Google DeepMind's $40M Genesis Mission: AI for Science Meets Ecommerce Video
By VEONIB | 2026-10-18
Quick Answer
Google committed $40 million in AI tokens and cloud credits to the U.S. Department of Energy's Genesis Mission, giving Genesis Mission awardees in-kind access to AlphaEvolve, AlphaFold 3, AlphaGenome, WeatherNext and AlphaEarth Foundations, plus one year of Gemini for Government seats and tokens for tens of thousands of National Laboratory staff. For ecommerce teams, the transferable value is not the science compute — it is the deployment pattern: agentic tooling, governed seats and measured removal of manual steps.
TL;DR
- Google DeepMind and Google Cloud pledged $40 million in AI tokens and cloud credits to the DOE Genesis Mission, announced at the DOE Genesis Mission Summit 2026 and published on 2026-07-23.
- Five science models — AlphaEvolve, AlphaFold 3, AlphaGenome, WeatherNext and AlphaEarth Foundations — are offered in-kind to Genesis Mission awardees, not to the general public.
- Gemini for Government seats and tokens will reach tens of thousands of DOE National Laboratory users for one year across research, operations and management teams.
- At National Laboratory of the Rockies, Gemini cut microscope calibration from over 90 minutes to about 13 minutes and reduced image-focus steps from up to 50 to two.
- Ecommerce relevance is indirect but real: agent-plus-vertical-model bundles, seat-based governance and step-count reduction are the same three levers that constrain AI product video production today.
Table of Contents
- What Google Actually Committed to the Genesis Mission
- The Five Frontier Science Tools, Explained
- Gemini for Government: The Enterprise Governance Signal
- Inside the Labs: What PNNL and NLR Actually Measured
- Is Any of This Useful for Ecommerce Video?
- Gemini and Veo Inside an Ecommerce Video Workflow
- Cost, Speed and Scalability: Reading the Token Economics
- Risks and Limits: Where the Hype Stops
- Outlook: What to Watch Through 2027
- Recommendations
- FAQ
According to Accelerating the frontiers of scientific discovery: Google's $40M commitment to the Genesis Mission published by Google Cloud Blog on 2026-07-23, Google DeepMind and Google Cloud are directing $40 million in AI tokens and cloud credits toward researchers working under the Department of Energy's Genesis Mission. The post, authored by Pushmeet Kohli and Karthik Narain, details five science models offered in-kind and a year of Gemini for Government access for National Laboratory staff.
On its surface, this is a public-sector research story with no obvious connection to product marketing. In practice, it is one of the clearest published descriptions of how a large AI vendor packages frontier models for institutional deployment: which tools are bundled together, how access is governed, and what success is measured against. Those three decisions are exactly the ones ecommerce teams face when they industrialise AI video production. This analysis separates the reported facts from VEONIB's assessment of what they mean for Shopify merchants, Amazon sellers, TikTok Shop operators and the agencies that serve them.
Hero Image Alt Text: Google DeepMind Genesis Mission AI research models and ecommerce AI video production workflow Caption: Google's $40M Genesis Mission commitment bundles frontier science models with governed Gemini access for 17 DOE National Laboratories. OG Image Title: Google's $40M Genesis Mission and What It Signals for Ecommerce AI Video Suggested Visual: A split-screen composition showing a scientific laboratory instrument on the left and a product video storyboard interface on the right, connected by a single shared workflow arrow labelled "agent → vertical model → measured output."
What Google Actually Committed to the Genesis Mission
Google's commitment is a compute and access package rather than a cash grant. The $40 million is denominated in AI tokens and cloud credits, distributed in two tracks: in-kind access to Google DeepMind's science portfolio for Genesis Mission awardees, and a year of Gemini for Government seats and tokens for tens of thousands of DOE National Laboratory users.
Original Fact
- The $40 million figure refers to AI tokens and cloud credits, not direct funding to researchers.
- The announcement was made at the DOE Genesis Mission Summit 2026.
- The Genesis Mission is described as a national effort to harness AI and double the pace of American scientific discovery within a decade.
- Google DeepMind had previously announced an early access program providing AI for science tools to all 17 DOE National Laboratories, and Google Public Sector had outlined how Gemini for Government could serve as an AI backbone for the DOE.
- The post states that the practical impact is already visible across the laboratory ecosystem, while acknowledging "a lot of work still to do."
