#FDAS Lower-Tertiary pipeline — revision comparison (V30 → V50 → wed)

Purpose. Establish the worldenergydata ("wed") toolset as the canonical, go-forward truth for the Frontier-Deepwater Lower-Tertiary economics, by (1) laying the three revisions side by side, (2) tracing every result to the script that produces it, and (3) surfacing every script difference, missed file, and latent bug — so the FDAS authoring group (Roy, Chuck, et al.) and wed converge on one definition of the model.

Revisions and dates

Both source archives are backed up at /mnt/ace/worldenergydata/reference/fdas_revisions/{v30,v50}/ (zips + extracted).

#0. Executive summary

1. D&C days are provably version-independent. extract_drilling_completion_days.py is byte-identical across V30, V50 and wed (md5 1b89c23e) and reads only WAR data — never production. So the drilling+completion day counts carry across revisions unchanged; only the OGOR-A production layer moves. 2. The V30→V50 economic delta is (in wed's reproduction) purely +11 months of production. V30 window ends 2025-05; wed's V50 window ends 2026-04. Holding methodology and costs fixed, that adds +14.1 % oil, +15.2 % revenue, +$432.5 MM aggregate NPV — with no field crossing into positive NPV. 3. **But Roy's shipped V50 script is a *different* model than the published V50 numbers. generate_financial_summary_V50.py adds an after-tax block (severance + ad-valorem + corporate 21 % with NOL carryforward) and its lease_assumptions.xlsx silently changes 13 cost cells (§3.4). The published V50 figures were produced by wed's own reproducer** (pre-tax, window-only), *not* by that script. This is the one thing to align on before adoption (§4.3). 4. wed reproduces V30 to ±0.1 % oil / ±1 % NPV (a passed gate) — the basis for adopting wed as canonical (§4.1). 5. ~30 latent bugs were found by adversarial review of the four wed scripts (§6). Almost all are *masked by the current Lower-Tertiary dataset* (so they do not change any published number today), but they must be fixed before wed is trusted as a general tool. Two are structural and worth immediate attention. 6. Several files were never migrated V50→wed (the V50 generator scripts, the updated cost assumptions, and any V50 result workbook — which was never generated anywhere) (§5). 7. Roy's V50 script has now been run and reconciled field-by-field against wed (§4.4–§4.6). D&C days match to the day (Δ = 0, every field). Economics match closely — seven producers within −$103 M…+$197 M NPV (the 13 cost cells), three exploration-only fields exactly. His after-tax block is a small, real effect (severance & ad-valorem rates are 0; the 21 % corporate tax hits the three fields with positive *lifetime* cash flow — Jack St Malo −$103.4 M NPV, Julia −$4.4 M, Buckskin −$3.9 M). The one material discrepancy is Jack St Malo (−$3.1 B): ≈ $3.0 B of it is the NPV discount-reference convention (Roy "from Day 1" vs wed from first cashflow) plus ~$103 M corporate tax — not a cost or data difference. Roy's own script also corrects the article's Table-2 errata wed flagged (§4.6).

#1. The pipeline and its results lineage

Four stages turn raw public BSEE data into the article's tables. This is the chain to defend end-to-end:

 BSEE OGOR-A (monthly production)          BSEE WAR (daily drilling/completion)
        │                                          │
        ▼                                          ▼
 [1] ogora_to_chronological.py            [2] extract_drilling_completion_days.py
        │  chronological_lease_analysis.xlsx        │  drilling_and_completion_days.xlsx
        │  (per lease/well/month oil)               │  (per-well D&C days — VERSION-INDEPENDENT)
        └───────────────┬───────────────────────────┘
                        ▼
             [3] build_multi_year_lease_matrix1.py   (per-lease monthly matrix)
                        │
                        ▼
             [4] generate_financial_summary_*.py  +  lease_assumptions.xlsx  +  wti_monthly.xlsx
                        │
                        ▼
              financial_project_summary.xlsx
                        │
        ┌───────────────┴───────────────┐
        ▼                               ▼
  Article Table 1                 Article Table 2
  (producing fields:              (exploration-only fields:
   NPV, recovered oil,             Kaskida, North Platte, Tiber —
   revenue)                        D&C spend only, no production)

What moves V30→V50 and what does not: stages [2] (D&C) is byte-identical; the exploration-only Table-2 fields (no production) are identical V30==V50. All movement is in stage [1]'s window feeding stage [4]. That is why the reconciliation is clean.

#2. Three-revision comparison (master table)

Blank cells are shown as "-" where a revision does not carry that item.

#3. Script-level differences (V30 → V50), in pipeline order

#3.1 extract_drilling_completion_days.py — identical

Byte-for-byte identical across all three revisions (md5 1b89c23e). Reads WAR (mv_war_main, mv_war_boreholes_view, mv_war_main_prop_remark) + leases.xlsx; emits per-well DRILLING_DAYS and COMPLETION_DAYS. Confirms the D&C-days-are-version-independent premise. The golden workbook holds 219 well-rows over 20 leases / 12 developments.

