The result, and the detail that was not in the headlines

[S] At the public session of the 4th cycle of the Oferta Permanente de Partilha on 7 October 2026, the ANP awarded seven of the thirteen production-sharing blocks on offer: Magnetita, Hematita and Azurita in Campos; Jade, Cruzeiro do Sul, Rubi and Rodocrosita in Santos. Signing bonuses — fixed in the bidding document, not bid — came to R$ 530,466,522.04, and committed exploration-phase investment to R$ 778,426,200.00. Both figures equal the sum of the seven rows in Quadro 8 of the edital, which we checked.

[S] The regulator's results page carries a bid sheet for each block. Every sheet lists one envelope. Six companies bid; all six won. Cerussita, Opala, Larimar, Granada, Aragonita and Turmalina got no bid at all.

[US] So the premium of 494.64% on Rodocrosita — the number in most headlines — was not won against anyone. Equinor (70%) and Galp (30%) wrote 15.52% against a floor of 2.61%, alone. A sealed single bid tells you what one company thinks the block is worth. It tells you nothing about what a competitor thought, because there wasn't one.

Two ways to average the same seven premiums

[S] The ANP's release gives an average premium of 108.38%. It does not say how that average is taken.

[US] There is more than one defensible way to average seven premiums, and here they give very different answers. Take the seven winning bids and the seven floors. Sum of bids over sum of floors, minus one: 108.38% — the ANP's number to the second decimal, so that is almost certainly how it is calculated, although the release does not say so. The simple mean of the seven per-block premiums is 178.26%. The gap comes from two low-floor blocks: Rodocrosita and Hematita, with floors of 2.61% and 1.73%, produce premiums of 495% and 315% on modest absolute bids, and they dominate a simple mean. Without them, the other five average 88%.

The lesson is older than this auction: a percentage of a percentage needs its formula stated. Our own pre-auction post quoted the 3rd cycle's "average premium of 91.20%" without one. We still don't know which average that is.

The correction: the annex has a neutral cell

[S] Item 7.5 of the edital says the bid must refer to Brent at US$ 50.00 and 10,000 bbl/d of oil per active producing well. ANEXO XX turns that into a grid, Brent bands down the side and productivity bands of 500 bbl/d across the top. In the row US$ 45.01–50.00, the cell 9,500–10,000 bbl/d does not hold a number. It holds the word "OFERTA" — the bid applies as written — and every other cell is a signed adjustment in percentage points.

[US] On 19 September we wrote that there was no neutral cell and that this one carried a small negative adjustment. That was wrong: on the rendered page the reference cell is highlighted and labelled, and we misread it. The neighbours are −1.57 pp (9,000–9,500) and +1.41 pp (10,000–10,500). The magnitudes we gave further out were right: in the same row the adjustment reaches +11.44 pp at 18,000–18,500 bbl/d and +11.65 pp at 18,500–19,000. At the bottom end it falls to −95.78 pp in the 0–500 band, and the low-price rows bottom out at a flat 1%. The 19 September post now carries a dated correction pointing here.

Why an engineer should care: well count is a fiscal variable

[S] In August 2026 the pre-salt produced 3,841,869.23 bbl/d of oil from 203 wells (ANP bulletin nº 192). [US] That works out to 18,925 bbl/d per well, up from 18,365 in July — both in the +11 pp region of the reference row.

[US] That is a national average across mature giants, not a forecast for any of these blocks — several of which may never produce. What it shows is the shape of the contract. The State's share does not depend on total field output. It depends on output per active producing well, averaged monthly (item 7.5) at the monthly Brent Dated average (7.6). Gas is shared at the same percentage (7.7).

That puts a fiscal number inside a technical decision. Two development concepts that deliver the same plateau — say, more wells at lower rate against fewer, longer-reach or higher-capacity wells — land in different columns of ANEXO XX. The well count, the completion design and the slot count on the manifold are no longer only reservoir and subsea choices; they move the profit-oil split. Under item 7.8, wells restricted for technical or operational reasons and booked as losses can be left out of the average, at PPSA's discretion. That discretion will matter the first time a field runs wells choked back for sand, water or flow-assurance limits.

If you work on the subsea production system — trees, manifolds and the wells they collect, the subject of our SPS episode — this is where the architecture shows up in the contract.

What the numbers do not say

The edital gives no water depth for any block, and neither do the results. Any depth you see attached to these names this week came from somewhere else. The results don't explain why six blocks drew no bid either. Cerussita had the highest floor of the thirteen (14.64%) and the largest fixed bonus (R$ 329.1 million); Opala had the second-largest bonus (R$ 249.1 million). Together the six unsold blocks carried R$ 704.5 million of bonus, more than the seven that sold. That is an observation from one cycle, not a cause.

What to watch

1. Winners' qualification documents due 14 October, adjudication by 18 November, per the published schedule. A single-bid block whose winner fails qualification has no runner-up.
2. Contract signature by 26 February 2027. We still have not confirmed that the 3rd cycle's contracts, due by May 2026, were signed.
3. Whether the six unsold blocks return in a later cycle, and at what floor.
4. The first development plan out of this cycle. When one of these blocks reaches a concept select, the productivity axis of ANEXO XX will be sitting in the well-count trade-off, whether or not the field-development team has been told.

The document, and when it actually appeared

[S] Lloyd's Register EMEA wrote Integrity of Connectors in Pipelines and Subsea Systems for Havtil, the Norwegian offshore regulator: report PRJ11100465363-1, revision 0.2, dated 17 November 2025, 55 pages. Havtil published it on 24 February 2026; LR's press release followed on 20 August 2026. [US] That is when the trade press picked it up — so this is not a September document. We are writing about it now because our episode on the subsea production system went out today and stops at the hub — and we have found no other public source that puts a failure rate on the thing that sits on the hub.

[S] Havtil's brief (Appendix C) asks about components "assessed as barrier elements and which, in the event of a single point of failure, can lead to weakening of the connection", under §11 (loads) and §12 (materials) of the Facilities Regulations. One sentence of it is worth keeping: "Traditionally, welded connections are favoured. For design, installation or operational reasons, mechanical connections are used." Welded joints are then excluded from the study (§1).

How many of these things exist

[S] Nobody counts connectors, so LR estimated them (§3): every pipeline, jumper and spool in the S&P Global Offshore Fields Database (3 October 2024) was assigned two connectors; shallower than 200 m counts as diver-assisted, deeper as diverless. Declared a lower bound — it ignores every connection inside a manifold or tree envelope — the estimate is more than 18,000 diver-assisted and more than 5,000 diverless connectors in service worldwide (§3.3). The Gulf of Mexico and South America hold most of the deep-water diverless experience; in the North Sea the UK has the most connectors and Norway the most deep-water diverless ones (§3.1–3.2). [US] Two per line is a floor, but the report says twice that these figures illustrate trends, not absolute numbers: 23,000 is "at least", not a count.

