← petro-agent · catálogoglm-5.2 + thinking · v14

Petrophysical Interpretation Report — 15-135-24,881-00-00


Well (UWI)	15-135-24,881-00-00
Log date	Sat Feb 28 00-46-47 2009
Service company	Log Tech
Acquisition date (header)	Sat Feb 28 00-46-47 2009
PLSS location	Sec 12 T20S R22W (map precision ±1.6 km)
Larionov variant	old_rocks (degraded)
Convergence status	DID_NOT_CONVERGE
Confidence tier	○ BRACKETED
Engine versions	calc_vsh 0.1.0 · calc_phie 0.1.0 · calc_sw 0.1.0 · formula_registry 0.1.0
Config hash (SHA-256)	`2a9cb78728e386ab…`
Git SHA	`7ded80e5dcb3`
Generated	autonomously, no human in the per-report loop

Confidence legend. Each result is tagged by parameter provenance: ● FIRM (core-calibrated) · ◐ QUALIFIED (offset-derived) · ○ BRACKETED (regional/global default — read as a range, dominant uncertainty stated). The system does not claim *always correct*; it states how well-supported each number is.

1. Executive summary

⚠️ ABSTENTION — this is NOT a confident estimate. The run did not converge to a defensible result; the numbers below are an uncalibrated engineering estimate, reported for transparency only:
- 1 unresolved MECHANICAL objection(s)

⚠️ ABSTENTION — this is NOT a confident estimate. The run did not converge to a defensible result; the numbers below are an uncalibrated engineering estimate, reported for transparency only. One unresolved mechanical objection remains.

Well 15-135-24,881-00-00 encountered a thick gross interval of 1198.2 m but a very lean net-to-gross of 0.042, meaning pay is thinly developed within a predominantly non-reservoir section. Within the net-pay population, average effective porosity of 0.181 and shale volume of 0.168 indicate relatively clean, porous intervals; water saturation averages 0.352, consistent with hydrocarbon-bearing rock where pay is present. Net pay is best expressed as a range: 18.3 m at the P10, 72.3 m at the P50, and 264.9 m at the P90, reflecting substantial uncertainty driven primarily by formation-water resistivity (Rw), which is an uncalibrated regional default with a swing of 69.5 m. This run did not fully converge and carries 1 unresolved mechanical objection, so these results are bracketed estimates rather than a confident point determination.

Net pay P10 / P50 / P90 = 18.3 / 72.3 / 264.9 m. Net pay is dominated by 'Rw', which is a regional DEFAULT (uncalibrated). This is the single largest uncertainty — the result is bracketed, not a confident point estimate.

Net pay P10 / P50 / P90 = 18.3 / 72.3 / 264.9 m.
Net pay is dominated by 'Rw', which is a regional DEFAULT (uncalibrated). This is the single largest uncertainty — the result is bracketed, not a confident point estimate.

2. Data inventory (from LAS)

Curves present (canonical ← raw): DCAL←DCAL, DRHO←RHOC, GR←GR, NPHI←CNLS, PHID_SVC←DPOR, RHOB←RHOB, RMED←RILM, RT←RILD, RXO←RLL3, SP←SP
Logged interval: 59.44–1382.57 m
Log date: Sat Feb 28 00-46-47 2009
Service company: Log Tech · Company: Berexco, Inc.
Field: Schaben
Not provided by LAS (out of scope): core, mud logs, pressure tests, production, completion, formation tops.

3. LAS quality control

Edit type	Curve	Detail
unit_conversion	NPHI	factor 0.01
hard_range_mask	RXO	count 1
hard_range_mask	RT	count 117
hard_range_mask	RMED	count 95
hard_range_mask	RHOB	count 1
hard_range_mask	DRHO	count 4
hard_range_mask	PHID_SVC	count 5
range_warn	GR	count 4
range_warn	RHOB	count 4
range_warn	NPHI	count 3
spike_removal	—	89 edits
degradation	—	3 edits

4. Standardization

Raw mnemonics mapped to canonical curve names:

Canonical	Raw mnemonic
DCAL	DCAL
DRHO	RHOC
GR	GR
NPHI	CNLS
PHID_SVC	DPOR
RHOB	RHOB
RMED	RILM
RT	RILD
RXO	RLL3
SP	SP

Unit conversions: NPHI×0.01.

