← Back to Decision Stack

Watch AI process a borrowing-base packet.

A synthetic lender packet, built for inspection. See what the model extracted, what it missed, what the human fixed, and what was actually ready to send.

The packet we tested

Four source documents simulating a real lender redetermination request. Contains 4 planted issues for detection testing.

1. Reserve Report Summary (Meridian Energy Partners — Fictional)

Key Metrics

Metric Value Notes
PV-10 Total $847M Discounted at 10% discount rate
PDP Reserve Value $612M Proved Developed Producing
PDNP Reserve Value $178M Proved Developed Non-Producing
PUD Reserve Value $57M Proved Undeveloped

Production Profile

Category Volume (BOE/d) % of Total
Oil 29,104 68%
Natural Gas 13,696 32%
Total Current Production 42,800 BOE/d

⚠️ Summary Table (PDP Production)

The summary production table on page 2 states PDP production as 28,400 BOE/d.

⚠️ Detail Exhibit (PDP Production)

The detailed well-by-well exhibit on page 11 states PDP production as 27,200 BOE/d.

Decline & Price Assumptions

Parameter Value
Base Decline Rate (Annual) 22%
WTI Price Deck $72/bbl
Henry Hub Price Deck $3.25/mcf
Forecast Period 5 years explicit + terminal
2. Covenant Compliance Worksheet

Current state of financial covenants against facility limits (LTM ended 2025-12-31).

Covenant Limit Current Status Headroom
Leverage (Debt/EBITDAX) ≤ 3.50x 3.48x ⚠️ Borderline 0.57%
Current Ratio ≥ 1.00x 1.23x ✅ Pass 23%
Interest Coverage (EBITDAX/Interest) ≥ 2.50x 3.12x ✅ Pass 25%
Asset Coverage ≥ 1.50x 1.87x ✅ Pass 25%

Supporting Detail

Item Amount
Total Debt Outstanding $620M
RBL Commitment $750M
Availability $130M
Cash on Hand $12M
Net Debt $608M
EBITDAX (LTM) $178M
3. Prior Cycle Lender Q&A (Sample Questions)

8 questions from prior redetermination. Responses reference exhibits in the packet.

# Question Response
1 Provide updated engineering report. See Exhibit A-1 (Dec 2025 reserve estimate).
2 What is the current RBL availability? $130M based on current borrowing base of $750M.
3 Confirm status of Williston Basin divestiture. Asset sale expected Q2 2026. See management presentation (Exhibit B-7).
4 Provide sensitivity analysis on borrowing base at WTI $60 and $55. See attached Exhibit J-4.
5 Detail all commodity hedges in place for 2026-2027. Detailed in Exhibit C-2 (Hedging Schedule).
6 What is the ratio of hedged to unhedged production? ~65% of 2026 production hedged (see Exhibit C-2).
7 Confirm Q4 2025 financial statements are audited. Yes, audited by KPMG. See Exhibit D-1.
8 Are there any material litigation matters pending? None disclosed beyond standard regulatory matters.
🔴 Missing Reference
Q&A #4 response cites "Exhibit J-4" for price sensitivity analysis. This exhibit does not exist in the packet. The model will flag this as a data quality issue.
4. Hedging Schedule (2026–2027)

Commodity derivative positions currently in place. Updated as of 2026-01-31.

Commodity Period Volume Instrument Strike/Range Counterparty
WTI Crude Q1–Q4 2026 18,000 bbl/d Swap $71.50/bbl Goldman Sachs
WTI Crude Q1–Q4 2026 8,000 bbl/d Collar $65–$80/bbl JPMorgan
Henry Hub Gas Q1–Q4 2026 85,000 mmbtu/d Swap $3.10/mmbtu Citigroup
2027 Rollover Positions
WTI Crude Q1–Q2 2027 12,000 bbl/d Swap $68.00/bbl Goldman Sachs

Analysis Summary

2026 Oil Hedges: 26,000 bbl/d (Swaps + Collars)

2026 Forecast Oil Production: 42,800 × 68% = 29,104 bbl/d

Hedge Ratio (2026): 26,000 / 29,104 = 89% (aggressive)

⚠️ 2027 Over-Hedge Risk: With 22% annual decline, 2027 forecast oil = 29,104 × (1 – 0.22) = 22,701 bbl/d. Rolling 2026 hedges plus new 2027 position (12,000 bbl/d) could create over-hedge if not carefully managed.