#3.2 ogora_to_chronological.py → …_V50.py — rewrite (the core change)

The window is *not* pinned in code — it is whatever OGOR-A files sit in the working directory. V50's "+11 months" came from newer BSEE downloads, not a code constant. wed still ships the old V23_001 reader.

#3.3 generate_financial_summary_V30.py → …_V50.py

• V50 adds a full after-tax block: severance + ad-valorem (% of revenue) + corporate tax at 21 % with monthly NOL carryforward, and new tax columns in the output.
• V50 adds a flexible production join (maps DEV_NAME from leases.xlsx when absent).
• Discount rate (10 %), royalty (0.1875) and WTI base ($60/$75) are unchanged.
• wed did not adopt the V50 generator — it ships the V30 script unchanged.

#3.4 lease_assumptions.xlsx — 13 cost cells changed V30→V50 (verified cell-by-cell)

Discount rate, royalty, WTI base and tax rates are unchanged. These cost changes were never propagated into wed, and they contradict the "cost assumptions unchanged from V30" statement in the published V50 reports — see §4.3.

#3.5 build_multi_year_lease_matrix1.py — V30-only

Present in V30 and wed (also ported to packages/…/fdas/data/production.py); absent from the V50 archive. Carries the most severe latent bug (§6, matrix-1).

#4. Results connection — how the numbers reconcile (the persuasion core)

#4.1 wed reproduces V30 (the gate)

v30_repeatability_report.md: production matches golden_baseline_v30 within ±0.1 % for all 7 producers (Jack St Malo +0.03 %, others 0.00 %); NPV within ±1 %. The V30 financial_project_summary.xlsx NPV USD column ties 1:1 to the V30 column of v30_vs_v50_comparison.md. So the shipped V30 workbook is the source of record, and wed regenerates it.

#4.2 V50 = V30 + 11 months (window-only, in wed's reproduction)

Per-field NPV @10 % ($MM), V30 → V50:

#4.3 The reconciliation ask — one definition of "V50"

There are currently two V50s:

• wed's V50 (what the published reports show): V30 methodology, pre-tax, costs frozen at V30, only the data window extended. Reproduces V30 exactly and moves only with new data.
• Roy's shipped V50 script: after-tax (severance + ad-valorem + corporate 21 % + NOL) on top of the 13 changed cost cells in §3.4.

Update (2026-07-07): Roy's V50 script has now been run (§4.4–§4.6). It was executed exactly as shipped — his ogora_to_chronological_V50.py + extract_drilling_completion_days.py + generate_financial_summary_V50.py, his 26-lease leases.xlsx, his 13-changed-cost lease_assumptions.xlsx, after-tax on — against BSEE OGOR-A through 2026-04 and WAR through 2026-02. Two surprises fall straight out: (a) his after-tax block is small but real — the severance and ad-valorem rates are 0, but the 21 % corporate tax (with NOL carryforward) is levied on the three developments whose *lifetime* cash flow is positive (positive cumulative taxable income even though NPV is negative): Jack St Malo (corporate tax $1,030.6 M → −$103.4 M NPV), Julia (−$4.4 M) and Buckskin (−$3.9 M); the other eight fields have negative cumulative cash flow and pay none; and (b) his own V50 script already corrects the article's Table-2 errata we flagged (Stones ≠ Tiber, Cascade NCF sign, zero-OPEX) — §4.6. So the V50-vs-wed gaps are the 13 cost cells, the production window, the NPV discount-reference convention (dominant), and a modest corporate-tax effect (up to ~$103 M on JSM). *(Common mistake avoided: NPV-negative does not imply zero corporate tax — a field with front-loaded capital can post a negative NPV yet a positive lifetime taxable income.)*

*Faithfulness of the run:* OGOR-A came from wed's pickled .bin (Roy's original zips are gone) exported to his delimited format; WAR .bin→.txt; run in both his own WTI deck (flat-fills the last 9 months at $75) and wed's extended deck (thru 2026-05) as a sensitivity. His JSM facilities ($7,850 M) reconcile with wed's golden ($7,400 M), confirming the run is his model, not an artifact.

#4.4 Days reconciliation — V50 vs wed is exact (Δ = 0)

Because extract_drilling_completion_days.py is byte-identical across V30/V50/wed and all read the same WAR feed, Roy's V50 D&C output equals wed's to the day — the discrepancy the article team asked us to isolate is zero for every field, so no well-level drill-down is required (the drill-down triggers only on a discrepancy; there is none). The V30 column is the frozen 20-lease golden (older WAR vintage, no Buckskin); it differs from V50/wed only by newer-WAR recency and the Buckskin addition — lineage, not an open gap.