Not specified in the original source: the split between tokens and credits, the number of Genesis Mission awardees, eligibility rules for applicants, the distribution timeline, or what happens to unused tokens at the end of the one-year term.
The sequencing matters. Google did not lead with a check. It led with an early access program across 17 laboratories, then a platform narrative around Gemini for Government, then a scaled credit commitment. That is a deployment ladder: prove the tool inside a controlled cohort, publish named results, then widen access with a metered commercial structure behind it. Enterprise buyers in any sector should recognise the shape, because it is the same ladder vendors use to sell AI production tooling into retail organisations.
VEONIB Insight
The most useful detail here is the unit of commitment. Google is not selling licences; it is selling metered capacity plus governed seats. That combination — consumption credits on top of a fixed seat layer — is precisely how AI video production is being priced in 2026, and it has a direct budgeting consequence for merchants. Seat costs are predictable and can be planned per team member; token and credit costs scale with retries, which means quality control is a financial control. A team that reduces prompt retries from five to two per shot does not just save time, it cuts the variable cost of every video. Treat your AI video stack as two line items from day one, and instrument retry rates the way you would instrument ad spend.
The Five Frontier Science Tools, Explained
The portfolio Google is offering is deliberately heterogeneous: one general-purpose agent plus four domain-specific models. AlphaEvolve is a Gemini-powered coding and discovery agent for designing advanced algorithms. AlphaFold 3 predicts the structure and interactions of proteins and other biomolecules. AlphaGenome interprets how DNA variation, including the non-coding genome, shapes biology and disease. WeatherNext is a family of AI weather forecasting models. AlphaEarth Foundations is a foundational model for mapping and understanding the planet.
| Tool | Core Function | Domain | Practical Ecommerce Relevance |
|---|---|---|---|
| AlphaEvolve | Gemini-powered coding and discovery agent for algorithm design | Software and mathematics | Indirect: demonstrates automated search over candidate solutions, the same loop used for prompt and script variant testing |
| AlphaFold 3 | Predicts structure and interactions of proteins and biomolecules | Biology | None direct |
| AlphaGenome | Interprets DNA variation including non-coding regions | Genomics | None direct |
| WeatherNext | AI weather forecasting model family | Geoscience | Indirect: seasonal campaign timing, shoot scheduling, logistics planning |
| AlphaEarth Foundations | Foundational model for planetary-scale mapping | Geospatial | Low to moderate: location scouting, background plate sourcing, market-level context |
The architecture is the story, not the individual models. Three of the five are single-domain specialists that would be useless outside a laboratory. AlphaEvolve is the outlier: it is a general reasoning and search agent pointed at a hard optimisation problem. The bundle therefore reads as "one horizontal agent, several vertical specialists" — a structure that is now standard in commercial AI product design.
VEONIB Insight
Do not build an ecommerce roadmap on AlphaFold or AlphaGenome access; that access is scoped to Genesis Mission awardees and has no commercial pathway for retailers. Build it on the pattern instead. In AI video production, the horizontal agent is the planning layer that analyses a product URL, extracts selling points and drafts a script; the vertical specialists are the image and video generators. Teams that treat one general LLM as a substitute for both layers consistently underperform, because script quality and render quality have different failure modes and need different tools. The Genesis portfolio confirms the division of labour rather than inventing it.
Gemini for Government: The Enterprise Governance Signal
The second half of Google's commitment is access, not science. Gemini for Government seats and tokens will be provided for one year to tens of thousands of users across the DOE National Laboratories' operations, research and management teams. The post describes this as a single secure foundation spanning work "from the research bench to the administration of specialised user facilities."
Original Fact
- Access is seat-based and lasts one year.
- The stated user base spans operations, research and management — not only scientists.
- Google frames the value as a unified, secure platform rather than as raw model capability.
That framing is worth pausing on. When a vendor sells an AI model to a public institution, the deciding criteria are rarely benchmark scores. They are data residency, access control, auditability, procurement compatibility and whether the same tenancy can serve a bench scientist and an administrative officer without separate contracts. Google is explicitly positioning Gemini for Government on those grounds.
For commercial teams, the lesson translates almost directly. AI video programmes stall at the same gate: individual creators using personal accounts with no shared asset library, no approved prompt set, no review trail and no way to prove which model version produced which asset. That is a governance failure, not a creativity failure.
