Climate Week NYC · September 20–27, 2026
Steps Ventures
Four open research programs on the ocean and climate. Independent and unfunded. Every input is public, every line of analysis code is archived, and the results that did not survive our own review are listed at the bottom of this page rather than quietly dropped.
The common method is computational reconstruction of things nobody can observe directly: where an animal spawned, where a century of ocean heat moved, what fraction of a purchased tonne is physically real. The common discipline is trying to break each result before anyone else does.
Mike German · mike@stepsventures.com · +1 443 924 9795
01
Marine carbon removal integrity
Tool live. Paper needs a co-author.
carbon.stepsventures.com
~5,000 t
issued across all ocean CDR to date
2 orders
of magnitude larger, the contracted book
11
removal methods scored on common integrity axes
What we established
- –Nearly all marine CDR realizes atmospheric removal only through rate-limited air-sea gas exchange. Where you deploy sets what fraction is ever realized, and over what timescale. A tonne contracted is not a tonne removed, and the gap is physics, not paperwork.
- –Verification, not physics or cost, is the binding gate. Efficiency and measurability are weakly positively coupled through the Revelle factor, so the most efficient water is not the easiest to measure.
- –We reclassified direct ocean capture as atmosphere-coupled rather than storage-only, which carries a testable prediction about its efficiency relative to alkalinity enhancement.
- –Satellite verification of supplier sites is possible for only a small minority of ocean CDR. The distribution of what cannot be checked from space is itself the finding.
What is still open, or did not hold
- –The manuscript scored 2 and 3 out of 10 with two independent frontier models acting as hostile referees. We rebuilt it and it still needs a domain co-author to carry it. We are not preprinting it.
- –The often-quoted "fraction realized" headline rests on a denominator we do not think is valid. We report issuance against contracts instead, and label the rest illustrative.
The ask
An oceanographer co-author. Registries and buyers willing to have the screen run against their actual book.
02
The European eel
Data and code archived. Preprint posting. Journal next.
doi.org/10.5281/zenodo.20980884
~0.4°/dec
poleward migration of the 22 to 24 °C spawning band
+0.095°/dec
movement of the SST-gradient front, not significant
>90%
collapse in recruitment we do not claim to explain
What we established
- –The eel spawns only in the Sargasso Sea. The dominant explanation for its collapse is that warming pushed the spawning ground poleward and broke the larval route to Europe. That has been asserted from index correlations and never tested against the front itself.
- –The spawning thermal band migrated poleward. The front that marks the spawning ground moved far less. This is a power-limited null, meaning no detected movement rather than proof of a fixed front.
- –Modeled larval delivery to Europe is governed by larval swimming behaviour, not by front latitude, across three independent current products and depth-resolved transport.
- –A CMIP6 single-forcing comparison leaves greenhouse gases as the only single forcing of the right sign and magnitude for the isotherm migration.
What is still open, or did not hold
- –We exclude one specific mechanism, not every oceanic pathway. The one we do not exclude is a decline in drift-route productivity: corridor chlorophyll down 5.3% per decade, net primary production down 3.3% per decade, same drift window, same sign.
- –An in-situ plankton record we obtained to extend that series back to 1958 does not cover the warm western drift path, so it can neither confirm nor refute it. We report that as a negative result.
The ask
Modern larval cruise data from the Sargasso spring surveys. A fisheries-oceanography reader before journal submission.
03
The sea turtle lost years
Data and code archived. Preprint posting. Journal next.
doi.org/10.5281/zenodo.21841875
AUC 0.91
forward model against real tracked turtles
~13 days
until an individual’s origin is physically erased
171,823
drifter pairs behind that measurement
What we established
- –After hatching, sea turtles vanish into the open ocean for years. A forward drift model reproduces where they go, predicting the positions of real tracked turtles at AUC 0.91 when compared like-for-like on matched durations.
- –The inverse problem is not merely hard. Trajectory pairs that begin 15 km apart separate beyond the 150 km spacing between nesting beaches in about 13 days. The ocean destroys individual origin information almost immediately, which is why the field uses genetics rather than back-tracking.
- –One relation governs both directions: inversion is possible only while age at observation stays below the diffusive mixing time across the origin spacing. Days-old eel larvae are invertible. Months-old turtles are erased. Same phase diagram.
- –Rookery genetics traces origin through a separate channel that the ocean cannot mix away. We quantified it rather than waving at it, and it resolves populations well and individuals poorly.
What is still open, or did not hold
- –We make no climate claim. Whether the delivery corridor drifts as the ocean warms is product-dependent and underpowered, and a turtle drifts for years before its endpoint is observable, so the recent window is a handful of cohorts in any dataset. We state the bound instead of a finding.
The ask
Long-duration oceanic-stage tracks. These are the single input that would move the corridor question.
04
Rebel Reef, Tela Bay, Honduras
Field program. Narrow paper drafted, needs co-authorship.
steps-re.github.io/rebel-reef
~68%
live coral cover at Cocalito
706 days
above the bleaching threshold, against 183 at Utila
DHW 19.9
peak heat dose, comparable to the Florida Keys die-off
What we established
- –A murky, sewage-loaded reef next to a river mouth holds high coral cover while its clear-water neighbours degrade. It took roughly four times their cumulative bleaching heat and a dose comparable to the Florida Keys mass mortality.
- –It is not a thermal refuge. Its temperature history is statistically indistinguishable from the neighbours that died.
- –Across sixteen reefs, turbidity, river outflow and carbonate subsidy all fail to sort thriving from degraded. Persistence is not explained by the coarse physical setting.
- –In June 2023 part of the bank died in about five days under a milky layer that sank. The event class is reasonably confident: acute oxygen loss, not bleaching and not disease.
What is still open, or did not hold
- –What triggered the oxygen crash is genuinely open. River flooding is contradicted by discharge data, and vessel dumping is not supported by vessel-tracking records. The decisive chemistry is bay-scale and sub-grid, so satellites cannot settle it.
- –Our own red team killed the first version of the heat-and-grazing interaction result. The narrower recovery-phase version survived. The site says so.
The ask
The Op Wallacea time series. In-water dissolved-oxygen and CTD monitoring. A leach assay on the bay’s heavy-mineral black sand.
What we retracted
Every headline claim in these programs was put through an adversarial recompute, often by independent models instructed to destroy it. A lot of them died. Publishing the casualty list is the point: a result that has survived a documented attempt on its life is worth more than one that has never been attacked.
Eel: the spawning front is pinned by a fixed wind-driven convergence
Dropped. The wind leg did not survive recompute with the bug fixed and the full record.
Eel: the front is stationary
Narrowed to a power-limited null. Two SST products disagree on its direction.
A cross-species climate-vulnerability index
Retired entirely. The literature-scored outcome is confounded by research effort at r = +0.79, stronger than any model we built. Cross-species severity scores measure how much a species is studied.
Turtle: warming in place
Cut from the title and demoted to a bound. Corridor drift flips sign between current products.
Turtle: the northward escape decline is AMOC-coupled
Retracted. A shared forced-trend artifact that collapsed under de-forcing and multiplicity correction.
Marine CDR: efficiency is decorrelated from verifiability
Overturned by our own robustness check. The original result was an artifact of counting the seasonal cycle as noise.
Coral: same heat, opposite fate
Killed by our own red team as tautological. The narrower recovery-phase mechanism survived.