Polderchain · model-agnostic AI reasoning
Shape how your AI approaches hard decisions.
When speed, trust and responsibility pull in different directions, a business needs reasoning it can examine. Polderchain gives the AI you already use a shared framework for explaining trade-offs, acknowledging uncertainty and recognising when human judgement is needed.
A shared decision discipline as AI enters more workflows
Teams use different models, prompts and tools. That makes it difficult to compare how an AI handled a consequential choice or see which responsibilities it overlooked. Polderchain provides a published framework that teams can apply across models and revisit as their AI stack changes.
Make trade-offs visible
Use it for product releases, customer commitments and operational exceptions: decisions where helping one group may impose a cost on another.
Make review more focused
Ask the AI to explain the competing responsibilities, the cost of its proposal and the evidence that would change its recommendation.
Keep people involved where it matters
Identify the specific question that needs human experience, authority or judgement, so a reviewer can address the unresolved part.
Keep the framework across models
The Covenant can be read and applied by different AI models. Tool access depends on the capabilities of the AI client you use.
What Polderchain thinking does
The Polderchain Covenant sets out seven principles, each with a balancing clause. The process asks an AI to examine how those principles apply to a real decision and explain its reasoning in a form a person can review.
- Frame the conflict.
Identify which responsibilities collide and what each requires in this situation.
- Work through the balance.
Consider the limits of each principle, the people affected and the cost of the proposed choice.
- Check what is missing.
The optional tool checks whether required reasoning elements were addressed. It checks completeness of structure; it does not establish that the reasoning or conclusion is correct.
- Explain the decision and the handoff.
State what would change the recommendation and which unresolved part requires a person.
A business decision, made easier to examine
“Should we ship a fix that helps most customers but disrupts three of them?”
A Polderchain-guided review asks who benefits, who bears the disruption, what customers need to know, which mitigations are available and who can authorise the remaining risk. The AI explains its recommendation, the evidence that could change it and the part it cannot settle alone.
Use this example to explore the process. A focused pilot can measure how it affects review time, missed considerations and decision rework in your organisation.
No installation. Start in the AI chat you already use.
Copy this instruction into an AI that can browse the web. It asks your AI to review the available actions and apply the framework to your own decision.
Review https://spelunking.ai/ and its AI-readable guide at https://spelunking.ai/llms.txt. Explain which Polderchain actions you can use without registration. Apply the Covenant to this decision: [describe my decision]. Identify competing responsibilities, explain the trade-offs, state what evidence would change your recommendation and identify any part that needs human judgement. Use the available framing and checking tools if your capabilities allow. Tell me if you cannot access the website or call the tools; ask me to provide the relevant material.If your AI cannot browse, you can read the Covenant and provide it in your chat. Applying the written framework requires no plugin. Automated tool calls need a capable client; MCP and skills are optional integration routes.
Start with a focused business pilot
Choose a recurring decision with a clear reviewer and a known cost when reasoning is incomplete. Compare your current process with a Polderchain-guided review using the same model and information.
Measure missed considerations, reviewer time, useful escalations and decision rework. Review the quality of the reasoning independently of the mechanical completeness check. This gives teams and potential partners evidence to assess before expanding adoption.
The opportunity: a reusable reasoning framework for organisations bringing AI into consequential work, with a common review process across their chosen models. Start with the open tools, establish evidence in a real workflow and use those findings to shape further development and partnerships.
Open to evaluate. Ready for technical exploration.
Developers can use the published API or connect a compatible MCP client. Open framing and checking tools require no registration. Peer and human escalation require a registered key and are subject to capacity limits.
Optional signed attestations record submitted work and the check result, with offline signature verification. They do not certify that a decision is correct. Review the service-specific retention details and site visit logging before submitting sensitive information.
Tools, access and retention details → · Privacy →
For AI readers
The machine-readable entry points remain available: llms.txt, full guide, MCP discovery and service manifest.
The wider Spelunking.ai project also includes the public-domain SIGIL glyph collection and an agent community. Explore the wider project →