Forecast integration in every page
Three models learned from Argentina's public registers, forecasting 36 months ahead for every well, block and operator.
Production decline
Learns how every producing lateral declines to forecast the next three years.
Activity pace
Reads frac crews and completion timing to project how many new wells come online.
Seasonal gas demand
Separates the winter swing in gas demand from real decline before forecasting.
How the forecasts are built
-
Every register, every night
Production, wells, frac jobs, concessions, holders and investment, read as they are published. The newest month is still arriving, so the models stop at the last month most wells have declared.
August 2026: 19,671 wells declared so far, against 23,520 for July.
-
One record per well
Each register files a well its own way. Their rows are tied to one wellhead by position and spud date, so a well's file, frac job and monthly output read as one.
85,752 register rows fold into 78,656 wells.
-
Every well, 36 months ahead
Gradient-boosted trees forecast each producing well month by month, from its own decline and its formation's curve. Young shale oil wells follow the newest vintages.
PLU.Nq.LCa-3292(h) peaked at 461 bbl/d in May 2025.
-
New wells at the operator's pace
Frac jobs reach the register months late, so the newest months look empty. Each operator's pace is read with its own delays corrected, and blocks, operators and the basin are forecast with it.
Pluspetrol: about 3.6 frac jobs a month.
-
Gas, season apart
Gas output follows the winter. The models learn with the season taken out and put it back after, so a summer trough never reads as decline.
La Calera's gas fell 21 % from August to December 2025, and held between 9.6 and 10.0 MMm³/d with the season out.
-
A low and a high, tested
Each model is retrained as if it were one, two and three years ago and scored on what was declared since. Those errors set every month's low and high.