Skip to main content

Documentation Index

Fetch the complete documentation index at: https://mintlify.com/XxYouDeaDPunKxX/decision-rain-library-project/llms.txt

Use this file to discover all available pages before exploring further.

Proposals are the primary output of the AI assistant in the Decision Rain Library Project. The assistant does not make final decisions — it prepares structured proposals for the operator to evaluate and approve. The quality of a proposal depends on the evidence gathered, the honesty of the certainty labels applied, and the care taken to surface conflicts before recommending any action.

What the Assistant Proposes

For every entry the assistant processes, it is expected to produce proposals covering all of the following where evidence supports them:
  • Interpretations — what the entry actually is and what problem it addresses
  • Classifications — proposed collection, type/*, domain/*, source/*, and scenario/* tags
  • Tags — proposed status/*, truth/*, authority/*, risk/*, stack/*, and fit/* tags
  • Verdict — a plain-language summary of the assistant’s assessment
  • Next — a proposed next/* action tag
  • Risks — adoption risks, billing risks, ecosystem risks, or trust concerns surfaced by evidence
  • Alternatives — other entries or tools that serve a similar purpose
  • Missing evidence — gaps in what the assistant was able to verify
  • Taxonomy gaps — cases where no approved tag fits and a new one should be considered

Certainty Levels

Every proposal item must be labeled with one of four certainty levels. The assistant must not present inferred conclusions as verified facts, and must not present uncertain assessments as inferred ones.
LevelMeaningExample use
verifiedConfirmed directly from official source material — docs, source code, release notes, or maintainer statementsPricing confirmed from the official pricing page; feature confirmed from the README
inferredReasonably concluded from available evidence, but not explicitly stated in a sourceStackFit inferred from documented setup steps and dependency list
uncertainEvidence is insufficient, ambiguous, or conflicting — the assistant cannot reach a reliable conclusionMaintenance status uncertain: last commit is 14 months ago with no archived notice
operator-decision-requiredThe assistant has reached the limit of what analysis can resolve — a judgment call that only the operator can makeWhether to extract the idea or test the tool depends on operator priorities
The goal is to make the operator’s decision easier, not to sound more decisive. A proposal that clearly labels what is verified, what is inferred, and what remains uncertain is more useful than one that presents everything with equal confidence.

Evidence Gathering Rules

Before the assistant may produce practical classification or adoption proposals, it must gather current evidence from both official and community sources. These two evidence streams are kept separate because they answer different questions. Official evidence shows intended behavior. It includes:
  • Documentation and README files
  • Maintainer notes and changelogs
  • Release notes and versioned specifications
  • Pricing pages and plan descriptions
  • Source code, examples, and tests
Community evidence shows real-world friction. It includes:
  • GitHub issues and discussions
  • User reports, forums, and developer communities
  • Hacker News and Reddit threads when relevant
  • Adoption signals and ecosystem activity
  • Breakage reports, pricing complaints, and setup friction reports
Official and community evidence must be kept separate in the proposal. A project’s documentation may describe a free tier; community evidence may reveal that the free tier requires a credit card or expires after 14 days. Both facts matter, and conflating them obscures the risk.

Access Failure Protocol

Sometimes the assistant cannot inspect the target well enough to form a reliable proposal. This is not a reason to guess or to pad the proposal with inferences. It is a reason to stop, report the problem clearly, and wait.
1

Record what was saved

Report the URL and any metadata that was successfully saved to 00_INBOX, including any tags already applied.
2

Report available evidence

List the evidence the assistant was able to gather before hitting the access barrier — partial docs, a cached page, a description, or community signals.
3

Report missing evidence

Describe what evidence is needed but could not be obtained — the full README, the pricing page, the source code, or the issue tracker.
4

Identify the tool or access problem

Name the specific barrier: a paywall, a login requirement, a dead link, a rate limit, a tool failure, or an access restriction.
5

State the minimum unblock step

Describe the smallest action the operator can take to unblock the research — sharing a direct link, providing credentials, or confirming that a cached version is acceptable.

Ambiguity and Conflict Handling

Not all ambiguity requires the assistant to stop. The rule depends on whether the ambiguity changes the outcome. Stop and ask the operator when ambiguity or conflicting evidence would change any of the following:
  • Collection placement
  • status/* tag
  • truth/* tag
  • StackFit assessment
  • next/* action
  • Adoption recommendation
  • Extraction vs. rejection decision
  • Promotion out of 00_INBOX
Surface the conflict before proposing whenever evidence from official and community sources contradicts each other. Community evidence can raise risk and justify a lower StackFit or a risk/* tag, but it does not automatically override official evidence unless the community signal is material, repeated, and directly relevant to the operator’s use case. Proceed with recorded uncertainty only when the ambiguity does not change the decision — and only if the uncertainty is clearly noted in the proposal rather than silently discarded.
The assistant must not classify, promote, archive, or write a full decision note from insufficient evidence.It must not infer missing substance from the entry’s title, URL, star count, popularity signals, marketing copy, model memory, or the behavior of similar projects. An entry that cannot be adequately researched must be marked with explicit gaps — not filled in with plausible-sounding estimates.

Build docs developers (and LLMs) love