Drilling days (D):

Completion days (C):

Drilling + completion days (D&C):

V30→V50/wed grows +2,926 D&C days = Buckskin +2,056 (added via the KC ingest) + +870 on Big Foot/Jack St Malo/Shenandoah/Stones from the newer WAR vintage (2026-02 vs 2025-09; consistent with §3.1's post-cutoff activity). None of that is a V50-vs-wed discrepancy — those two are identical.

#4.5 Economics reconciliation — V50 vs wed, per column

Roy's V50 (his script, 2026-04, after-tax) beside wed canonical and the article. wed-latest = wed's committed latest-window component baseline at the same 2026-04 window (golden_baseline_v50.yml, plus the Buckskin V50-KC recompute because frozen V30 has no Buckskin row). wed-frozen is retained only in the NPV table as the audited V30 reference. Article = the published WO Table 2 (thru Nov-2025). "–" = no article value or no latest-window producing case.

Produced oil (MMbbl):

Oil is fully reconciled V50↔wed (both read the same OGOR-A to 2026-04); the article is lower because it stops at Nov-2025 (Shenandoah, still ramping, shows the biggest gap: 21.2 vs 8.7).

Revenue ($MM):

Revenue is now compared on the same 2026-04 window. The small residual differences are price-deck/rounding effects (and Big Foot's 0.5 MMbbl oil delta); Buckskin still uses the V50-KC recompute because there is no frozen V30 Buckskin row.

Royalty ($MM):

OPEX ($MM):

Net cash flow ($MM):

MIRR (annual):

OPEX = variable + fixed. Net cash flow is undiscounted. The V50 columns are Roy's after-tax script output; wed-latest remains the pre-tax V30-methodology reproduction unless and until the V50 tax/cost assumptions are formally adopted.

NPV @ 10% ($MM) — the headline, with the reason for each V50↔wed gap:

\* Article value is one of the known Table-2 errata — see §4.6.

Reading it: for the seven small-gap producers the V50↔wed NPV difference is only −$103 M to +$197 M, entirely attributable to the 13 cost-cell changes (Roy's V50 generally lowers host CAPEX, making his NPVs less negative). The three exploration-only fields match exactly. Jack St Malo is the sole material discrepancy, and it is not a cost or data difference at all — it is the NPV discount-reference convention, amplified by JSM's uniquely long lead time.

#4.6 Roy's V50 script confirms wed's article-errata findings

Running Roy's *own* newer script resolves the D1–D5 discrepancies wed flagged against the published article — his V50 output disagrees with his article exactly where wed said the article was wrong:

This is the strongest "same page" evidence in the pack: wed and Roy's latest script already agree that the published article carries these errors.

#5. Missed / non-migrated files (inventory)

Stale artifact: field_economics_cascade_chinook.md (un-suffixed) still uses first-oil 2014-01-01 / NPV −1,480.5 MM — it was not re-copied from its _v50 version after the first-oil correction. Regenerate.

#6. Bugs found (adversarial review of the four wed scripts)

Ranked by severity. "Affects current LT numbers?" answers the only question that matters for the published article: almost every finding is *latent* — masked by the current dataset — so no published figure changes today. They matter for making wed a general, canonical tool.

Full per-finding failure scenarios are in the session record; the actionable set (B1–B7) should become tracked issues before wed is declared the general canonical tool.

#7. Recommendation — wed as canonical

1. Adopt wed as the single toolset. It reproduces V30 to ±0.1 % and carries the extended window on one code path. 2. Settle the V50 definition with Roy (§4.3): pre-tax vs after-tax, and accept/reject the 13 cost-cell changes. Commit exactly one generator + one assumptions file. 3. Port the V50 ogora improvements (all-members read, de-dup, positional parsing) — these fix real latent risks (B2/B3) with no effect on current numbers. 4. Generate a V50 result workbook in Roy's financial_project_summary.xlsx schema so wed's V50 can be byte-validated (today there is none). 5. Fix the critical/high bugs B1–B7 with regression tests; regenerate the stale Cascade-Chinook report. 6. Verify the refreshed 2026 WTI values before any economics rely on them; decide gas-deck usage.

The D&C layer needs nothing — it is already identical and version-independent, and is the cleanest part of the story to put in front of Roy.

#8. Provenance

• Source archives: /mnt/ace/worldenergydata/reference/fdas_revisions/{v30,v50}/ (original zips + extracted).
• Revision dates confirmed against Gmail (Roy Shilling threads, Sept 2025 and June 2026) and file vintages.
• Every reconciliation number above is drawn from the shipped V30 workbook and the wed reports (v30_vs_v50_comparison.md, v30_repeatability_report.md, field_economics_*_v50.md); the 13-cell assumptions diff was recomputed cell-by-cell from the two lease_assumptions.xlsx files.