5. Per-curve log QC

GR baseline: gr_p5=34.4, gr_p95=110.4
Bad-hole: GOOD=0.844, DEGRADED=0.127, EXCLUDED=0.029
Low-resistivity: rt_threshold=1.79, n_flagged=848, intervals=[[278.6, 282.2], [284.2, 339.1], [345.5, 346.7], [361.8, 362.4], [366.8, 367.1], [399.9, 401.7], [419.9, 421.8], [458.6, 458.7], [459.8, 467.4], [478.7, 480.2]]

6. Data preparation

Edits applied before compute (by type): degradation: 3, hard_range_mask: 6, range_warn: 3, spike_removal: 89, unit_conversion: 1.

7. Interval definition

Logged interval: 59.44–1382.57 m
Analysis window (analyst-restricted): 184.4–1382.6 m — gross and NTG are measured over this window
Gross evaluated interval: 1198.2 m
Computed net-pay runs (pre-merge): 68
Zonation is computed by depth (no formation tops in LAS).

8. Methodology

All numbers are produced by the deterministic, golden-tested engine. The LLM only selects methods/parameters and writes prose — it never computes a number. Figures are deterministic renderings of these computed numbers for the human reader; the agent reasons over the numeric EDA digest, not images (the models have no vision).

Step	Method (frozen)	Version
Vsh	Larionov 1969 (old rocks)	`vsh_larionov_old 0.1.0`
PHIE	Density-neutron crossplot, shale-corrected	`phie_density_neutron 0.1.0`
Sw	Simandoux 1963 (shaly sand)	`sw_simandoux 0.1.0`
Net pay	Vsh/PHIE/Sw cutoffs → net sand → net reservoir → net pay	`netpay 0.1.0`
Uncertainty	Monte Carlo P10/P50/P90 + parameter sensitivity	`mc 0.1.0`

9. Gamma-ray analysis

Clean baseline gr_min = 20.0 API · shale gr_max = 120.0 API
Shale index IGR = (GR − gr_min)/(gr_max − gr_min), the basis for Vsh.
GR baseline check: gr_p5=34.4, gr_p95=110.4

10. Resistivity analysis

Low-resistivity scan: rt_threshold=1.79, n_flagged=848, intervals=[[278.6, 282.2], [284.2, 339.1], [345.5, 346.7], [361.8, 362.4], [366.8, 367.1], [399.9, 401.7], [419.9, 421.8], [458.6, 458.7], [459.8, 467.4], [478.7, 480.2]].

Invasion profile (multi-depth resistivities, 8471 samples): fraction with shallow reading above deep (Rxo/RT > 1.2, invaded/permeable indication): 0.604; reversed (< 0.8): 0.098; median Rxo/RT 1.301; median Rmed/RT 0.961. Raw-curve facts; naming moveable hydrocarbon is the analyst's judgement.

11. Density-neutron crossplot lithology

Matrix point-shares (fraction of points near each matrix line): sandstone: 0.091, limestone: 0.219, dolomite: 0.247. Engine crossplot output; naming the dominant lithology is the analyst's judgement. Comparison with core/mud log: not available (LAS-only).

12. Water resistivity (Rw)

Rw = 0.0400 ohm-m · provenance: default
Sourced by the engine (Pickett / SP / default); never authored by the LLM.
Net-pay sensitivity to Rw: 69.5 m swing.

