Most industrial R&D never becomes citable. The work happens, the deliverable ships, the model goes into a workflow, and the methods stay locked inside a data room nobody outside the account will open. That is a quiet waste: the interesting part of applied research is usually the method, and the method is almost never the confidential part. The confidential part is the reservoir.
This is a milestone recap of one engagement that went the other way. A single multi-year programme with a mid-sized Middle East carbonate operator produced two peer-reviewed papers and a conference poster: one paper on fracture and bedding detection, one on vug quantification, and the poster carrying the whole detection pipeline. None of it breached client confidentiality, because no field data left the manuscripts. What made it possible was a decision taken at proposal stage, before a line of the detector was written.
The clause that came before the science
Look at the actual origin. The proposal that opened this account (reference DK2018015) was a machine-learning scope for the operator's well logs. Buried in it, past the technical phases and the cost lines, was a marketing section. It contracted for an approved joint case study, a video interview, progress blogs, joint award entries, and conference talks. In plain terms: the right to publish, negotiated and signed at proposal stage, years before the first manuscript existed.
That single clause is the whole story. Without it, every method we developed would have been a work-for-hire deliverable owned inside the account, publishable only with a fresh permission conversation that tends to arrive too late or not at all. With it, publication was a pre-agreed pathway rather than a favour to ask. The programme that ran under that proposal was substantial: a scope of work signed in 2021, three phases over roughly twenty months, spanning more than eighty processed and interpreted borehole-image wells. Large engagements produce a lot of method. The clause is what let that method surface.
The instrument above is the argument in one frame. Toggle the orange gate shut and the same three outputs turn back into withheld deliverables. That is not a rhetorical flourish. It is the literal counterfactual: identical work, zero citable results, because the precondition for turning confidential R&D into science was contractual, not technical.
Two papers, one dataset, different physics
The two papers came out of the same borehole-image corpus but attack different rock. The first is a fracture-and-bedding detector we call GeoBFDT. It treats an unwrapped borehole image as a scene and detects the sinusoids that planar features trace when a cylinder is cut open and flattened. A plane crossing the borehole at an angle projects to a sine curve, and its dip and azimuth fall out of the curve's amplitude and phase:
This paper was accepted at a reservoir-engineering journal and is in copyediting. Acceptance did not come easily. One reviewer initially recommended it as unpublishable, and the associate editor floated an unusual remedy, that we upload the code to a public repository so a third party could verify it on open data. We could not. The confidentiality agreement blocked a code release; we re-approached the client for permission and were declined. The point here is narrow: the paper survived because everything a reviewer needed to judge the method could be shown with depths masked and the reservoir kept private.
The second paper attacks vugs, the secondary-porosity cavities that read as dark speckles on a borehole image and control permeability in carbonate rock. Different problem, different toolkit: mode extraction, a locally applied Gaussian-weighted adaptive threshold, contour extraction, and Laplacian contrast validation, reporting vug count, area, and shape every 0.1 m. That paper is in its second revision at the same journal. During that revision the technical editors confirmed, in writing, that the fracture paper had been accepted, which is how one manuscript's review file came to certify the other's status.
The poster that carried the pipeline
The third output is a conference poster, presented on 19 April 2024. It carries the full detection pipeline on one board, with six authors across five affiliations spanning a geoscience department, our research lab, the academic partner, the operator, and a petroleum-engineering centre of excellence. The poster also substantiates the pipeline's most quotable number: it interprets a five-metre section in under ten seconds, against a semi-automatic prior art that measured the same task in hours. That is the speed claim we now stand behind in outreach, and it lives in a citable, dated, co-authored artifact rather than a sales slide.
A poster is worth more than it looks. It is the fastest way to plant a dated public record of a method, it clears the same confidentiality bar as a paper, and it names the collaboration to a room full of the exact audience we want. Because the marketing clause already covered conference talks, putting it up required no new negotiation.
How the method publishes while the reservoir does not
The engineering discipline that makes this repeatable is anonymisation at the source. Two habits do most of the work.
First, depth masking. Every depth in every figure and table is written with its leading digits replaced, in the style XX17.04. A reader sees the interval structure, the spacing between picks, the relative geometry that a method has to reproduce, and learns nothing about where in the field the interval sits. The masking is applied once, in the publication-grade figures, so the same before-and-after art is reusable across the paper, the supplementary material, and the poster without a second scrub.
Second, a hard line on field data. The reservoir's resistivity and salinity values were disclosed to reviewers privately, under the confidentiality agreement, but never entered the manuscripts. A reviewer's private question can be answered off the record; a published paper cannot carry a number that identifies the field. Keeping those two channels separate is what let the review conversation be candid while the paper stayed clean.
The rule that makes confidential R&D publishable
Publish the method, never the field. A detector's dip-azimuth accuracy, a threshold sweep, an ablation over well count, a sensitivity curve against depth error, are all method properties and none of them locate a reservoir. Depths, resistivities, salinities, and field names do. Mask or withhold the second set at the figure-generation step, and the first set publishes freely.
How to structure an engagement so this can happen
The transferable lesson is not about carbonate geology. It is about how a research engagement is set up before the research starts.
Negotiate publication and marketing rights in the proposal, not at the end. By the time a method is worth publishing, the commercial relationship has moved on, the sponsor who would have said yes may have rotated out, and a fresh request reads as scope creep. A pre-agreed clause converts publication from a favour into a plan.
Design anonymisation as a deliverable, not an afterthought. If depth masking and the field-data firewall are decided at figure-generation time, publication is a straight path. Retrofitted onto figures built with raw depths, every exhibit has to be rebuilt, and the temptation to skip the paper grows with the rework.
Treat a poster as a first-class output. It is cheaper than a paper, plants a dated public record, and lets a collaboration show its work while the papers are still in review.
Limitations
This is one engagement, and the arrangement that made it work was specific: a research-oriented operator, an academic co-author who could carry manuscripts through peer review, and a proposal drafted early enough to include a publication clause. Not every client will sign one, and a purely operational contract with no research intent is a different animal. The two papers are, at the time of writing, one accepted-in-copyediting and one in second revision, so the citable record is real but still settling. The speed claim of under ten seconds per five-metre section is a poster-substantiated figure for the detection model on this corpus; it is a method-level marker on our runs, not a benchmark to quote against a different tool stack. And the anonymisation discipline described here clears a peer-review confidentiality bar, which is not the same as clearing every client's specific data-handling policy. The precondition we are naming, rights negotiated at proposal stage, is necessary here. It is not, on its own, sufficient anywhere.
References
-
Machine-learning proposal for well-log analytics (reference DK2018015), including the marketing and publication section contracting for a joint case study, conference talks, and joint award entries; internal engagement archive, negotiated at proposal stage.
-
Signed scope of work (2021): three phases over approximately twenty months across more than eighty processed and interpreted borehole-image wells; internal engagement archive.
-
Conference poster on deep-learning fracture and bed detection in subsurface geological analysis, presented 19 April 2024, six authors across five affiliations, substantiating a detection speed of under ten seconds per five-metre section; internal engagement archive.
-
Vug-quantification manuscript, second-revision response to the technical editors, which confirms the fracture-and-bedding manuscript was accepted at the same reservoir-engineering journal and was in copyediting; internal engagement archive. Field resistivity and salinity values were disclosed to reviewers under confidentiality and withheld from both manuscripts.