The number: connectors leak less than what they join

[S] §9 extracts OREDA@Cloud, Phases VI–XII, data version 3.7.4.3, subsea data 2000 to 2017, filtered to loss of containment only. Table 4: Flowline & Pipe, 1,587 units, mean failure rate 5.18 × 10⁻⁶; Connections (Flowline), 1,278 units, 1.04 × 10⁻⁶. Table 6: Manifold Piping, 737 units, 4.75 × 10⁻⁶; Manifold Process Valves, 3,266 units, 6.89 × 10⁻⁶; Connections (Manifold), 1,769 units, 1.35 × 10⁻⁶. The report calls these "approximately five times lower" than the flowline body, "approximately four times lower" than manifold piping and "approximately five times lower" than manifold valves. [US] Our division: 4.98, 3.52 and 5.10 — the "four" is a generous rounding of 3.5. And mind what a unit is here: a whole flowline against a single connection; even at two connections per line the pair comes in at 2.1 against 5.2.

[S] What the events were matters more than the rate. Flowline connections: 8 incidents in over 1,200 units — 4 material failure, 2 mechanical failure, 1 corrosion, 1 leakage (Table 5) — seven of them filed as external leakage and the corrosion case under "Other"; the text nonetheless states that all eight involved leakage. "The mechanical failures of connections were only associated with external leakage of utility fluids rather than process fluids" — hydraulic fluid, not hydrocarbons. "No catastrophic structural failures of the connections were reported as all failures involved leakages. In contrast, burst failure modes were found in flowline and pipe body data" (§9.1). Manifold connections: 10 incidents — 5 leakage, 2 looseness, 2 material failure, 1 breakage — and "regardless of the reported failure mechanism, the resulting failure mode is leakage" (§9.2). The report's explanation is mechanical, not statistical: correctly designed and installed, connections "typically possess greater thickness than the connecting pressure-containing parts and are more frequently manufactured from superior materials" (§9). Its caveat: root causes "are not known" (§9.3).

The unit label that does not survive arithmetic

[US] Tables 4 and 6 print "Mean Failure Rate /unit/year". Literally, a fleet of 1,278 connections at 1.04 × 10⁻⁶ per unit-year leaks once every 750 years; to accumulate the 8 events behind the table, OREDA would need about 7.7 million connector-years — impossible in a 2000–2017 window. OREDA's customary unit is failures per 10⁶ hours. Read that way, 8 events at 1.04 × 10⁻⁶ sit on the order of 10⁷ hours — some hundreds of connector-years, a surveillance population OREDA could plausibly hold. OREDA's mean is a multi-sample estimate, not events divided by exposure, so these are orders of magnitude, not exposures. Per year, the figures would be roughly 9 × 10⁻³ for flowline connections and 4.5 × 10⁻² for the pipe body.

We have not seen the OREDA page behind the extract, so this stays our inference — but per 10⁶ hours is the only reading that is physically possible. What does not depend on it is the ratio: connection and pipe are in the same unit, whichever it is, and five-to-one holds either way. Lifting "10⁻⁶ per year" into a QRA unchecked would be wrong by a factor of nearly ten thousand.

What the rate cannot tell you

[S] The data stops in 2017 and does not separate flanged, clamp-hub, collet and proprietary systems; §12 concedes that "detailed failure data for different connector types is not readily available in the public domain." What the report offers instead is qualitative. Collet connectors are "widely used for jumper spool connections and wellhead connectors" (§5, opening list — four families, five counting gripping adapters). Some proprietary diverless clamp connectors "may rely upon a single locking screw to maintain the clamping force on the hubs", and "these single point of failure elements may not be readily inspectable during the service life" (§7). An ILI tool runs through a connector and reads nothing of it: "it is not possible to inspect connectors in such detail due to high thickness of components and the inherent variation and complexity of geometries of proprietary connectors", so inspection is visual plus cathodic protection, concentrated on "non-redundant parts which may become a single point failure source, such as locking screws" (§11.3).

[US] That is the engineering shape of the finding: the connector is the thick, alloy, over-designed part of the system, and the part you can least inspect. Its integrity is bought in design, qualification and make-up. The report's framing (§10) is API RP 17N's technical risk categorisation, scored A to D per component and lifecycle stage, with API RP 17Q setting the qualification effort; the connector standards it leans on are API Spec 17D, ISO 13628-7 (which ISO has since withdrawn; the API equivalent is RP 17G), API 17TR7 and API RP 17R (Appendix B).

Where this touches the episode

[US] Today's episode ends the jumper at "the two hubs it connects" and gives it API RP 17R. What this report adds is what the episode could not say for lack of a source: on the only public extract of a connection failure rate we have found, the connection is not the weak link — the pipe body and the valves are — and every recorded failure was a leak, never a rupture. A jumper bent to absorb hub movement and locked with a connector nobody can later inspect is exactly the item where the qualification file is the integrity file.

What to watch

1. Whether Havtil turns this into supervision under §§11–12 of the Facilities Regulations; the report asks operators to keep design and qualification records for life extension (§11.4, NORSOK U-009, ISO 12747).
2. OREDA's post-2017 subsea phases — nearly a decade of newer connector designs is absent here.
3. A per-type split. If an operator or JIP publishes collet against clamp against proprietary, five-to-one becomes a design choice rather than a comfort.

The extent of what we read, stated before any conclusion

[S] One PDF, case 2025/983, published 24 September 2026, 72 pages. Body §1 to §11 on printed pages 4 to 63 — the table of contents puts §11.3 Granskingsgruppen on page 63, and that page, like VEDLEGG C, is redacted out of the published file; §12 lists exactly three attachments, A to C; the last printed number is 71 (VEDLEGG C, a heading with nothing under it — the published file is redacted, as its own filename says: skjult innhold). That is the end of the document, not the end of a screen.

[S] §9 «Avvik og forbedringspunkt» has exactly four items, §9.1 to §9.4, every one classified Avvik — non-conformity. No improvement point. We read past the boundary: printed page 57 closes §9.4, page 58 opens §10. Havtil's English news item agrees in one line — "four non-conformities within Equinor."

[S] Tabell 1 (§4, printed page 19) is one page, three columns (Tidspunkt · Hendelser · Involverte personer), five period rows, the last being June 2025; the caption bounds the axis at 17.6.2024 – 13.6.2025. [US] We did not add up the incidents in it, because [S] §9.3 says "it is challenging to state an exact number of incidents reported in the period June 2024 – June 2025."

The loop that ends in an atmospheric vent

[S] Hammerfest LNG takes the well stream from Snøhvit, Askeladd and Albatross through a 143 km pipeline from the subsea installations (§3.1). The MEG system exists so that "the multiphase pipeline from the production wells to shore is protected against hydrate formation at all times" (§3.5, quoting the plant's own system description). Rich MEG returns, is pre-treated to strip hydrocarbon gases and water, and goes to tank 13-TA-101 or, if salt-laden, 13-TA-102 before regeneration. The tanks are nitrogen-blanketed, and excess pressure is vented to atmosphere from a valve on top of the tank.