VEONIB Insight
Most ecommerce brands do not need government-grade tenancy, but they do need the four properties it implies: one workspace, versioned prompts, an approval step before publish, and a record of which model and asset version produced each deliverable. Brands generating more than roughly 20 product videos per month should establish this before scaling volume, because retrofitting provenance onto a year of ungoverned output is expensive and sometimes impossible. The DeepMind analysis of securing AI agents covers the same control problem from the agent side, and it applies equally to a creative pipeline.
Inside the Labs: What PNNL and NLR Actually Measured
Two laboratory case studies carry the article's evidence. At Pacific Northwest National Laboratory, senior scientist Dr. Henry Kvinge uses AlphaEvolve to explore massive mathematical systems too complex for manual analysis, describing systems like AlphaEvolve as "perfect for this search" because they pair broad mathematical knowledge from LLMs with automated exploration of countless angles.
At National Laboratory of the Rockies, Dr. Steven R. Spurgeon's team deployed Gemini inside physical instruments for autonomous materials discovery. The reported results are specific: microscope calibration dropped from over 90 minutes to about 13 minutes — roughly eight times faster — and the manual steps required to focus an image fell from as many as 50 to two.
Original Fact
- PNNL is using AlphaEvolve for combinatorial and mathematical search; Dr. Kvinge states the team is "still experimenting" but that discoveries are already shaping future research.
- NLR reports an eight-fold reduction in calibration time and a reduction from up to 50 manual steps to two.
- Both results are attributed to laboratory researchers in a vendor-published blog post.
The numbers that matter are not accuracy percentages. They are elapsed time and step count. An eight-times speed-up arrived not from a smarter microscope but from removing 48 of 50 manual interventions. That is a workflow result dressed as a model result, and it is the single most transferable finding in the post.
VEONIB Insight
Audit your own video pipeline in steps before evaluating any model. A typical product video flow runs from URL intake through product analysis, script, storyboard, image prompts, video prompts, generation, retries, voiceover, subtitles and QA — often 40 to 60 discrete manual actions per asset. The NLR result suggests where to look: not at the highest-quality generator, but at the handoffs between stages. Templating the script-to-storyboard handoff and the storyboard-to-prompt handoff usually removes more total time than switching video models, and it compounds across every SKU rather than applying once.
Is Any of This Useful for Ecommerce Video?
Directly, no. Nothing in the Genesis Mission portfolio generates product footage, writes ad hooks or renders a TikTok creative. The access is restricted, the models are domain-specific, and the evaluation criteria are scientific validity rather than conversion performance. Any claim that this announcement changes AI video for commerce would be unsupported.
The comparison is still instructive, because it makes the differences in operating constraints explicit.
| Dimension | Genesis Mission AI | Ecommerce Video AI | Implication |
|---|---|---|---|
| Access model | Restricted to DOE awardees | Commercial APIs, seats and credits | Commerce teams buy capacity; they do not apply for it |
| Primary objective | Scientific discovery | Conversion, watch time, return on ad spend | Optimisation targets rarely overlap |
| Evaluation standard | Scientific validity and reproducibility | Hook rate, click-through, creative fatigue | Commerce tolerates subjective judgement |
| Input data | Experimental and sensor data | Product catalogues, UGC, brand guidelines | Commerce data is messy and rights-constrained |
| Iteration cycle | Months to years | Hours to days | Commerce needs versioning, not validation |
| Cost of failure | High; experiments are expensive | Low; failed creative is a sunk media test | Commerce can afford volume experimentation |
The most meaningful overlap is methodological. Both domains need closed loops: observe, reason, decide, act, measure. That loop is what makes NLR's autonomous experimentation notable, and it is what separates an AI video programme that improves month over month from one that produces a fixed quality ceiling.
VEONIB Insight
Borrow the loop, not the tools. A closed-loop commerce video process looks like this: publish a variant, read hook rate and hold rate at three seconds, rewrite the script structure rather than the visuals, regenerate, and record which change caused which delta. Teams that only regenerate visuals plateau quickly, because the constraint is usually the first three seconds of the script. This is also why self-reported vendor benchmarks deserve scepticism in commerce: a model can win on a generic aesthetic score and still lose on a specific product category where geometric accuracy matters more than polish.
