Arps every well. Re-fit every shift.

Decline curve software fits a mathematical decline model (typically Arps exponential, hyperbolic, or harmonic) to each well's historical production rate, then projects future production. It is the foundation of reservoir engineering, reserves estimation, and short-term operational decisions like when to intervene on a well whose actual production has dropped below forecast. Most operators run decline analysis quarterly; the next-era discipline is continuous monitoring against forecast every shift.

Arps (1945) defined three forms. Exponential: q = qi × exp(-D×t). Hyperbolic: q = qi / (1 + b×D×t)^(1/b). Harmonic: q = qi / (1 + D×t). qi is initial rate, D is the nominal decline coefficient, b is the hyperbolic exponent (0 = exponential, 1 = harmonic, 0.4-0.6 typical for unconventional). Most onshore US horizontals are well-fit by hyperbolic with b around 0.5 for the first 12-18 months, then transitioning to exponential for terminal decline.

Quarterly review catches a well that has been underperforming for 90 days. Continuous monitoring catches the same well in days. The math is the same; the cadence is the difference. WorkSync runs the decline forecast nightly, checks every well's actual against the lower confidence band every shift, and flags the gap when it materializes. The actionable window goes from "the reservoir engineer's next slide deck" to "Tuesday's ranked work plan."

Traditional fitting picks qi, D, and b from historical data, fixes them, and projects forward. AI-assisted analysis (1) re-fits the curve continuously as new data lands, (2) detects regime changes (a well transitioning from hyperbolic to exponential), (3) overlays event-driven adjustments (workover, choke change, shut-in), and (4) attributes underperformance against the curve to mechanical (rod pump fillage), reservoir (pressure depletion), or facility (compression) categories. The reservoir engineer keeps the final call; the AI shortens the loop between data and decision.

The Arps forecast is the baseline that anomaly detection runs against. Per-well ML models learn the well's normal variability around the forecast curve, then flag deviations beyond expected noise. The two systems work together: Arps gives the long-term shape, anomaly detection catches the short-term break. See /capabilities/anomaly-detection for the equipment-class failure signatures the ML layer detects.

Two compounding wins. (1) Catch-time: when a well drops below forecast and gets flagged in days instead of a month, the intervention cost is a fraction of what it would be after a month of compounded deferred production. (2) Better reserve estimates: continuously calibrated decline curves produce more defensible reserves bookings and clearer 1P/2P/3P boundaries for the next investor deck.