[US] That is the point worth carrying away: the last item in a subsea hydrate-control loop is an atmospheric vent on a storage tank, and what leaves it is whatever survives the pre-treatment upstream of it — which the incidents showed to be less settled than the design assumed.

[S] Summarising Equinor's own investigation, Havtil writes (§10) that "it was originally assumed that only nitrogen and water vapour would occur when venting from the MEG tanks under normal operation", and that the contents and stratification in the tanks proved more dynamic — quantity, frequency and composition of the vented gas varying far more than was previously known. [S] §9.3 adds the paper trail: a 2022 "safe location analysis for vent emission points" concluded that vent points pose no exposure risk to personnel during operation; Havtil finds this "does not necessarily hold", and that there was no plan to update it. [S] §6.3 names the document — E066-AB-S-RS-0028, dated 7 December 2022, itself an update of one belonging to the plant's construction, and followed by four further updates up to 16 October 2025 — and records what it concluded: that there was no basis to assume that the various vent points at Hammerfest LNG, "including the MEG tanks", could contribute to exposure risk for personnel "present on the gravel area".

[US] So the receptor was never the missing piece. Personnel standing on the gravel beside the tanks were inside the analysed case from the start, and the document was revisited four times after it was written. What the incidents falsified was not the layout. It was the conclusion.

Two gases, two opposite detection problems

[S] §3.3.2 prints the numbers. H₂S: odour threshold "as low as 0.008 ppm", most people smell it at 0.5 ppm, 8-hour limit 5 ppm, ceiling 10 ppm. The H₂S sensors in the personal 4-gas detectors had a sensitivity of 1 ppm, which "meant that the gas could be smelled at a level that did not register on the personal monitors." Benzene runs the other way: 8-hour limit 0.2 ppm today, 0.1 ppm from 2028, while "most people can smell benzene at levels above 1.4 ppm."

[S] And §4.2 carries Havtil's own assessment, not just the numbers: "most people will detect the smell of H₂S at levels lower than the measuring instruments used at HLNG can register."

[US] Those are opposite failure modes of the same barrier. For H₂S, at the odour threshold the nose is roughly two orders of magnitude more sensitive than the instrument — and even at the 0.5 ppm level where most people notice it, still twice as sensitive — so a crew can be right and the log clean. For benzene the nose is about seven times less sensitive than the limit value, so a crew can be comfortable and wrong. A procedure that treats smell and reading as one signal will be wrong about one of the two.

The escalation chain, and its latency

[S] §3.4: benzene is the risk indicator, on the premise that control of benzene implies control of the rest. Screening is by ION CUB (photoionisation, VOC only, cheap); confirmation by Dräger X-pid (photoionisation plus separation, benzene-specific, costly, needs training) — and the Områdeansvarlige, the area-responsible personnel, "were the only ones who had the necessary competence and who held the X-pid on the plant." (§7.8 calls the same role områdeoperatør; the report uses both terms.)

[S] §7.8: after June and July 2024, fire, entry and security guards — one-day course, contractor-employed — walked L201 in pairs with CUB meters. Above 0.3 ppm VOC they withdrew and called the area operator, an Equinor process operator, who measured and decided whether to clear the area. "In every incident where the BES guards got a reading on the VOC meter and contacted the area operator to have a benzene measurement carried out, the area operator could not confirm the presence of benzene" — except on 13 June 2025. In one case the gap between reading and confirmation was as much as 1½ to 2 hours.

[US] A screening-then-confirmation chain only works if the confirmation arrives faster than the phenomenon decays. Here the latency ran to hours while the events ran to minutes — [S] §4.1 records one run of odour episodes lasting "2 to 5 minutes at a time" (other incidents ran 30 and 40 minutes). A negative measured under those conditions is not evidence of absence, and it was used to clear areas. [S] §9.4 charges exactly that: no systematic assessment was made of the instrument's limitations, of environmental influence on it, of instrument fault or of delayed response, and no occupational-hygiene expertise was involved.

The mitigation that made each release larger

[S] §7.7: venting was originally automatic — frequent, short and small, because the valve is only partly open. After June 2024 the two rich-MEG tanks went to manual venting, prepared at 60 mbar overpressure, announced over the PA, area cleared by the guards. Manual venting overrides the automatic process control and "entails larger releases per venting, since the valve will be in the fully open position" — a change in consequence potential that §9.4 records as "not documented as having been systematically assessed or compensated". Twice it failed on its own terms: on 3 December 2024 the venting appears to have been set to auto by mistake (§4.2, §7.4 — the report hedges this one), and in April 2025 tank 13-TA-101 was again "by an error set to vent automatically over a longer period" (§9.4). On 17 March 2025 personnel worked 40 minutes on top of 13-TA-102 while the neighbouring tank vented toward them; they noticed the smell in the last three minutes (§4.3). No symptoms were reported from that incident.

One caveat for anyone checking this against the report: §3.5 and §7.7 contradict each other on automatic venting. §7.7, the dedicated section, says it vents more often but with less duration and volume because the valve is largely only partly open; §3.5 says automatic venting releases over a longer period. We follow §7.7.

[US] Swapping an automatic technical barrier for a procedural one is paid for somewhere. Here it was paid in a larger release per event, into an area whose occupancy depended on an announcement reaching the right people.

What the regulator charged, and what it did not establish

[S] The four hang on Styringsforskriften §§11–12, Arbeidsmiljøloven §§4-3 and 4-4, and Forskrift om utførelse av arbeid §3-1. §9.1 carries the sharpest line. A 2021 order arising from the Johan Castberg project (Ptil 2020/1704) led Equinor to undertake that dimensioning of project manning would be a mandatory measure when engaging suppliers new to the company, and that undertaking was carried into the Snøhvit Future PDO/PIO. Equinor then took part in selecting, for a large sub-project, a contractor "with limited experience of work on petroleum facilities in operation" — and no change was made to the project organisation's manning as a result. At peak the site held about 800 people, against roughly 600 in normal operation (§3.6).

[S] On outcome: more than 30 people reported odour and/or symptoms; about 300 worked in the areas across the incidents; symptoms were mostly mild and temporary, some people were sick-listed, one for six weeks; no permanent health effects have been identified, and it has not been possible to establish with certainty which substances or levels people were exposed to. [US] Note the wording: the Norwegian §1 says flatly that no permanent health damage has been demonstrated, while Havtil's English landing page adds a qualifier the Norwegian does not — "linked solely to chemical exposure". [S] Havtil attributes much of the severity of the outcome to the working environment around the project. Identifying the source was not an objective of the investigation (§11.1). Equinor's own level-3 report, received 18 November 2025, defines 15 individual incidents over June 2024 to August 2025 — a wider window than Havtil's — and makes MEG-tank venting the probable cause of 8 of the 15; the last of them, on 2 August 2025, is expressly outside Havtil's scope. [US] We have not read that report, and pass those two figures on only as Havtil states them.

[US] And for anyone quoting this case: Havtil's three surfaces disagree on the period investigated — report body and landing-page summary say June 2024 to June 2025, the landing-page standfirst says July 2025, the news item says April 2025. Cite the caption of Tabell 1: 17.6.2024 – 13.6.2025.