What happened step by step

Seven-step walk-through showing extraction, conflict detection, analysis, and output quality. Timestamps are illustrative.

Step 1: What the model pulled out

Model extracts key fields from all four source documents. Quality signal via color coding.

Field Extracted Value Quality Signal
PV-10 Total $847M ✅ High confidence
PDP Reserve Value $612M ✅ High confidence
PDP Production Volume 28,400 vs 27,200 BOE/d 🔴 Conflict detected
Oil % of Total Production 68% ✅ High confidence
Base Decline Rate 22% annual ✅ High confidence
WTI Price Deck $72/bbl ✅ High confidence
Henry Hub Price Deck $3.25/mcf ✅ High confidence
Total Debt Outstanding $620M ✅ High confidence
RBL Commitment $750M ✅ High confidence
Cash on Hand $12M ✅ High confidence
EBITDAX (LTM) $178M ✅ High confidence
Leverage Ratio 3.48x ✅ Medium confidence*
Exhibit J-4 (Price Sensitivity) Not found 🔴 Missing
2026 Oil Hedges 26,000 bbl/d ✅ High confidence
2027 Oil Hedges 12,000 bbl/d ⚠️ Incomplete view

* Leverage = $608M net debt / $178M EBITDAX = 3.416x. Reported 3.48x suggests different EBITDAX adjustments applied by lender. Model flags this discrepancy.

Step 2: Where the packet contradicted itself

Model checks extracted fields against each other and identifies internal inconsistencies.

🔴 Conflict #1: PDP Volume Mismatch
Summary table (page 2) states PDP production = 28,400 BOE/d. Detail exhibit (page 11) states 27,200 BOE/d. Discrepancy: 1,200 BOE/d (4.2%).

Action: BLOCKER. Model flags for human review. Cannot auto-reconcile; requires confirmation from engineering team.
🔴 Conflict #2: Missing Exhibit Reference
Q&A response #4 cites "Exhibit J-4" for WTI sensitivity analysis ($60 and $55 scenarios). Exhibit J-4 does not exist in packet.

Action: BLOCKER. Model cannot validate lender's prior request satisfaction. Requires document location or creation.
⚠️ Concern #1: Potential Over-Hedge (2027)
2027 forecast oil production (after 22% decline) = 29,104 × (1 – 0.22) = 22,701 bbl/d. Model assumes worst-case rollover of all 2026 hedges (26,000 bbl/d from page) plus new 2027 swaps (12,000 bbl/d), resulting in gross hedged volume of 38,000 bbl/d against 22,701 bbl/d production.

Action: FLAG. Does not constitute a breach but indicates hedging strategy requires review. Lender hedge coverage policy likely permits 60–80% of production.
⚠️ Concern #2: Covenant Proximity
Leverage covenant at 3.48x with limit of 3.50x = 0.57% headroom. Covenant is not currently breached, but highly sensitive to commodity price changes. Model calculates: a $3/bbl decline in realized WTI from $72 to $69 would reduce EBITDAX by approximately $15M, pushing leverage to ~3.65x (breach).

Action: FLAG. Not a current covenant breach, but warrants disclosure to lender in "Covenant Sensitivity" section of memo.

Detection Rate: 4 issues identified (of which 2 are blockers, 2 are flags). Model successfully auto-detected 3 of 4 planted issues. One false positive: Model flagged a 200 mmbtu/d rounding variance in gas volumes as a conflict (resolved after human review).

Step 3: What the math said

Model recalculates leverage ratio independently and notes discrepancies with lender worksheet.

Leverage Ratio Calculation: ─────────────────────────────── Total Debt Outstanding: $620M Less: Cash on Hand: ($12M) ───────────────────────────── Net Debt: $608M EBITDAX (LTM): $178M Leverage = Net Debt / EBITDAX Leverage = $608M / $178M Leverage = 3.416x Covenant Limit: 3.50x Model Calculation: 3.416x Lender Worksheet Reports: 3.48x ─────────────────────────────── Discrepancy: +0.064x (1.9%)
Analysis Note
The 3.416x vs 3.48x gap: Model has visibility into total debt ($620M), cash ($12M), and reported EBITDAX ($178M). The lender's higher 3.48x ratio likely reflects EBITDAX adjustments that reduce the denominator (e.g., excluding certain non-recurring revenue items or applying stricter covenant definitions) or includes off-balance-sheet obligations the model doesn't see.

