Janauba-Jaiba, Rio Novo do Sul |
| H42 | Confirmed | FPOTPPA is the second NE-SE bottleneck | 2 circuits, 5,780 MW. Jan/2026: 75.5 GWh REL from poorly planned maintenance |
| H43 | Refuted | Concentrated curtailment = solvable by few interventions | HHI -41%. Monotonic deconcentration. Targeted solutions no longer sufficient |
| H44 | Partial | Wind VP correctable by ML | 50 per-cluster XGBoost. MAE -41.1%. Systematic and correctable |
| H45 | Partial | Solar VP correctable by ML | 146 plants. MAE -48.1%. Outperforms wind. Irradiance + hour = top features |
| H46 | Confirmed | NE load forecast BG outperforms ONS | -8.6% MAE in the NE. Stacking ECMWF+ETA40+GEFS50 |
| H47 | Confirmed | DESSEM models zero curtailment | 0 of 32,142 records. val_geracao = val_geracao_operada. Total blindness |
| H48 | Refuted | PDP captures all planned curtailment | PDP captures only 26%. 74% = real-time surprise |
| H49 | Refuted | DECOMP projects SE solar adequately | 2,190 MW projected vs 5,854 MW actual (+167%) |
| H50 | Refuted | NE thermal aligned with DECOMP | +65.5% above projected (1,223 vs 739 MW) |
| H51 | Partial | NEWAVE has adequate constraints for renewables | 57 constraints vs 168 in DESSEM. Renewables as exogenous input, not variable |
| H52 | Partial | Model cascade generates R$ 4.2B/year in surprise | 87% unplanned. Estimated cost of blindness: R$ 4.2B/year |
| H53 | Partial | Integrated correction would significantly reduce cost | VP -41/-48%, load forecast -8.6%, D+1 R2=0.70. Recoverable portion |
| H54 | Refuted | PDP gap useful for D+1 | Partial r=-0.07. Autocorrelation artifact. Degrades prediction |
| H56 | Partial | Curtailment by state has distinct dynamics | BA accelerating, RN saturating, CE decelerating, MG emerging |
| H57 | Partial | Precipitation correlated with hydro/curtailment | EAR r=-0.761. Precipitation drives hydro which influences curtailment |
| H58 | Partial | MMGD inference possible via energy balance | Hourly residual detectable. +5.9% ONS underestimate vs ANEEL |
| H59 | Partial | D+0 wind forecast MAE ~18% | MAE 701 MW. IPDO confirms -21.5% gap (Feb/2026) |
| H60 | Partial | DESSEM bus apportionment per plant | ~3,000 plant-bus relationships with apportionment 0-1 |
| H61 | Partial | GIBR group organizes constraints | 4 groups, ~15-20 relationships. Constraints grouped regionally |
| H62 | Partial | DECOMP over-dispatches hydro | South hydro +131.6% above plan |
| H63 | Partial | NE thermal coincident with curtailment | rho=+0.398. Must-run + internal NE constraints |
| H64 | Partial | Porto Sergipe I operates above optimum | R$ 300/MWh natural gas + R$ 844M coincident cost |
| H68 | Confirmed | DESSEM schedules zero NE wind curtailment | 32,142 rows, zero delta. Absolute confirmation |
| H69 | Partial | Solar forecast bias in DECOMP | +167% above projected. Systematic underestimate |
| H70 | Partial | NEWAVE-DECOMP transition loses fidelity | 57 to 168 constraints. Too much abstraction |
| H71 | Partial | BA REL explosion is specific | 190x increase Jan/2026. SGI events on FNESE and Rio Novo |
| H74 | Confirmed | FPOTPPA as second NE-SE bottleneck | 334-338 km, 5,780 MW. Poorly planned maintenance Jan/2026 |
| H75 | Partial | MG solar explosive growth | +469% YoY. Plants 110 to 195. 5.18 TWh in 2025 |
| H76 | Confirmed | MG solar has 3 independent bottlenecks | Itabira, Janauba-Jaiba, Rio Novo do Sul. 43.2% with attribution |
| H77 | Confirmed | Hourly curtailment profile changes seasonally | Q3 midday-peaked (night 14.7%), Q1 FLAT (night 26.4%) |
| H78 | Confirmed | Significant weekend curtailment bias | +44%/day. CNF 2.5x weekday, ENE 1.7x weekend |
| H79 | Confirmed | Curtailment transitioning to baseload | CV -27%, P10 27x, 24h curtailment 41% of days |
| H80 | Confirmed | D+1 curtailment predictable by persistence | Lag-1 ACF 0.75. Hybrid R2=0.70, MAE -14.9% |
| H81 | Refuted | MG solar is weather-invariant | r=0.957 vs generation. Identical to NE wind r=0.950 |
| H82 | Refuted | PDP gap useful for D+1 curtailment | Partial r=-0.07. Confound with autocorrelation |
| H83 | Refuted | Nighttime curtailment grows faster | Night share DROPPED 26.2% to 21.9%. Q3: 0% night-dominant |
| H84 | Confirmed | Curtailment deconcentrating (HHI dropping) | HHI -41%. 145 to 86. Top 5: 17.1% to 10.1% |
| H85 | Partial | NE-SE congestion cost quantifiable | R$ 1.77B. Average spread R$ 41, max R$ 1,154/MWh |
| H86 | Confirmed | BESS value-window differs from volume-window | 3h difference. Volume 07-10 BRT, value 12-15 BRT. 2.4x revenue |
| H87 | Refuted | 3-day momentum improves D+1 | r=0.28. Mean reversion, not momentum. Degrades prediction |
| H88 | Confirmed | Geographic divergence: 3 distinct fronts | NE saturating ~55%, BA accelerating +215%, MG emerging +434% |
| H89 | Partial | Wind-solar complementarity in portfolio | Daily r=+0.459. Intraday: wind night 16K MWh when solar=0 |
| H90 | Partial | Curtailment rate has saturation ceiling | Wind ~55-60%, solar ~45-50%. CE wind decelerating: +5.1pp |
| H91 | Explored | IPDO consistently over-forecasts | Single snapshot: NE wind -21.5%, solar +7.4%. Insufficient for time series |