What to watch next

[S] Equinor was asked to report by 29 September 2026 on how the four non-conformities will be addressed — that deadline is stated on Havtil's English landing page, not in the report itself — the first place a technical measure on the MEG tank vents would appear. Three questions for that reply: is the 2022 vent-point study re-run with the new site layout as receptor; does venting return to automatic through a modification rather than a procedural barrier; and does the specific instrument end up where the screening alarm is, instead of an hour away?

[US] And one that generalises past Melkøya: if your rich MEG comes back from a long multiphase tieback, has anyone measured what actually leaves the tank vent — or is it still assumed to be nitrogen and water vapour?

On 13 August 2026 the UK regulator published its annual decommissioning update. Its own headline carried two numbers: £2.6 billion spent on UK Continental Shelf decommissioning in 2025 — a basin record, against £2.4 bn in 2024 and £2.0 bn in 2023, in-year prices — and £43.4 billion of estimated remaining cost, down from £43.6 billion a year earlier, each survey in its own year's prices.

Read straight: £2.6 billion spent, outstanding bill down £0.2 billion. The report's own waterfall says why — the forecast is revised upward at every cycle, by more than the year's spend in three of the four: +£3.6 bn against nothing, +£5.4 bn against −£2.0 bn, +£2.6 bn against −£2.4 bn, and last cycle +£2.4 bn against −£2.6 bn. Inflation is one of the causes named.

Our arithmetic on the published figures

This section is subtraction and division performed by us on numbers printed in Figures 15 and 16. The inputs are the regulator's; the differences and percentages are ours.

Figure 16 restates the same scope — the whole programme from 2023 onwards — at one price basis (2025 prices), once per annual stewardship survey. The total reads:

£44.0 bn (UKSS 2021) → £45.7 bn → £48.8 bn → £49.7 bn → £50.5 bn (UKSS 2025).

Over the same cycles, spend of £2.1 + £2.5 + £2.6 = £7.2 billion was recorded — Figure 5 prints the same three, labelled — and the remaining forecast moved £44.0 bn → £43.4 bn. The report's "approximately £7 billion" is the in-year sum.

So: £7.2 billion was spent, the remaining bill fell by £0.6 billion, and the programme grew by £6.5 billion — 14.8% — in constant money. Figure 15 shows £36.3 bn → £41.8 bn in 2021 prices, and it has to: both panels rebase one HM Treasury deflator series (30 June 2026 release), so the base year cancels and the 15.2% there is the same number after rounding. Real growth against economy-wide inflation — though a GDP deflator does not remove rig and heavy-lift rates, which the report names among the causes. Scope and rates, not the index.

The report's own caveat: "Comparison against historic NSTA cost estimates is on a full portfolio best estimate basis and not like-for-like per project."

Where the money actually sits

Figure 10 breaks the 2023-onwards programme by WBS, in 2025 prices — totals, with what is still to spend in brackets:

• Well P&A (subsea) — £11.1 bn (£9.8 bn remaining)
• Well P&A (platform) — £9.2 bn (£7.3 bn)
• Topside and substructure removal — £7.4 bn (£7.0 bn)
• Subsea infrastructure — £5.4 bn (£5.0 bn)
• the remaining four categories run from £4.3 bn down to £2.3 bn

Subsea well P&A is the largest single line item in the basin, and the least advanced: about £1.4 bn of its £11.1 bn has been spent in three years. Removals are almost entirely ahead: £7.0 bn of £7.4 bn on topsides and substructures, £5.0 bn of £5.4 bn subsea. Wells took over 50% of spend since 2023 and about 40% of what remains — the balance shifts toward steel, vessels and yards.

Mind the scope label: decom survey only. The eight categories sum to £40.0 bn remaining (our addition) against the £43.4 bn headline, which is total decommissioning and also carries onshore terminals, major pipelines and non-producing fields. Quoting one against the other is a way to be wrong by £3.4 bn.

The backlog is physical

About 3,000 wells remain to be fully plugged and abandoned. Figure 12a profiles them in seven five-year windows running to 2056–2060, and the cumulative Figure 12b ends at 1,989 platform, 1,030 subsea and 81 suspended E&A wells, 3,100 by our addition. The first window, 2026–2030, carries 902 platform, 404 subsea and 60 E&A wells.

Against that: in 2025 operators worked on 257 wells and 114 reached final AB3 status, up from 238 and 103 in 2024. Well abandonment cost £1,273 million in 2025, 49% of all decommissioning spend — the same share as 2024. The backlog awaiting final abandonment is "broadly unchanged" at about 500.

AB3 is a cutting job, not a barrier job

AB2 (derogated): in-well isolation complete, with derogation granted or pending to leave the well origin or the equipment above it, such as the conductor. AB3: in addition to the permanent barriers, "the well origin and all conductor equipment associated with the well origin have been removed."

That last step is not a pressure-containment operation. It is severance and lift at the seabed — and the regulator's stated aim is to "release valuable rig capacity for earlier-stage well abandonment work". The wellhead severance charter signed by the regulator and 17 operators in August 2026 is careful to stay neutral on the means, "including both vessels and rigs"; it "is not intended to have binding legal effect", and commits signatories to a workgroup, to data sharing and to early supply chain engagement.

The numbers for the prize: roughly 1,000 subsea wells left, £700,000 average per wellhead removal over five years on a sample of more than 100 wells, and a 30%, or about £200 million, reduction if vessels replace rigs — assuming every wellhead is removed and excluding inflation. Our check: 1,000 × £700,000 × 30% = £210 million, so £200 million is rounded down. It is also about 0.5% of the £43.4 bn (our division). The prize is not the money — it is the rig-weeks released for barrier work that only a rig can do.

The constraint is the fleet

Six semi-submersibles are marketed in the UKCS, down from a peak of eleven in 2023 — the NSTA's figure, sourced in its appendix to Westwood Global Energy, July 2026. The same units are wanted for production-adding interventions and for CCS drilling. The report also records a "marked increase in the cost of large (> 3kTe) removals of jackets and topsides", attributed to global demand for heavy lift vessels. Decommissioning is forecast to overtake capital expenditure from 2029, taking 24% of UK upstream spend in 2026–2031 against 13% in 2021–2025.

What we could not confirm

• Appendix 1, Figure 17 forecasts removal quantities, tonnage and kilometres for 2026–2037 in seven panels, with no data labels — we quote nothing from it.
• Figures 8a and 8b (operator concentration) carry no values in the PDF text layer. That one operator now accounts for 17% of the decade's expenditure and the top 10 for more than 80% comes from the body text; the report does not name them, and neither do we.
• The Benchmarking Report 2026 is an interactive application we did not read; its ">1,280 wells" scope and the cited saving of over £250,000 per well on an eight-well campaign are the regulator's.