13. Figures

Composite log

Composite log

Pickett plot

Pickett plot

Neutron-density crossplot

Neutron-density crossplot

Buckles plot

Buckles plot

Hingle plot

Hingle plot

Distributions

Distributions

Sensitivity tornado

Sensitivity tornado

Net-pay Monte-Carlo distribution

Net-pay Monte-Carlo distribution

14. Shale volume (Vsh)

Mean Vsh by method (selection is the engine's; the LLM authors no number):

Method	Mean Vsh	Selected
vsh_larionov_old	0.364	✓
vsh_larionov_tertiary	0.277
vsh_linear	0.495
vsh_clavier	0.345
vsh_steiber	0.299
vsh_neutron_density	0.321
vsh_multimineral	0.280

15. Porosity

Mean porosity by method (effective where shale-corrected; the LLM authors no number):

Method	Mean porosity	Selected
phie_density_neutron	0.086	✓
phi_density	0.129
phi_neutron	0.203

Service-company porosity curves (cross-check evidence, full-column means): phid_svc: 0.197. Matrix assumptions differ per curve; agreement/disagreement is the analyst's reading.

16. Water saturation

Mean Sw (Simandoux 1963 (shaly sand)) = 0.662 (a=1.00, m=2.00, n=2.00, Rw=0.0305 ohm-m). Electrical parameters are engine-sourced; alternative Sw models are optional sections.

Mean Sw by method (engine-computed comparison; the LLM authors no number):

Method	Mean Sw	Selected
sw_archie	0.846
sw_simandoux	0.662	✓
sw_indonesia	0.652

17. Parameters and provenance

Parameter	Value	Unit	Provenance	Source
gr_min	20.000	API	default	—
gr_max	120.000	API	default	—
rho_ma	2.710	g/cc	default	Schlumberger 1989
rho_fl	1.000	g/cc	default	—
phie_max	0.450	v/v	default	—
phi_sh_d	0.100	v/v	default	—
phi_sh_n	0.350	v/v	default	—
a	1.000	-	default	Winsauer et al. 1952
m	2.000	-	default	Archie, G.E. 1942
n	2.000	-	default	Archie, G.E. 1942
Rw	0.040	ohm-m	default	Kansas Geological Survey 2000
rt_hydrocarbon_floor	5.000	ohm-m	default	—
vsh_cutoff	0.350	v/v	default	—
phie_cutoff	0.100	v/v	default	—
sw_cutoff	0.500	v/v	default	—
bit_size_config	7.875	in	default	—
qc_abort_threshold	0.800	-	default	—
circuit_breaker_n	3.000	-	default	—

The citations table (not RAG) gives each cited parameter exactly one frozen source. Parameters tagged default are regional/global — they drive the bracketed tier.

18. Petrophysical cutoffs

Boundary-inclusive criteria applied by the engine (netpay) to flag reservoir:

Stage	Criterion	Cutoff	Provenance
Net sand	Vsh ≤ cutoff	0.350	default
Net reservoir	+ PHIE ≥ cutoff	0.100	default
Net pay	+ Sw ≤ cutoff	0.500	default

Cutoff values are engine parameters (never LLM-authored); the dominant net-pay uncertainty is Rw — see Uncertainty for the swing.

19. Data quality and validator objections

Curve provenance (canonical ← raw mnemonic): DCAL←DCAL, DRHO←RHOC, GR←GR, NPHI←CNLS, PHID_SVC←DPOR, RHOB←RHOB, RMED←RILM, RT←RILD, RXO←RLL3, SP←SP.

QC edits applied before compute — degradation: 3, hard_range_mask: 6, range_warn: 3, spike_removal: 89, unit_conversion: 1.