MEDIUM CONFIDENCE in the 3.416x figure pending confirmation of EBITDAX adjustments. Model flags the gap for human reconciliation.

Sensitivity Analysis (Model-Generated)

Scenario WTI Price Est. EBITDAX Impact Implied Leverage vs. Covenant
Base Case $72/bbl $178M (no change) 3.42x PASS (0.08x headroom)
$69 WTI $69/bbl –$15M (8% impact) 3.64x BREACH (+0.14x)
$65 WTI $65/bbl –$35M (20% impact) 4.08x BREACH (+0.58x)

Sensitivity assumes constant production and does not account for hedges. Actual impact would be muted by 89% of 2026 oil being hedged.

Step 4: The draft memo AI wrote

Model outputs draft memo with confidence markers. This is the raw AI output before human review.

MERIDIAN ENERGY PARTNERS
Covenant Compliance Summary – Q1 2026 Redetermination

HIGH CONFIDENCE Meridian Energy Partners remains in compliance with all financial covenants under its $750M revolving credit facility as of the Q1 2026 redetermination date. The company's leverage ratio stands at 3.48x debt-to-EBITDAX, below the covenant maximum of 3.50x. Interest coverage and asset coverage ratios provide adequate cushion at 3.12x and 1.87x, respectively. Current ratio of 1.23x reflects solid near-term liquidity.

MEDIUM CONFIDENCE Reserve values, as reported by the company's third-party reserve engineer, total $847M on a PV-10 basis (10% discount rate). This valuation reflects a commodity price deck of $72/bbl WTI and $3.25/mcf Henry Hub. Current production runs at 42,800 BOE/d with a base decline rate of 22% annually. The company has hedged approximately 89% of 2026 forecast oil production (26,000 bbl/d out of 29,104 bbl/d), providing downside protection at price levels of $65–$71.50/bbl.

FLAG Data Quality Issues Identified: A discrepancy exists between stated PDP production volumes in the summary table (28,400 BOE/d) and the detailed well-by-well exhibit (27,200 BOE/d). This 1,200 BOE/d variance (4.2%) requires reconciliation with the engineering team before final submission. Additionally, a reference to "Exhibit J-4" in the Q&A section (price sensitivity analysis) cannot be located in the current packet and must be resolved.

FLAG Covenant Sensitivity: While in compliance, the leverage covenant operates with minimal headroom (0.57%). A $3/bbl decline in realized WTI from the $72/bbl assumption would reduce EBITDAX by approximately $15M and push leverage to 3.64x, above the 3.50x limit. However, this scenario is partially hedged; approximately 89% of oil production is protected. Management should monitor commodity prices and consider additional hedging or deleveraging actions if WTI falls below $69/bbl.

Recommendation: Approve covenant compliance certification upon receipt of (1) engineering confirmation of correct PDP production figure and (2) location or creation of Exhibit J-4.

Quality Checklist

Element Status
Covenant calculations correct ✅ Yes (vs. lender report)
Reserve values cited correctly ✅ Yes
Hedge coverage calculated correctly ✅ Yes (89%)
Data quality issues flagged ✅ Yes (2 flagged)
Sensitivity analysis included ✅ Yes
Confidence levels disclosed ✅ Yes

Step 5: What the human had to fix

Treasury analyst (Raj) reviews draft and makes corrections. Shown as before/after diff.

Change #1: EBITDAX Adjustment

EBITDAX (LTM): $178M
EBITDAX (LTM): $184M ($178M reported + $6M non-cash hedge loss add-back per lender covenant definitions)

Reason: The model counted $6M in unrealized losses on open hedges as real cash impact. Lender's covenant definition excludes non-cash mark-to-market. Raj recalculates: true EBITDAX for covenant purposes is $184M ($178M reported + $6M adjustment), yielding leverage of 3.30x (not 3.48x).

Change #2: PDP Volume Resolution

A discrepancy exists between stated PDP production volumes in the summary table (28,400 BOE/d) and the detailed well-by-well exhibit (27,200 BOE/d).
Engineering team confirmed PDP production is 27,200 BOE/d. Summary table reflects stale Q3 2025 data and should be updated to 27,200 BOE/d for consistency.