What to watch

1. The rate gap. The 2026–2030 window holds 1,366 inactive wells (our sum of the three categories); 114 reached AB3 in 2025. Either the rate multiplies or the profile slides right.
2. Whether severance actually moves to vessels. A non-binding charter is worth one campaign: the test is a multi-operator vessel campaign with a published unit cost.
3. The next survey's "change in forecast" term. In three of four cycles it exceeded that year's spend; in the fourth it fell £0.2 bn short — and that £0.2 bn is the whole fall in the headline.
4. CCS-driven P&A scope. Under net zero, reuse and repurposing, the risk register flags wells that may need decommissioning to a higher standard to enable CCS — "additional intermediate barriers and or intersect wells". A different well design, decided long before the rig arrives.

What happens on 7 October

Brazil holds the public bidding session of the 4th cycle of the Oferta Permanente de Partilha de Produção — the production-sharing permanent offer — on 7 October 2026. Thirteen exploratory blocks are on the table: five in the Campos basin (Magnetita, Turmalina, Hematita, Larimar, Azurita) and eight in Santos (Jade, Aragonita, Cruzeiro do Sul, Opala, Granada, Cerussita, Rubi, Rodocrosita).

Adding up the per-block figures in the regulator's own bidding document gives 25,717.531 km² of acreage, R$ 1,235,012,266.54 in signing bonuses and R$ 1,414,648,800.00 of committed minimum investment. Those three totals are our addition of the thirteen rows; every row is the ANP's.

A separate concession cycle — the 6th OPC, with 24 areas across nine basins — runs alongside it. The two are different regimes and get confused constantly. Concession and production sharing are not variants of one auction.

One variable

In a concession round you can win on signature bonus, on work programme, on local content. Here you cannot. The signing bonus is fixed per block, printed in the annex. The minimum exploratory programme is fixed, expressed in work units — 266 UT for most of these blocks, over a seven-year exploration phase. Local content is fixed.

The only number a bidder writes down is the percentage of profit oil offered to the State, to two decimal places. Highest percentage wins.

Each block carries a floor. Across the thirteen, that floor runs from 1.73% (Hematita) to 14.64% (Cerussita) — a factor of about 8.5 between blocks in the same auction, in two adjacent basins. The floor is not a judgement about generosity; it is the regulator pricing geological risk block by block before anyone bids.

The part that does not travel

Here is the clause that decides what those percentages mean, and it is almost never quoted. Item 7.5 of the bidding document states that the percentage offered by bidders must refer to a Brent price of US$ 50.00 per barrel and to an average daily production of 10,000 barrels of oil per active producing well.

The winning percentage is therefore not a share. It is a coordinate. ANEXO XX of the same document is a grid: rows are Brent price bands, from the lowest up to above US$ 200.01; columns are bands of average production per active producing well. Each cell carries an adjustment in percentage points to be applied to the percentage that won the auction. One cell is neutral: at Brent US$ 45.01–50.00 and 9,500–10,000 bbl/d the annex prints the word "OFERTA" — the bid applies as written. Its neighbours already move it: −1.57 pp at 9,000–9,500 bbl/d, +1.41 pp at 10,000–10,500 bbl/d, and +1.02 pp in the same productivity band one price row up, at Brent US$ 50.01–55.00. Wherever else the field actually lands, the State's share moves. [Corrected 12 October 2026. This passage previously read: "There is no neutral cell. […] the cell at Brent US$ 45.01–50.00 and 9,500–10,000 bbl/d already carries a small negative adjustment, and the band above the reference price carries a small positive one."]

Two features of that grid matter to anyone reading auction results.

The adjustments are large. They are quoted in percentage points to four decimals, and in the low-productivity bands they run to tens of points negative — enough to drive the State's share to the 1% floor the table applies in its top-left corner. A bid of 14% into a field that underperforms is not 14%.

The productivity axis does not stop at 10,000 bbl/d per well. It carries on, in 500 bbl/d bands, across five consecutive pages of the annex, to a last column of 50,000–50,500 bbl/d. And in July 2026 the Brazilian pre-salt produced 3,728,074 bbl/d from 203 producing wells — an average of 18,365 bbl/d per well, which is division of two figures published in the same ANP bulletin. That average lands well inside the table, not off the end of it.

Read along the reference price row — Brent US$ 45.01–50.00 — the adjustment is zero in the 9,500–10,000 bbl/d reference cell and negative below it, and by the bands around 18,000 bbl/d it has risen to roughly +11 percentage points. So productivity does not leave the State's share where the bid put it: at pre-salt performance it lifts it substantially. The winning percentage is neither a floor nor a ceiling — it is the reference the table moves from, downward when the field disappoints and upward when it does not.

We read the annex directly in the bidding document's PDF. We quote the magnitude and not the four-decimal cell values, because aligning a row and a column by hand across a five-page table is exactly the kind of reading that can be off by one of each. [Corrected 12 October 2026: the sentence before this note previously said "the adjustment near the 10,000 bbl/d reference is negative".]

That two-way movement is the design. The mechanism protects the bidder against a disappointing reservoir, and hands the State the upside on both axes — when crude is expensive and when the reservoir is exceptional.

Three more numbers with engineering consequences

Cost recovery is capped at 80%. Item 7.9: each month the contractor takes the cost oil parcel up to 80% of the gross production value of each block, with the excess carried forward to later years. That is a materially higher ceiling than the one documented for Libra, the first sharing contract, where the contract manager PPSA has described a limit of 50% of monthly production value, falling to 30% once past costs are recovered. We have not re-verified the Libra figures against PPSA this week; the 80% is the current edital's, read directly. If you carry "50/30" in your head as the pre-salt cost recovery rule, it is not the rule being auctioned in October.

Local content is highest on the subsea scope. Quadro 10 sets the minimums identically for every block: 30% in the exploration phase, and in the development stage 30% for well construction, 40% for the gathering and flow system, and 25% for the stationary production unit. The gathering and flow system — flowlines, risers, umbilicals, the manifolds and jumpers that connect them — carries the highest minimum of the three development macro-groups. The hull can be built abroad more freely than the SURF scope can.

Petrobras took no preference. Quadro 9 lists the statement of the operator preference right for all blocks in the permanent offer inventory, and the answer is "Não" for every one, with 100% of each block offered to the market. Since 2016 the company has a right of first refusal on sharing blocks, not an obligation to operate; on this inventory it declined across the board.

What is not in the document

The block table gives area in km². It does not give water depth. Bathymetry sits in the paid technical data package and in the general map, not in the parameters annex — so any water-depth figure you see attached to these blocks in coverage this month did not come from the bidding document.

And a small warning for anyone reading the edital themselves: items 7.4 and 7.20 both point to column (xiii) of Quadro 8 for the minimum percentage, while in the table's own header (xiii) is the exploration phase in years and (xiv) is the minimum profit oil. Read the column heading, not the cross-reference.