Gemini and Veo Inside an Ecommerce Video Workflow
Gemini models, together with Google's Veo video generation family, are relevant to commerce less because of this announcement and more because of how multimodal reasoning models fit into a production chain. Gemini's role is upstream and analytical; Veo's role is downstream and generative. The following assessment is VEONIB's evaluation based on published capabilities and production use, not a claim from the source article.
| Capability | VEONIB Assessment | Practical Note |
|---|---|---|
| Product consistency | Moderate | Requires locked reference images and explicit geometry constraints in the prompt |
| Character consistency | Moderate | Stronger within a single generation run than across sessions; use a fixed persona reference |
| Motion quality | High on short clips | Sustained complex motion degrades; keep shots under five seconds and cut in the edit |
| Camera movement | Prompt-level control | Pan, push-in and orbit are achievable; compound camera moves need staged prompts |
| Text rendering | Unreliable | Never rely on generated on-screen text; add captions and price overlays in editing |
| Prompt controllability | Strong with structured prompts | Storyboard-derived prompts outperform freeform descriptions |
| Editing flexibility | Clip-level | Regeneration is faster than repair; treat each clip as a replaceable unit |
| Production speed | Minutes per clip | Full sequence time is dominated by prompt iteration, not render time |
| Cost efficiency | Moderate | Credit consumption is driven by retry count; quality control is cost control |
| Commercial readiness | Yes, with human review | Legal and brand review remain mandatory before paid media |
| Scalability | High when templated | Scales on script and prompt templates, not on ad-hoc prompting |
Recommended ecommerce use cases: lifestyle product shots, UGC-style talking-head formats, short product demos, category-level brand story sequences, and hook variants for TikTok Ads and Meta Ads. Recommended video types: YouTube Shorts, TikTok Ads, Meta Reels placements, Shopify product page loops, Amazon supplementary product video where policy permits, and storyboarded brand films. Where it fits less well: long-form demonstrations requiring precise mechanical behaviour, regulated claims requiring exact on-screen text, and any format where the product's geometry must be pixel-accurate without a real photograph as a base.
Inside the VEONIB workflow — Product URL → Product Analysis → Script → Storyboard → Image Prompt → Video Prompt → AI Video → Voice → Subtitle → Publishing — reasoning models do their heaviest work at the analysis, script, storyboard and prompt stages, while video models handle the generation stage. Voice, subtitle and publishing sit outside the Google stack in most implementations. The critical dependency is that each stage produces a structured artefact the next stage can consume; without that, every stage becomes a manual re-briefing.
VEONIB Insight
Adopt now if your bottleneck is scripting and prompt authoring rather than rendering. Most merchants who report disappointing AI video output are not limited by model quality — they are feeding unstructured prompts with no storyboard, no reference image and no shot specification, then blaming the generator. Wait on full automation if your catalogue includes complex assemblies, transparent materials, reflective packaging or size-critical apparel, where an unreviewed generative pipeline will produce plausible but wrong product depictions that create returns and compliance risk.
Cost, Speed and Scalability: Reading the Token Economics
Google's decision to fund research with tokens rather than cash is a statement about where it believes the binding constraint sits. In a metered AI economy, capacity is the scarce resource, and credits are the mechanism that converts a strategic relationship into predictable consumption.
The same logic governs AI video budgets. Credit consumption in video generation is dominated by retry overhead, not by output length. A ten-second clip that succeeds on the third attempt costs roughly three times a first-pass success. This makes prompt discipline the highest-leverage cost lever available to a merchant, well ahead of negotiating seat pricing.
| Cost Driver | Typical Impact | Practical Control |
|---|---|---|
| Retry rate per shot | Multiplies generation cost directly | Storyboard-locked prompts and reference images |
| Shot length | Longer clips compound failure probability | Generate short clips and assemble in the edit |
| Resolution | Higher tiers cost more per second | Draft at low resolution, finalise only approved shots |
| Variant count | Linear cost growth per hook tested | Test script variants before visual variants |
| Human review time | Usually the largest hidden cost | Define a single approval owner per asset |
Original Fact from the source: the $40 million commitment is structured as tokens and credits rather than cash. VEONIB assessment: this structure is now the default procurement model for AI capability at institutional scale, and brands should expect the same metered structure to persist in commercial creative tooling rather than reverting to flat licensing.