Reason: Raj spoke with the engineering firm and resolved the conflict. The detail exhibit (27,200) is correct; summary table is simply out of date from a prior quarterly update.

Change #3: Hedge Coverage Language

The company has hedged approximately 89% of 2026 forecast oil production (26,000 bbl/d out of 29,104 bbl/d), providing downside protection at price levels of $65–$71.50/bbl.
The company has hedged approximately 89% of 2026 forecast oil production (26,000 bbl/d out of 29,104 bbl/d), providing downside protection across collars ($65–$80/bbl) and fixed swaps ($68–$71.50/bbl). 2027 rollover positions begin in Q1 2027 with 12,000 bbl/d of new swaps at $68/bbl.

Reason: Model was overly general. Raj adds specificity to show lender that hedging is properly layered and 2027 positioning is deliberate, not accidental.

Change #4: Strategic Context

Management should monitor commodity prices and consider additional hedging or deleveraging actions if WTI falls below $69/bbl.
Management should monitor commodity prices and consider additional hedging or deleveraging actions if WTI falls below $69/bbl. Additionally, the planned Williston Basin asset sale (targeted for Q2 2026, ~$75–$100M gross proceeds) is expected to improve leverage by 0.2–0.4x upon closing, providing additional covenant cushion.

Reason: Model had no visibility into the asset sale (mentioned in Q&A but not emphasized). Raj adds context that lender will appreciate: near-term deleveraging event reduces covenant risk.

Summary of Edits

Type Count Severity
Factual corrections 1 (EBITDAX) High
Data quality resolutions 1 (PDP volume) Medium
Precision improvements 2 (hedge language, strategic context) Medium

Step 6: What was actually ready to send

After human review and corrections, the clean final version ready for lender submission.

MERIDIAN ENERGY PARTNERS
Covenant Compliance Summary – Q1 2026 Redetermination
Prepared by: Treasury Department | Date: 2026-04-01

Meridian Energy Partners remains in compliance with all financial covenants under its $750M revolving credit facility as of the Q1 2026 redetermination date. The company's leverage ratio stands at 3.30x debt-to-EBITDAX, well below the covenant maximum of 3.50x, with 0.20x of headroom. Interest coverage and asset coverage ratios provide adequate headroom at 3.12x and 1.87x, respectively. Current ratio of 1.23x reflects solid near-term liquidity.

Reserve values, as reported by the company's third-party reserve engineer, total $847M on a PV-10 basis (10% discount rate). The valuation reflects a commodity price deck of $72/bbl WTI and $3.25/mcf Henry Hub, with current production at 42,800 BOE/d and a base decline rate of 22% annually. The company has hedged approximately 89% of 2026 forecast oil production (26,000 bbl/d out of 29,104 bbl/d), providing downside protection across collars ($65–$80/bbl) and fixed swaps ($68–$71.50/bbl). 2027 rollover positions begin in Q1 2027 with 12,000 bbl/d of new swaps at $68/bbl, representing disciplined hedge coverage.

The Williston Basin asset sale, targeted for Q2 2026 with expected gross proceeds of $75–$100M, is anticipated to further improve the leverage ratio by 0.2–0.4x upon closing, providing additional covenant cushion in the second half of 2026.

Engineering Note: PDP reserves are confirmed at 27,200 BOE/d. Production documentation was updated from an earlier quarterly revision and is now consistent across all exhibits.

Covenant Sensitivity: Should WTI fall below $69/bbl (from the $72/bbl assumption), unhedged exposure in gas production and the 11% unhedged oil position would begin to pressure EBITDAX. Management has appropriate visibility into price sensitivity and maintains sufficient liquidity ($130M availability) to manage any near-term working capital needs.

Conclusion: Meridian Energy Partners is well-positioned from a covenant and liquidity perspective. The leverage ratio includes reasonable cushion, commodity hedges are appropriately layered, and upcoming deleveraging events provide additional confidence in sustained compliance through 2026.

Document Metadata
Prepared by: AI Model (Draft) + Raj Singh, Treasury (Review & Edits)
Time to completion: 48 minutes total (45 seconds draft + 47.5 minutes human review)
Quality certification: Approved by Treasury Manager for lender submission

Step 7: Was it good enough to use?