What to watch

1. The spread of the premiums on 7 October. The 3rd cycle, in October 2025, closed with an average premium of 91.20% over the minimum percentages. If that repeats against floors as low as 1.73%, the headline percentages will look small and mean more than they look.
2. Whether the low-floor blocks attract a bid at all. Hematita at 1.73% and Rodocrosita at 2.61% are priced as frontier risk. Empty blocks are as informative as contested ones.
3. Signature by 26 February 2027, with the signing bonus due by 30 December 2026. The 3rd cycle's contracts had a May 2026 deadline; confirming that those were actually signed is still an open item for us.
4. Whether anyone quotes the reference conditions. When the results are reported as "X% to the State", the number will be a coordinate at US$ 50 Brent and 10,000 bbl/d per well. It will be reported as a share. It is not one.

On 18 September 2026 the Greensand project began commercial CO₂ storage in the Danish sector of the North Sea. The operator's figure for this first commercial phase is up to 400,000 tonnes of CO₂ per year, injected into the Nini West reservoir — a depleted oil field described as approximately 1,800 metres beneath the seabed.

There is no pipeline. CO₂ is captured onshore, liquefied, trucked to a terminal at Port Esbjerg, loaded onto a single purpose-built carrier and sailed out to a platform. That is the whole chain, and it is the reason this project is worth an engineer's attention. The reservoir is ordinary. The delivery is not. The CO₂ itself is biogenic — the European Commission's opinion on the draft permit records it as coming from biogenic sources, with a CO₂ content above 95 % and no water, and therefore no expected SOx or NOx.

The chain, as published

Four numbers define it, all from the project or the shipbuilder:

• the terminal at Port Esbjerg holds 6,000 t and is rated for 60 to 72 trucks per day;
• Carbon Destroyer 1 carries about 5,300 t per cargo;
• the round trip from Esbjerg to the field and back takes about three days;
• the builder puts the vessel's annual transport capacity at about 600,000 t.

The arithmetic nobody publishes

The following is our arithmetic on those published figures, not a claim by anyone involved.

Four hundred thousand tonnes a year at 5,300 t a cargo is about 75 cargoes, one every 4.8 days. A three-day round trip means one ship can in principle make roughly 122 trips a year, or about 645,000 t — which lands close enough to the builder's stated 600,000 t to say the two figures describe the same vessel doing the same job. So the first phase runs this ship at roughly two-thirds of its own annual capability. The slack in the system is in the ship, not in the store.

Now the buffer. Steady-state inflow at 400,000 t/y is about 1,096 t/day. The terminal holds 6,000 t — about 5.5 days of that inflow, or a little over one shipload. Put the same numbers the other way: 1,096 t/day arriving in 60 to 72 trucks would mean 15 to 18 t a load, which is under what a CO₂ road tanker carries — so the terminal's truck rating is not the binding constraint in this phase. It has headroom. The ship does not.

That 5.5-day figure is the one to sit with. A North Sea weather window that keeps a DP2 vessel off station for a week fills the terminal. When the terminal is full, the trucks stop; when the trucks stop, the capture plants upstream have nowhere to send product. In a pipeline-fed system the storage site absorbs that. Here the buffer is one tank farm and one hull.

What the well sees

The project states that the injectors are existing wells, originally drilled as oil producers, reconfigured by running wire about 1,800 m down to install new equipment and barriers — work done from the platform, with crew arriving by helicopter or by walk-to-work gangway. It does not say how many wells.

Two things follow, and both are our reading rather than anyone's statement.

First, re-used producers carry their history. The reservoir here is the Frigg Member — twelve to thirty metres of porous, glauconitic sandstone on the western flank of the Nini salt structure at about 1,700 to 1,800 m TVDSS, sealed below by the Balder Formation, as the Danish regulator's own permit describes it. A well completed for oil in that sand was not specified for decades of wet CO₂ against carbon steel. Legacy casing and legacy cement become the barrier envelope for a new fluid, and the wireline-set equipment is the visible part of a qualification argument that is mostly about what was already downhole. Worth noting that the permit counts three legacy wells in the storage complex — the Nini-4 exploration well, the former oil producer NA-03 and the field's water injection well NA-05 — for five reservoir penetrations in total, and that the 2022 pilot injected through NA-05. "Existing oil producer" is the project's description of the commercial injectors, not of every well in the complex.

Second, and more particular to this project: ship-fed injection is batch injection unless something smooths it. Seventy-five cargoes a year is seventy-five arrivals. Whether that translates into seventy-five start-stop cycles at the wellhead depends on whether there is buffer storage on the platform between cargoes — and that is precisely what none of the published material says. It matters because a depleted reservoir sits at low pressure, and liquid CO₂ crossing a large pressure drop cools sharply. Cyclic pressure and temperature at the completion, rather than the reservoir itself, is where a ship-fed injector differs from a pipeline-fed one.

The 2022 pilot is the one place where the cycle is on the record. Its permit allowed up to fourteen round trips with cargoes of up to 880 t, pumped from ISO tanks through a flexible hose, and CO₂ injected cyclically at about 40 m³/h — each injection lasting roughly twenty hours and placing about 750 t, at a maximum injection pressure of 315 bar. The commercial phase is some fifteen times that annual volume through a fixed offloading system, and nothing published says whether the rhythm changed with it.

The comparison the press release does not make

"First full-scale CO₂ storage site in the EU" is the operator's phrase, and the project defines full-scale as more than 100,000 t/y. Greensand is four times that.

The qualifier doing the work is EU. Norway is not an EU member, and Northern Lights has been injecting since August 2025, at a phase 1 capacity of 1.5 Mt/y — 3.75 times Greensand's first phase — with phase 2 taking it to at least 5 Mt/y from 2028. Northern Lights ships CO₂ to a shore terminal and then sends it out through a 100 km pipeline. Greensand sails it to the platform.

Those are two different answers to the same problem, and the trade is legible: the pipeline costs capital up front and gives you continuous flow at the well; the ship costs less up front and gives you batches. Greensand's own ambition of 4–8 Mt/y by 2030 is 10 to 20 times the current phase. By our arithmetic that is 7 to 13 ships of this size — or a pipeline.

What the draft permit allowed

One number the coverage did not carry. Before Denmark issued the permit, the European Commission gave an opinion on the draft, under Article 10 of the CO₂ storage directive. That opinion, of 19 November 2025, records the draft as allowing 2.4 million tonnes of CO₂ over eight years, starting by the end of 2026 — which the Commission itself converts to about 0.3 Mt a year.

Set that beside the 400,000 t/y the operator announced for this first commercial phase. The ceiling is cumulative, not a rate cap: 400 kt/y breaks nothing, it simply exhausts the permitted 2.4 Mt in six years rather than eight. What the arithmetic says is that the constraint on this site is not the reservoir — it is what the paperwork allows to be put into it. The draft also fixed pressure limits — reservoir pressure below 280 bar, tophole pressure never above 200 bar — from geomechanical study.

Two cautions, both ours. First, every one of those figures is described in the Commission's text as being in the draft permit; the final Danish decision is in the same 127-page bundle, but on scanned pages we have not read. Article 10 obliges the national authority to give reasons if it departs from the opinion, which makes it likely the numbers survived — likely is not a citation. Second, a permitted total is a permitted total, not a forecast: permits get varied.