VEONIB Insight
Price your video programme per published asset, not per generation. If a merchant publishes 40 product videos a month and burns three times the necessary credits on retries, the inefficiency rarely shows up clearly in a monthly invoice — it hides inside a workflow that feels busy. Track retries per approved asset as a first-class metric alongside cost per video and time to publish. A reasonable early target is fewer than 2.5 generations per approved clip; teams above 4 should invest in template and storyboard work before adding volume.
Risks and Limits: Where the Hype Stops
Several limits deserve explicit statement, because the announcement is easy to over-read.
- Access is not universal. The science models are offered to Genesis Mission awardees. There is no stated pathway for commercial retailers.
- Results are vendor-published. The PNNL and NLR outcomes are attributed to named researchers in a Google blog post. They are credible as directional evidence, not as independent peer-reviewed benchmarks.
- Timelines are unspecified. The post does not state when credits are disbursed, how awardees are selected, or what happens after the one-year Gemini term.
- Domain transfer is not automatic. A model that accelerates microscope calibration does not imply anything about generating a compliant product video for a regulated category.
- Vendor concentration is a real exposure. Brands that standardise entirely on one provider's stack inherit that provider's pricing changes, policy shifts and deprecation schedule. Maintain a documented fallback model for every pipeline stage.
The competitive landscape frames this risk. OpenAI and Anthropic compete on reasoning and agent capabilities, Runway, Pika and MiniMax compete on video generation, and HeyGen competes on avatar-led presentation. No single vendor leads on every stage of a commerce pipeline, and the brands that scale successfully treat these as replaceable components behind a stable internal prompt and asset layer.
VEONIB Insight
Build for portability before you build for speed. Concretely: store prompts and storyboards in your own system rather than in a vendor's editor, keep a canonical asset library outside any single tool, and re-test your primary generator quarterly against a fixed ten-SKU benchmark set. The teams caught out by model deprecations are almost always the ones whose creative process existed only inside one interface. The Genesis Mission structure — restricted specialist access alongside widely available general models — is a reminder that capability tiers will keep shifting under you.
Outlook: What to Watch Through 2027
Three developments are worth tracking. First, whether Google publishes awardee-level outcomes from the Genesis Mission, which would test whether the credit model produces durable scientific throughput or one-off demonstrations. Second, whether Gemini for Government tenancy features — audit trails, shared workspaces, approval gates — migrate into commercial Google Cloud offerings for retail and media teams. Third, whether agent-plus-specialist bundles become standard commercial packaging, since that structure is what the science portfolio already implies.
For ecommerce specifically, the directional signal is that frontier capability will keep arriving through metered, governed channels rather than open licences. Brands should plan for a permanent mix: one reasoning model for analysis and scripting, one or more video generators for rendering, and an internal system of record holding scripts, storyboards, prompts and published assets.
VEONIB Insight
Position your team for a world where the model changes every quarter and the workflow does not. The durable assets are your product analysis templates, your script frameworks and your storyboard conventions — the parts that are yours. If a new release arrives and your first question is "how do we prompt it," your process is under-documented. If your first question is "which stage does it replace in our pipeline," you are ready to adopt it in days rather than months.
Recommendations
Shopify Merchants Start with 10 best-selling SKUs and build one reusable storyboard template per product category. Generate short clips and assemble in the edit rather than attempting long single renders. Track time to publish and retries per approved asset for four weeks before expanding the catalogue.
Amazon Sellers Prioritise product accuracy over visual flair. Use reference-based image conditioning wherever the generator supports it, and add all on-screen text, claims and compliance statements in editing rather than generation. Keep a human review gate for anything touching regulated categories.
TikTok Shop and DTC Sellers Test script variants before visual variants. Build a hook library of five opening structures per category and rotate them against fixed visuals to isolate what is actually driving hold rate at three seconds.
AI Developers Treat the science portfolio as an architectural reference: one reasoning agent, several specialists, structured artefacts between stages. Define JSON schemas for storyboard and prompt objects so stages remain swappable as models change.
SaaS Founders The metered-plus-seat pricing model used here is the one your buyers now expect. Offer a predictable seat layer for planning and a consumption layer for generation, and expose retry analytics so customers can see where their credits go.
