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.

A decision in Decision Rain Library Project is an explicit operator action — not an AI output. The assistant proposes; you validate. Once you approve an entry, it becomes a canonical part of your library with a confirmed verdict, finalized tags, and a recorded next action. Until you approve it, it is a proposal — pending, not canonical, and not to be treated as decided.

The Operator Gate

The assistant requires your explicit approval before making any of the following changes. This list is not optional — if approval is absent, the entry stays pending and the assistant asks rather than proceeding. Operator approval is required for:
  • final collection placement
  • final status/* tag
  • final truth/* tag
  • final StackFit assessment
  • final next/* tag
  • promotion out of 00_INBOX
  • any 20_LIBRARY placement
  • archiving an entry to 90_ARCHIVE
  • changes to SYSTEM documents
  • taxonomy changes, including new tag values or families
  • batch normalization runs
  • overwriting existing decision notes
Silence is not approval. Prior similar approval is not approval. Confidence is not approval. If your explicit approval is missing, the entry stays pending. The assistant will not interpret inaction or apparent agreement as permission to proceed.

Validating a Proposal

When the assistant produces a proposal, your role is to read it critically and either confirm it, correct it, or ask for more analysis. The validation step is short in normal cases — read the proposal, check the three key fields, and give your instruction.
1

Read the proposal

Review the assistant’s proposed verdict, tags, and next action. The Verdict and Next fields appear near the top of the note; check those first.
2

Check the key fields

Ask three questions: Does the Verdict match the evidence presented? Is the Truth assessment accurate given what was found? Does StackFit reflect your real situation and constraints?
3

Correct any fields that are wrong

You may update any field directly — verdict, tags, truth, StackFit, or next action. The assistant adjusts the note to match your corrections before moving the entry.
4

Instruct the assistant where to move the entry

Tell the assistant the destination: 10_REVIEW, 20_LIBRARY, or 90_ARCHIVE. This is the approval action. Without it, the entry does not move.
5

Assistant updates and moves

After your instruction, the assistant applies the final tags, updates the note if corrections were made, and moves the entry to the collection you specified.

Collection Outcomes

Each destination collection has a distinct meaning. Placing an entry correctly is part of the decision.

10_REVIEW

The entry needs more thought. It is worth keeping visible and may warrant further research, but it is not yet fully decided. You may leave it here until context improves or a concrete need arises.

20_LIBRARY

The entry is evaluated and canonical. You have validated the verdict, tags, and next action. This is the only collection where fully decided entries live. Nothing reaches here without your explicit approval.

90_ARCHIVE

Rejected, obsolete, duplicated, or historical. The entry is kept for traceability — so you can see what was evaluated and why it was set aside — but it is not active. Nothing is silently deleted.

Handling Ambiguity and Conflict

The assistant is expected to surface ambiguity and evidence conflict rather than resolve it silently. When evidence is incomplete or sources disagree, the assistant’s job is to make the conflict visible so you can make an informed decision. When the assistant stops and asks:
  • If ambiguity changes collection, status, truth, StackFit, next action, adoption, extraction, rejection, or promotion, the assistant stops and asks before proceeding.
When evidence conflicts:
  • The assistant surfaces the conflict in the proposal before recommending a course of action.
  • Community evidence can raise risk. It does not automatically override official evidence unless the conflict is material, repeated, and relevant to your use case.
  • The assistant distinguishes what is verified, what is inferred, what is uncertain, and what requires an operator decision.
If ambiguity does not change the decision — if the outcome is the same regardless of which interpretation is correct — the assistant may proceed. But only if the uncertainty is recorded clearly in the note. Unrecorded uncertainty is not acceptable.

What Makes a Serious Entry

A fully validated library entry in 20_LIBRARY must include all of the following. Entries missing required fields should be returned for completion before promotion.
FieldRequirement
CollectionExactly one collection
status/*One status tag reflecting evaluation outcome
type/*One type tag describing what the item is
domain/*One or more domain tags
truth/*One truth tag reflecting evidence quality
fit/*Required when adoption is relevant
next/*One next-action tag — the most specific applicable action
Cataloged dateDate the entry was first cataloged
NoteComplete note in template format
An entry that reaches 20_LIBRARY without all required fields is incomplete. The validation step is the right time to catch gaps — not after the entry has been marked canonical.

Build docs developers (and LLMs) love