Validator	Type	Detail
vsh_phie_anticorrelation	support	Vsh-PHIE Pearson 0.96 > 0.3 (dirty rock + high porosity)
rt_sw_consistency	mechanical	114 depths with Sw<0.4 but RT<5.0 ohm-m

20. Methodology (decision graph)

flowchart TD
    dec_1["decision: step: depth_quality"]
    act_1["tool_call: depth_quality"]
    dec_2["decision: step: compare_methods"]
    act_2["tool_call: compare_methods"]
    dec_3["decision: step: rw_evidence"]
    act_3["tool_call: rw_evidence"]
    dec_4["decision: step: compare_methods"]
    act_4["tool_call: compare_methods"]
    dec_5["decision: step: review_attempts"]
    act_5["tool_call: review_attempts"]
    dec_6["decision: step: mhi_scan"]
    act_6["tool_call: mhi_scan"]
    dec_7["decision: step: compute_vsh"]
    act_7["tool_call: compute_vsh"]
    dec_8["decision: step: compare_methods"]
    act_8["tool_call: compare_methods"]
    dec_9["decision: step: compute_phie"]
    act_9["tool_call: compute_phie"]
    dec_10["decision: step: compare_methods"]
    act_10["tool_call: compare_methods"]
    dec_11["decision: step: set_zone_of_interest"]
    act_11["tool_call: set_zone_of_interest"]
    dec_12["decision: step: compute_sw"]
    act_12["tool_call: compute_sw"]
    dec_13["decision: wasted no-op: compute_sw"]
    dec_14["decision: DETERMINISTIC RECLOSE (stale chain): apply_cutoffs"]
    act_13["tool_call: apply_cutoffs"]
    dec_15["decision: DETERMINISTIC RECLOSE (stale chain): run_uncertainty"]
    act_14["tool_call: run_uncertainty"]

21. Data gaps for calibration

This is a LAS-only study. Data that would calibrate the interpretation is absent:

Core (routine + SCAL): would anchor PHIE and Archie m, n, a.
Pressure tests (RFT/MDT): would give true fluid contacts and gradients (log-based).
Production/flow data: would test the net-pay flag against deliverability.
Dominant net-pay uncertainty driver (from the sensitivity): Rw (calibrated by produced-water salinity or a Pickett/SP Rw calibration).
Standard curves absent: DT, PEF, CALI.

22. Limitations

LAS-only study: no core calibration, no pressure tests (no true fluid contacts), no production, no mud logs, no formation tops (zonation computed by depth).
Standard curves absent in this well: DT, PEF, CALI.
Results are uncalibrated; accuracy depends on data not provided.

23. Conclusions

This well penetrates a thick section of 1198.2 m in which reservoir-quality rock is sparse, yielding a net-to-gross of 0.042. Where pay criteria are met, the rock is relatively clean (Vsh 0.168) and porous (PHIE 0.181), with water saturation of 0.352 suggesting hydrocarbon charge in those intervals. However, the run did not converge, 1 unresolved mechanical objection remains, and the dominant uncertainty — formation-water resistivity (Rw), sourced as an uncalibrated regional default with a 69.5 m swing — propagates into a wide net-pay bracket of 18.3 m (P10) to 264.9 m (P90). These numbers are bracketed estimates for transparency only and should not be treated as a confident point determination. No core, mud log, pressure, or production data were available from the LAS to calibrate or independently validate any parameter.

Appendix A — Ledger excerpt (traceability)

{
  "net_pay_total_m": 50.44440000000003,
  "net_pay_p10_p50_p90": [
    18.272760000000012,
    72.31380000000004,
    264.9016800000002
  ],
  "driving_params": {
    "a": 1.0,
    "m": 2.0,
    "n": 2.0,
    "Rw": 0.04
  },
  "claim_verifier": {
    "result": "PASS",
    "flags": [],
    "tone_flags": []
  }
}

Appendix B — Completeness gate

Item	Present
QC edits recorded before compute	✓
Every number ledger-traced	✓
Confidence tier on the run	✓
Parameter citations frozen	✓
Validator objections listed, not hidden	✓
Uncertainty propagated (Monte Carlo)	✓
Claim verifier run on prose	✓