What we could not confirm

Stated plainly, because none of this is in the material we could reach:

• how many injection wells the commercial phase uses. The project's own news item does not say;
• whether there is buffer storage offshore. The transfer arrangement itself is on the record: the Danish Energy Agency approved an offloading system coupled to the Nini A platform, part fixed and part flexible pipeline with a riser, to take CO₂ off the ship. What no published document says is whether anything holds that CO₂ between cargoes;
• the water depth at Nini West, and the platform's type;
• the distance offshore, given as 250 km by INEOS and 200 km by the project site. The two probably measure from different points, but we did not establish that;
• the Aurora reservoir depth, given as 2,500 m by Equinor and 2,600 m by Northern Lights.

What we got wrong on the way here is worth stating too. The Danish permit is not missing: the Energistyrelsen publishes the decision approving the storage plan and permitting CO₂ storage at Nini A (Nini West) under licence C2023/01, alongside the offloading decision of 23 September 2025, the 2022 pilot storage permit, and a run of 2026 decisions including a 2D seismic baseline survey at Nini Vest on 4 June 2026. The English-language pages we tried first returned 404; the Danish publications page has it all. A URL that will not open is a fact about us, not about the archive.

What to watch

1. Whether injection runs continuously between cargoes. This is the single fact that decides whether Greensand is a pipeline project with a ship in front of it or a genuinely batch-cycled injector.
2. The first full year of injected tonnage, against the 400,000 t nameplate. One ship, one terminal and one weather-exposed transfer make availability the number to beat, not capacity.
3. The second vessel, or the pipeline. More than half the first phase is already spoken for by a single Swedish contract of 210,000 t/y from 2028. Growth beyond one hull has to show up as steel somewhere.
4. Whether the final permit kept the draft's ceiling. The 127-page bundle Denmark publishes contains the Commission's opinion as machine-readable text, and the national decision itself as scans. The volume ceiling, the eight-year window and the pressure limits are all legible in the opinion; the monitoring plan and the injector count are not. For an industry that will be asked to re-use a great many old wells, the reasoning that qualified these ones is worth more than the ribbon.

On 15 September 2026 Strohm announced a contract from Petrobras to supply thermoplastic composite pipe — TCP — for Brazil's deepwater post-salt province. The scope includes qualification testing, engineering, field services and support for offshore trials and installation. The applications named are water injection and gas lift, at water depths up to 1,500 m. The pipe is carbon or glass fibre in a polymer matrix, spoolable, and to be trialled for handling and installation from pipe-lay vessels Petrobras already runs.

Read the applications again, because they are the story. Not production. Not gas injection. Not pre-salt.

To see why that matters, you have to know what is wrong with the steel.

The failure mode, from the regulator

Brazil's ANP issued a safety alert — Alerta de Segurança 012-ANP/SSM — on carbon dioxide stress corrosion cracking, SCC-CO₂.

The alert states that incidents began in 2016, in gas injection and production flexible pipes carrying high CO₂ content, and that failures appeared from the second year of operation against a manufacturer's stated design life of 20 years. That is the sentence to sit with: a twenty-year pipe failing in year two.

Three conditions must coincide, all of them:

• CO₂ fugacity — a property of the reservoir, not of the design;
• stress or strain in the pipe — the alert is explicit that both operating load and residual stress from manufacture were sufficient to activate the mechanism;
• a flooded annulus — water in the space between the internal pressure sheath and the outer sheath.

The geometry surprises people who have not worked on flexibles. The steel that cracks never touches the produced fluid. CO₂ permeates outward through the internal polymer sheath into the annulus. If that annulus is dry, the mechanism does not initiate; the annulus is normally gas-filled and vented. Flooding is an operational event — a damaged outer sheath, a connector, a vent valve. A new pipe does not have a flooded annulus. It acquires one.

The numbers in the alert that nobody quotes

The alert does not stop at the mechanism. It carries a working envelope, from the operator's own design measures, reported under lessons learned:

• For high-strength wire — 1,200 to 1,500 MPa UTS — there may be a cracking threshold between 3 and 5 bar of CO₂ fugacity in the annulus. Lower-strength steels are less susceptible. Strength is a liability here.
• No cracking was identified at temperatures of 25 °C or below.
• Susceptibility was not found in oil production lines to the degree seen in gas injection lines.
• Top sections live longer. In the flooded condition, the alert reports significantly longer SCC-CO₂ life in the upper riser sections, because CO₂ pressure in the annulus there is lower — surface venting continuously removes gas from the annulus.

The damage is not uniform along the line, and the reason is a boundary condition at the top.

Why Brazil, and the one CO₂ number that is primary

The reservoir supplies the first condition, and in the Santos Basin it supplies it generously. Petrobras' Form 20-F for 2025 describes Mero in its own words, in the Exploration & Production — Mero field section: "The project faces notable challenges, considering the high gas/oil ratio (415 std m³/std m³) and CO2 content in the associated gas (44%)."

44%. That is Petrobras' own figure in a document filed with the SEC — a place where being wrong has consequences — and it is worth stating plainly, because the range that circulates in commentary — 38 to 42% — is lower and unsourced.

One caution: this is Mero's number, not the pre-salt's. There is no such thing as "the CO₂ content of the pre-salt". It varies by accumulation, across orders of magnitude, within the same play. Búzios has no official CO₂ figure at all — the values quoted come from trade press and from a paper reporting a single well.

What the operator says it costs

SCC-CO₂ is not a background worry at Petrobras. It is a reported performance indicator. The 20-F glossary defines "SCC-CO2 production losses" as the measure of "the absolute production loss resulting from stress corrosion cracking, induced by CO2 in production pipelines." The company reports that in 2025 those subsea SCC-CO₂-related production losses were 34% lower than forecast after inspection campaigns and engineering actions, and that it has contracted and executed three Technological Development Orders (ETECs) aimed at creating flexible pipes immune to SCC-CO₂.

Two supplier programmes are public. NOV announced a pre-commercial agreement in March 2025 covering stainless steel armouring, for 6 in and 8 in production and gas injection risers and flowlines. Baker Hughes announced one the same month, targeting 30 years of service life in a high-CO₂ environment.

When mitigation fails, the bill arrives at the end of life. A 2026 Petrobras paper documents the decommissioning of the SPA-1 system at Mero — lines that served the FPSO Pioneiro de Libra between 2017 and 2022, 21 km of flexible pipe. Retrieval began in 2023 and was suspended. One section failed during a proof load test in November 2023, which proved the test worked. A later section failed after passing it, at the connector, on the work table of a pipe-lay vessel, with people nearby on routine accessory removal. The paper reports no injuries. The revised method splits the job in two: where the exposure can be designed out, accessories come off subsea by ROV and the proof load carries a 1.25 factor; where it cannot, the retrieval load is capped so the line must come in over the PLSV stern, under a stricter test. The pilot ran four vessels simultaneously to recover one interconnection.

Embrittled pipe is expensive to own and expensive to let go of.

So what did the award actually signal?

TCP has no tensile armour wires. Remove the steel and you remove the thing that cracks — the SCC-CO₂ locus does not exist in a carbon-fibre laminate.

