auto-forecasting

Auto Forecasting

Use this skill when the task is to decide how production data should be forecasted before or after running decline fitting.

This is a limited-data production forecasting skill, not a full reservoir-engineering simulator. In normal Tauris workflows, required reservoir, PVT, completion, pressure-transient, and operating details may be incomplete or unavailable. State those limits explicitly and use them as QC reasons.

Workflow

1. Profile the dataset before fitting:
   • well/date columns
   • daily, weekly, monthly, sparse, or unknown cadence
   • usable oil, gas, and water signals
   • usable casing pressure, flowing tubing pressure, and pump inlet/intake pressure
   • candidate forecast origin dates, including first production and possible post-recompletion or post-stimulation starts
2. Classify method eligibility per well:
   • daily production with usable pressure can run pressure-aware residual diagnostics
   • daily production without pressure should use decline methods plus fit-window scanning
   • monthly production should prefer simpler reserves-style decline fitting
   • short or sparse histories should be yellow/red QC and routed to human review
3. Run candidate methods only after eligibility is clear.
4. Test fit-origin candidates before selecting a champion curve.
5. Score candidates with both numeric fit and engineering guardrails.
6. Return a recommendation with alternatives, QC status, and reasons.

Forecast Origin Guidance

Do not assume day 1 or month 1 is the correct forecast origin. A recompletion, stimulation, artificial-lift change, constraint removal, or cleanup period can make full-history fitting misleading.

When origin candidates exist:

• fit from first positive production as a baseline
• fit from each separated sustained uplift as a candidate
• use pressure step changes as context for fit-window selection
• mark QC yellow when the selected origin materially changes EUR or forecast shape
• ask for human review when origin choice dominates the forecast

Pressure Guidance

Use pressure as context first:

• rate down and pressure down: decline fit may be depletion-driven
• rate down and pressure flat/up: possible constraint, downtime, lift, choke, facility, or data issue
• rate up and pressure down: possible drawdown change or cleanup period
• volatile rate and volatile pressure: operationally unstable, human review likely needed

Do not let pressure-aware models run when pressure coverage is sparse, stale, or poorly aligned with production dates.

Pressure should influence projection choice this way:

• classify whether rate decline is depletion-supported or potentially operational
• choose fit windows and candidate forecast origins
• decide whether pressure-aware residual methods are eligible
• raise or lower QC confidence
• explain why a human should review a well

Do not use pressure to force a physics-based projection when bottomhole pressure, PVT, completion, reservoir, and operating details are unavailable.

Limited-Data Guardrail

When reservoir-engineering inputs are missing, do not invent:

• drainage area
• permeability
• skin
• fracture geometry
• flowing bottomhole pressure
• PVT properties
• material balance
• pressure-transient interpretation

Use production and surface/downhole pressure signals to write better method logic, fit-window choices, and QC explanations. Do not present the result as a physics-complete reservoir calculation.

Industry Alignment Checks

Use references/industry-alignment-checks.md when the user asks whether the workflow is directionally consistent with petroleum engineering practice. Do not claim formal compliance. The goal is to stay aligned with limited-data empirical DCA review by checking decline convention, forecast origin, pressure/rate consistency, method eligibility, history-window sensitivity, and visual sanity.

Decline Convention Export

Keep internal fitting math separate from commercial app entry values.

When exporting Arps parameters to ARIES, ComboCurve, PHDWin, Mosaic, or similar tools:

• pin the time basis to years before converting decline values
• do not export nominal Di directly when the app expects effective annual decline
• convert initial nominal Arps decline to effective secant annual decline:
  • exponential: 1 - exp(-Di_annual)
  • hyperbolic: 1 - (1 + b * Di_annual)^(-1 / b)
• convert terminal nominal exponential Dmin to effective annual exponential decline:
  • 1 - exp(-Dmin_annual)

Use convert_decline_convention in the forecasting MCP to show the numbers an engineer should type into commercial software.

MCP Tool

For CSV profiling and method recommendation, use:

areas/forecasting/mcp-servers/forecasting-mcp/forecasting_mcp.py

Primary tool:

profile_and_recommend(csv_path)

Use pressureProjectionDiagnostics and fitOriginCandidates from its output to decide which ArpsForecasting methods and fit windows should be applied later.

Alignment tool:

validate_industry_alignment(profile, recommendation)

Convention tool:

convert_decline_convention(nominal_di, b_factor, terminal_dmin, input_time_unit)

Chart Reference

Use areas/forecasting/assets/engineering_log_decline_plot_reference.png as the canonical final-plot style. Do not treat generated SVG samples as authoritative. A production chart should match the reference image's engineering log plot behavior, colored series conventions, dense grid, and history/forecast orientation.