automated-log-analysis

Overview

Automated Log Analysis turns a vendor LAS 2.0 delivery into an auditable petrophysical evaluation. Every number comes from a bundled script: the parser resolves vendor mnemonics through an alias table, the evaluator computes Vshale, porosity, water saturation, net pay, pay zones, and OOIP with declared parameters, the plotter renders the composite log, and the report writer produces a Word document with a Markdown twin. The model explains the figures, decides how to proceed on warnings, and drafts the email; it never recomputes a value. When no email delivery exists, a deterministic synthetic well (SYN-LAS-1) exercises the whole pipeline with locked golden values.

Workflow

<Definition - Script execution recipe> Read references/script-interface.md first. For every script: read its source with file_read, then in run_python_with_write import it by file stem and call run(request); for example import run_evaluation as script; result = script.run(request). Always pass workspace_dir (the sandbox workspace directory Amazon Quick supplies) in the request. Use a unique script_name per execution. Do not read or assign __file__ or other dunder attributes, do not compile or exec source text, do not modify sys.path or os.environ, and do not import one bundled script from another. Treat the returned JSON as authoritative. </Definition - Script execution recipe>

<Definition - Canonical request> {"workspace_dir":"{workspace_dir}","well_id":"{well_id}","data_root":"automated-log-analysis-data"} plus the per-script fields listed in references/script-interface.md. well_id must match ^[A-Za-z0-9._-]+$; default SYN-LAS-1. All paths are relative to workspace_dir; the scripts reject absolute paths and parent traversal. </Definition - Canonical request>

<Definition - Default parameters> Starting defaults. Override only when the user specifies formation-specific values, and record every override in the report methodology.

<Definition - OOIP formula> OOIP (STB) = 7758 x area_acres x net_pay_ft x avg_phie x (1 - avg_sw) / boi. Recoverable = OOIP x rf. Computed only by run_evaluation.py; null when no water saturation exists. </Definition - OOIP formula>

<Definition - Synthetic golden values> SYN-LAS-1 from create_synthetic_las.py in create mode: 1,501 samples at 0.5 ft from 8,000.0 to 8,750.0 ft; vendor mnemonics GRMA, ROBB, TNPH, P40H, DTCO, HCAL resolve to GR, RHOB, NPHI, RT, DT, CALI. Evaluation with defaults: clean GR 35.0 API, shaley GR 120.0 API; net pay 325.0 ft of 750.5 ft gross, net-to-gross 0.433; three pay zones 8,100.0–8,199.5, 8,300.0–8,424.5, and 8,550.0–8,649.5 ft; average pay PHIE 0.2286 and Sw 0.1184; OOIP 195,437,364 STB and recoverable 68,403,077 STB; QC grade good with a CALI spike warning. no_resistivity mode omits P40H. </Definition - Synthetic golden values>

<Definition - Limited-accuracy result> When required_curves.missing contains RT, evaluation still succeeds: SW samples are null, net_pay_summary.sw_cutoff_applied is false, net_pay_summary.accuracy begins with limited, avg_sw_pay and every zone avg_sw are null, and ooip.ooip_stb is null with a note. Report net pay as porosity-and-shale screened only and recommend resistivity acquisition before volumetrics. </Definition - Limited-accuracy result>

<Definition - Report structure> Sections in order: title, well information table, methodology with parameters, cutoffs table, composite log image, net pay zones table, summary metrics, OOIP estimate, QC notes, recommendations. Summary metrics include a Net pay basis line built from net_pay_summary.accuracy; when resistivity is missing the report states porosity and shale cutoffs only with limited accuracy, renders Sw as n/a (no resistivity), and shows the ooip.note in place of a volume. The literal None never appears. Recommendations are supplied by the agent in the request only after the user has seen the evaluation; the script never invents them. </Definition - Report structure>

<Workflow - Email pipeline description="End-to-end evaluation from an Outlook LAS delivery to a draft results email." tools=[file_read, run_python_with_write, open_in_session_tab, get_current_time] triggers=["process daily drilling logs", "check for new MWD data", "run petrophysics on incoming email", "analyze vendor log delivery", "run log analysis pipeline"]