Quantitative assessment of model performance on this task.

Extraction Accuracy

Metric Score Notes
Fields extracted correctly 34 / 38 (89%) 4 fields flagged for conflict or missing data
False positives (errors flagged incorrectly) 1 / 34 (3%) Rounding variance in gas volumes treated as conflict
Conflicts detected (of planted issues) 3 / 4 (75%) Caught PDP volume, missing exhibit, covenant sensitivity. Missed over-hedge in isolation (flagged only as sensitivity).

Draft Quality

Element Correct Without Edit Required Correction
Leverage ratio calculation ✅ Yes (3.416x) Variance to reported (3.48x) flagged; human confirmed EBITDAX adj.
Reserve value summaries ✅ Yes
Production forecast ✅ Yes (42,800 BOE/d)
Hedge coverage % ✅ Yes (89%)
Data quality flags ✅ Yes (2 flagged)
Covenant sensitivity ⚠️ Partial Model lacked context on asset sale; human added for lender confidence
Hedge specificity ⚠️ Partial Model was generic; human clarified collar vs. swap structure
Professional tone ✅ Yes

Efficiency Gains

Task Component AI Time Manual Equivalent Compression
Document review & extraction ~25 seconds 1.5 hours 217x faster
Conflict detection ~15 seconds 2 hours 480x faster
Initial draft memo ~5 seconds 2.5 hours 1800x faster
Total AI execution ~45 seconds 6–8 hours 480–640x faster
Human review & edits: ~47.5 minutes (iterative, not parallelizable)

Residual Human Judgment Required

Task Why Human Judgment Matters
EBITDAX adjustment methodology Requires understanding of lender's specific covenant definition (non-cash items treatment, working capital norms). Not deterministic; requires relationship knowledge.
Conflicting source resolution (PDP volume) Model cannot determine which source is authoritative without calling engineering team. Human makes judgment call based on context (timing of updates, prior conversations).
Strategic context (asset sale impact) Mentioned in Q&A but not emphasized in documents. Human knows ongoing discussions with management and realizes this is material to lender confidence. Model can only cite what's explicitly documented.
Hedge policy & tolerance Model detects 89% hedge ratio and flags. But human knows company's risk appetite, board preferences, and market conditions. Judgment over whether 89% is "aggressive" or "appropriate" requires context.
Tone & audience messaging Model produces competent prose. Human refines to match lender's expectations, relationship tenor, and prior communication style. Credibility depends on this calibration.

Overall Assessment

Summary
Use Case Match: This task is ideal for AI assistance. Extraction, conflict detection, and drafting are high-volume, rule-driven activities where AI excels and where speed matters operationally.

Quality Outcome: AI draft was 89% extraction-accurate and caught 3 of 4 planted issues. Final document quality improved significantly after human review but started from a strong foundation.

Efficiency Gain: ~480–640x faster for mechanical steps. Total solution (draft + review) compressed 7–9 hour process into ~48 minutes of wall-clock time.

Key Insight: Humans retained all judgment calls that mattered: EBITDAX methodology, data conflict resolution, strategic context, hedge policy, and tone. AI handled the "expansion" phase perfectly — preparing the working room for human decision-making rather than making decisions itself.

What this proves — and what it doesn't

This is a synthetic proof trace — a complete walk-through using a fictional company (Meridian Energy Partners) and realistic but invented documents. No real data, companies, or lenders are involved.

What the model can do: Extract structured fields from natural language documents, cross-check for internal consistency, perform mathematical calculations, identify missing exhibits, flag covenant sensitivity, and draft professional prose.

What the model cannot do: Make judgment calls about which conflicting source is "correct," apply company-specific EBITDAX methodologies without documentation, contextualize strategic events not explicitly mentioned in the packet, or determine if a hedge ratio is "appropriate" without knowing the company's risk policy.

The planted issues: Four intentional errors were embedded to test conflict detection. The model caught 3 of 4 in isolation:

False positives: The model incorrectly flagged a 200 mmbtu/d rounding variance in gas volumes as a conflict. Human review resolved this as immaterial.

Human review is essential not because the AI output was wrong, but because it lacked organizational context: EBITDAX adjustment rules specific to this lender, knowledge of which data sources are authoritative, awareness of upcoming events, and calibrated tone for the relationship. The final memo improved meaningfully after human editing.