Content Marketers and Video Creators Document prompts and storyboards outside your generation tool. A creative process stored entirely inside one vendor interface is a single point of failure, and it cannot be audited when a client asks why a claim appeared in a video.
FAQ
What is the Genesis Mission? The Genesis Mission is a U.S. national effort, launched via presidential action in November 2025, that aims to use AI to double the pace of American scientific discovery within a decade. Google DeepMind committed to supporting it in December 2025 and expanded that commitment in July 2026.
How much did Google commit and in what form? Google committed $40 million in AI tokens and cloud credits, announced at the DOE Genesis Mission Summit 2026. The commitment is delivered as in-kind access to five DeepMind science models for Genesis Mission awardees, plus one year of Gemini for Government seats and tokens for tens of thousands of DOE National Laboratory users.
Can ecommerce businesses access AlphaFold 3, AlphaGenome or AlphaEvolve through this programme? No. The in-kind access described is scoped to DOE Genesis Mission awardees. The source article states no commercial pathway, eligibility route or public application process for retailers or agencies.
Does Gemini for Government mean Gemini is unavailable for commercial use? No. Gemini for Government is a separate, secure deployment tier for public-sector use. Commercial Gemini access through Google Cloud and consumer products is a distinct offering, and the article does not state any restriction on commercial availability.
What does the National Laboratory of the Rockies result actually prove? It proves that embedding a reasoning model inside an instrument workflow can compress a task from over 90 minutes to about 13 minutes and cut manual image-focus steps from up to 50 to two. It is a workflow result reported by a named researcher in a vendor blog, and it has not been independently benchmarked.
Should ecommerce teams change their AI video stack because of this announcement? Not directly — none of the five science models generate commerce video. The actionable takeaway is structural: adopt agentic scripting and structured storyboards upstream, keep prompt and asset ownership in-house, and measure retry rates as a cost metric.
Related Reading
- How Google DeepMind Securing AI Agents Impacts Ecommerce Video Production
- SynthID Bio and AI Content Provenance for Ecommerce Video Teams
- Google DeepMind Robotics Models Reshape AI Video for Ecommerce Content
- AI Video Ad Creation With GPT-6 Astra: Higgsfield AI Case Study
References
- Google DeepMind - official site of Google's AI research division
- Google Cloud - official site of Google Cloud
- Google AI - official site of Google's AI division
- U.S. Department of Energy - official site of the DOE
- Pacific Northwest National Laboratory - official site of PNNL
- National Laboratory of the Rockies - official site referenced in the source article
- OpenAI - official site of OpenAI
- Anthropic - official site of Anthropic
- Runway - official site of Runway
- Pika - official site of Pika
- HeyGen - official site of HeyGen
- MiniMax - official site of MiniMax
Sources
- Source Article: Accelerating the frontiers of scientific discovery: Google's $40M commitment to the Genesis Mission - Google Cloud Blog, published 2026-07-23
- Official Website: Google DeepMind
- Related Documentation: AlphaFold 3 announcement and AlphaEvolve availability on Google Cloud
- Related Documentation: Launching the Genesis Mission - White House presidential action
Try VEONIB
VEONIB converts a product URL into product analysis, video scripts, storyboards, image prompts, video prompts and finished AI marketing videos through a single automated pipeline. Teams that want the structured, template-driven workflow described in this article can review the platform at VEONIB AI video generation.
Credibility Assessment
From the source: the $40 million figure and its token-and-credit structure, the five named science models and their described functions, the Gemini for Government seat and token provision for tens of thousands of DOE National Laboratory users, the PNNL and NLR case studies with their quoted researchers, the eight-fold calibration reduction and the reduction from up to 50 manual focus steps to two, and the prior early access program across all 17 DOE National Laboratories. These are reported facts from a vendor-published blog post dated 2026-07-23.
VEONIB analysis: all assessments of ecommerce relevance, the Gemini and Veo capability evaluation, the workflow-fit analysis, cost and retry guidance, portability recommendations, and the competitive framing of OpenAI, Anthropic, Runway, Pika, HeyGen and MiniMax. These are interpretations, not claims made by Google.
Uncertain: the token-versus-credit split, awardee counts and selection criteria, disbursement timing, post-term access, and whether Gemini for Government governance features will reach commercial tiers. These are not specified in the original source, and no independent benchmark of the laboratory results was available at the time of writing.