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.

Every entry in Decision Rain Library Project follows the same path: a link is saved, an AI assistant inspects it and proposes a review, and the operator decides what to do with it. The assistant never skips ahead to classification on its own, and no entry reaches the permanent library without an explicit human action. That sequence — save, inspect, propose, validate — is the entire operating model.

The Core Principle

The foundational rule of the project is this: do not catalog links; catalog decisions. A raw bookmark records only that something existed. A decision entry records judgment. For every evaluated link, the library stores:
  • Authority — where the supporting evidence comes from
  • Truth — how confident the review is, using a controlled vocabulary from truth/verified to truth/unknown
  • Practical fit — whether this is realistically usable given your tools, accounts, budget, and setup tolerance
  • Operational status — the current lifecycle state of the entry
  • Risk — what could go wrong if you act on this
  • Next action — the single concrete step that follows from the verdict
This shape makes the entry useful the second time you look at it, not just the moment you saved it.

The Five Collections

Collections are lifecycle states, not topical folders. Each collection has a specific meaning, and entries should only appear in the collection that matches their current state.
CollectionRole
00_SYSTEMCanonical governance documents only: agent contract, system spec, tag registry, decision rules, note template, golden examples, and usage flow.
00_INBOXEvery new external link starts here, no exceptions. The entry is preserved but not yet evaluated.
10_REVIEWLinks worth analysis before becoming decisions. The assistant works here; entries do not move without operator action.
20_LIBRARYEvaluated and operator-validated decision entries only. Nothing enters this collection without explicit operator approval.
90_ARCHIVERejected, outdated, duplicated, inactive, or historical entries. Kept as a traceable record, not an active working set.
The numbering is intentional. 00_ sorts before everything else so governance documents and unprocessed links are always visible at the top of the list.

The Ingestion Flow

1

New link arrives

You give the assistant a URL. It saves the original link to 00_INBOX immediately, tagging it with status/pending-review and at least one next/* tag. The trace is preserved before any analysis begins.
2

Assistant processes the entry

The assistant reads the page, repository, documentation, or available sources. It then fills the full note template — producing a Verdict, Next, Claim, Evidence, Authority, Truth, and StackFit assessment — and proposes a collection and tag set.
3

Operator validates the proposal

You review the proposed verdict, tags, and collection. You may approve, correct, or reject. This is the only moment that counts as validation. The assistant waits; it does not move the entry on its own.
4

Entry is promoted or archived

After your approval, the entry moves to 10_REVIEW for further consideration, directly to 20_LIBRARY if it is ready, or to 90_ARCHIVE if it is rejected or obsolete. The assistant updates the entry only after this explicit instruction.
The rule “never skip INBOX” is not a preference — it is a hard constraint. Skipping it allows the assistant to treat capture as classification, which collapses the signal/analysis/decision distinction the entire system depends on.

The Operator Gate

The operator is the only validation authority in this system. The AI assistant may:
  • Preserve a new link in 00_INBOX
  • Inspect the page, repository, documentation, and available evidence
  • Research official and community sources independently
  • Analyze what the link provides, what supports or weakens that claim, and what the practical fit looks like
  • Propose a verdict, tags, collection, next action, risks, and alternatives
The AI assistant may not, without explicit operator approval:
  • Mark a decision as canonical
  • Promote an entry into 20_LIBRARY
  • Archive an entry
  • Change SYSTEM documents
  • Create or use new tag values outside the approved taxonomy
  • Normalize entries in batch
  • Overwrite existing decision notes
  • Treat silence as approval
That last point matters. If you do not respond to a proposal, the entry stays where it is. Absence of objection is not a green light. Previous approval for a similar entry is not approval for this one.

Why This Matters

The separation between signal, analysis, and decision is the core safety property of the project. Collapsing any two of these levels — treating a saved link as a decision, or treating a well-formed AI proposal as validation — lets unreviewed guesses silently become canonical library state. Once an entry is in 20_LIBRARY, it looks authoritative. The operator gate exists to ensure that appearance is earned, not assumed. A clean-looking proposal is not a decision. Analysis is not validation. The library is only as trustworthy as the discipline with which the operator gate is enforced.

Build docs developers (and LLMs) love