1. [Agent] Read README.md ## Pre-requisites and references/script-interface.md; confirm file_read and run_python_with_write are available and check whether the Outlook connector is available. Validate: Both required tools present; Outlook availability known. If fails: Stop on a missing required tool. If only Outlook is missing, say so and switch to <Workflow - Standalone LAS analysis>.
2. [Agent] Find unread emails in {{email_folder}} whose attachments end in .las or .dlis. Validate: At least one qualifying email. If fails: Report No new log deliveries found in {{email_folder}} and stop.
3. [Decide] If more than one qualifies, apply Rule 7. Validate: One delivery selected or all queued. If fails: Process the most recent.
4. [Agent] Read subject and body for well name, interval, and formation context. Save the first LAS attachment beneath {workspace_dir}/automated-log-analysis-data/ using a validated well_id as the file stem. Validate: A non-empty .las file exists at a relative path. If fails: For DLIS, warn per <Gotchas> and move to the next email; otherwise report the download error and stop.
5. [Agent] Continue with steps 2 through 7 of <Workflow - Standalone LAS analysis> using that relative path. Validate: Evaluation, plot, and report all returned status of success or created. If fails: Follow that workflow's failure paths.
6. [Agent] Draft, do not send, a reply to the vendor email using <Template - Email body>, attach the report (Word if created, otherwise Markdown), CC {{recipients}} when provided. Validate: Draft exists with the attachment. If fails: Create the draft without the attachment and include the full summary in the body.
7. [Agent] Print the console summary: well, interval, curve count and alias mapping, GR baselines, net pay and net-to-gross, zone count, OOIP and recoverable (or the limited-accuracy note), QC grade, warnings. Validate: Summary printed. If fails: Informational only.

</Workflow - Email pipeline>

<Workflow - Standalone LAS analysis description="Parse, evaluate, plot, and report a LAS file already in the workspace, or the synthetic well when none is supplied." tools=[file_read, run_python_with_write, open_in_session_tab] triggers=["run petrophysics on this LAS file", "analyze this LAS file", "evaluate this well log", "run the log analysis on the synthetic well"]

1. [Agent] Read README.md ## Pre-requisites and references/script-interface.md; confirm file_read and run_python_with_write. Determine the LAS source: a user-supplied relative path beneath workspace_dir, or none. Validate: Required tools present; source decided. If fails: Stop and name the missing tool.
2. [Decide] With no user LAS, read scripts/create_synthetic_las.py and run it with <Definition - Canonical request> and mode create (or no_resistivity when the user asks for the degraded case). Validate: status is created, sample_count is 1501, curves include GRMA, ROBB, TNPH. If fails: Report the script error and stop.
3. [Agent] Read references/curve_aliases.json, parse it, read scripts/run_las_parse.py, and run it with las_path and aliases. Validate: status is success; alias_mapping, depth_range, required_curves, and output_path present. If fails: Report the parser error message and stop.
4. [Decide] Inspect required_curves.missing and apply Rule 3. Validate: A continue, degraded-continue, ask, or stop decision is stated to the user. If fails: Stop.
5. [Agent] Read scripts/run_evaluation.py and run it with parsed_path and every parameter from <Definition - Default parameters>, substituting user overrides. Validate: status is success; gr_baselines, net_pay_summary, zones, ooip, qc_summary present; for SYN-LAS-1 defaults the values match <Definition - Synthetic golden values>. If fails: Report warnings and the error; stop.
6. [Agent] Read scripts/create_log_plot.py, run it with parsed_path, evaluation_path, and downsample; open the PNG with open_in_session_tab. Validate: status is created, bytes under 2,097,152. If fails: Note the plot failure and continue without the image.
7. [Agent] Present the evaluation, then ask the user for or propose recommendations; read scripts/create_report.py and run it with parsed_path, evaluation_path, plot_path, and the agreed recommendations; open the report. Validate: status is created; markdown_path present; docx_path present or a python-docx unavailable warning. If fails: Report the error and provide the summary in chat.
8. [Agent] Print the console summary as in the email pipeline. Validate: Summary printed. If fails: Informational only.

</Workflow - Standalone LAS analysis>

<Template - Email body> Subject: RE: <original subject> - Petrophysical Evaluation Results

<well_id> - Log Evaluation Summary Interval: <start> - <stop> <depth unit> Evaluated: <date from get_current_time>

KEY METRICS

• Net pay: <net_pay_ft> ft (N:G <net_to_gross>) <accuracy note when limited>
• Average porosity (pay): <avg_phie_pay>
• Average Sw (pay): <avg_sw_pay or "not available: no resistivity">
• OOIP: <ooip_stb> STB (recoverable <recoverable_stb> STB at RF <rf>) <or the ooip note>

PAY ZONES <one line per zone: id, top, base, thickness, avg PHIE, avg Sw>

DATA QUALITY: <overall_quality> <qc warnings>

RECOMMENDATIONS <agreed recommendations>

Parameters: Rw <rw>, m <m>, n <n>, matrix <matrix_density> g/cc. Cutoffs: PHIE >= <phie_min>, Sw <= <sw_max>, Vsh <= <vsh_max>. Full report attached. This is an automated evaluation; review before distribution. </Template - Email body>