But Strohm's release does not mention CO₂ at all, and the duties bought are water injection and gas lift in the post-salt. That is not a hedge on the technology; it is how qualification is supposed to work. You do not debut a new pressure-containing material on the highest-consequence, highest-CO₂, highest-GOR duty in the country. You put it in a benign service at a depth you have already trialled, and you collect evidence.

Meanwhile the steel is not being abandoned. In the same fortnight, trade press reported bids in a large Petrobras tender for up to 518.3 km of flexible lines, in two lots, with SCC-CO₂-resistant pipe explicitly in scope. Reported bid values put Baker Hughes lowest, at about R$ 3.1 bn and R$ 3.2 bn for the two lots. The tender scope and the bid values are both a single trade-press report; no primary source, no award announced. Treat them as an order of magnitude and nothing more. The direction is not in doubt: Brazil is buying a great deal of flexible pipe, and specifying it against a failure mode it had not seen before 2016.

What to watch

Three signals, none of which has happened yet:

1. TCP moving up in duty — to production or gas injection service, or into the pre-salt. That is the qualification step that would matter.
2. An ETEC pipe qualified and ordered. "Immune to SCC-CO₂" is a design target in an annual report. The test is a purchase order against a real riser specification.
3. The alert being revised. The envelope in ANP's alert — the 3 to 5 bar threshold, the 25 °C observation — came out of dissections and lab work on a young failure mode. It should move as the dataset grows, and when it moves, so do a lot of life-extension calculations.

In July 2026 Brazil produced 4,498,423 bbl/d of oil and 214,968 thousand m³/d of natural gas — 5,850,529 boe/d together. The pre-salt share was 3,728,074 bbl/d and 173,715 thousand m³/d, or 4,820,711 boe/d: 82.4% of national output.

That came from 203 producing wells. The country has 6,724 producing wells: 613 offshore and 6,111 onshore.

Both figures sit in the same ANP monthly bulletin, pages apart, and they are rarely read side by side. Read together they are the governing fact of the Brazilian upstream, and a surprising number of engineering decisions offshore Brazil follow from them.

The arithmetic nobody prints

The first and third rows are the bulletin's. The middle row is subtraction, and the per-well column is division; the inputs are not ours to adjust. Counting the associated gas, the pre-salt average rises to 23,747 boe/d per well. The onshore fleet — 6,111 wells producing the 1.9% of national oil that does not come from offshore fields — averages under 15 bbl/d each.

The gap between the two averages is a factor of about 155.

A rate is a productivity index and a drawdown

A producing well is a pressure drop with a coefficient in front of it. Whichever inflow model you prefer, rate scales with a productivity index multiplied by the difference between reservoir pressure and flowing bottomhole pressure. The drawdown term is the one you are not free to expand: it is bounded by near-wellbore rock strength, by the bubble point you do not want to cross in the reservoir, by sand, and by the asphaltene envelope. So an 18,365 bbl/d average is, before anything else, a statement about the productivity index — about permeability-thickness, about skin, and about how much rock the completion actually contacts.

The bulletin reports no reservoir properties, and this post will not invent any. But the direction of the inference is not optional: rates of that order per completion are not available from an ordinary reservoir, and they are the reason a couple of hundred wells can carry a country. The same arithmetic explains the cost structure. Capital is pulled forward into a small number of deep, long, expensive completions instead of being spread across thousands of modest ones, which is why a single well result moves a field plan here and would be noise almost anywhere else.

One caveat matters to anyone sizing equipment from these tables: the bulletin reports oil, not gross liquids. Produced water is not in the figures. Separation and water-treatment duty on the host is larger than the oil number by whatever the water cut is, and the water cut is precisely what the bulletin does not give you.

What it does to the host facility

The highest-producing installation in the country in July 2026 was the FPSO Almirante Tamandaré: 263,178 bbl/d of oil and 11,114 thousand m³/d of gas — 333,085 boe/d — from eight wells. The highest gas producer, the FPSO Marechal Duque de Caxias, made 200,802 bbl/d and 13.60 million m³/d from five wells: 57,272 boe/d per well.

Three consequences follow, and none is optional.

One well is a material fraction of the plant. At eight producing wells per hull, shutting one in removes more than 12% of throughput; at five, a fifth. That changes what an intervention is worth, how hard you argue for a rig slot, how much you are willing to spend on subsea metering to know which well is drifting, and whether you accept a completion design that cannot be re-entered. Redundancy that would be gold-plating on a 40-well platform is ordinary prudence here.

The gas is not incidental. Mero produced 50,926 thousand m³/d of pre-salt gas alongside 763,893 bbl/d of oil — about 67 m³ of gas per barrel. Itapu produced 1,455 thousand m³/d alongside 104,860 bbl/d — about 14. Nearly five to one, and both fields carry the same "pre-salt" label. Gas-oil ratio sets compression train count and power, and once you add the CO₂ content reported for these fields by operators and trade press — figures the ANP bulletin does not publish — it also sets the reinjection duty, the metallurgy and the membrane or amine selection. Nationally the picture is good: 97.8% gas utilisation and 4.67 million m³/d flared, down 29.5% on the month. It is good because the reinjection plant was designed for it.

Flowlines and risers are sized at rates most engineers never meet. Concentrating a third of a million barrels of oil equivalent per day into a handful of tiebacks pushes you against erosional-velocity limits and slug behaviour at the top of the riser, in 1,900 to 2,300 m of water, with the field 150 to 290 km from shore. Cool-down time, hydrate margin and the chemical injection rate over that length are not secondary calculations on this kind of network; they are what decides whether a shutdown is a four-hour event or a four-day one.

Capacity is not production

The same bulletin is the cure for the commonest mistake made with these numbers. The FPSO Bacalhau is rated at 220,000 bbl/d. In July 2026 the Bacalhau and Bacalhau Norte fields together produced 102,461 bbl/d — roughly 47% of the hull's capacity. Nothing is wrong. Ramp-up is a well-by-well curve, not a switch, and with a small well count each new producer is a visible step rather than a smooth trend.

Three ways to misread the same table

Field production is gross. The bulletin reports 100% of the field. An operator's release usually reports that operator's share. The two numbers are different and are not comparable.

A field is not a contract. Four pre-salt fields have production split across two contractual regimes. Búzios is 26.1205% transfer-of-rights and 73.8795% production sharing. Adding "Búzios production" to "production under sharing contracts" counts the same barrel twice.

Pre-salt is not the Santos basin. Ten fields in the Campos basin produce from the pre-salt, and Jubarte produces from both sides of the salt at once — 126,235 bbl/d pre-salt and 65,395 bbl/d post-salt in the same month, through the same infrastructure.

The short version

203 wells, 82.4% of a country. It is an enviable position and a narrow one. Every one of those wells is a single point of failure in a national statistic, which is the real argument for the redundancy, the instrumentation and the flow-assurance conservatism that visitors to these projects sometimes mistake for over-engineering. It is not over-engineering. It is what 18,365 bbl/d per well costs.