Evaluate a log entry for note creation
Type: kb/types/instruction.md
Use this instruction when a kb/log.md entry looks like it wants promotion into a new note. Treat the log entry as a proposal, not as evidence that a new note should exist.
Do not start drafting the note until the evaluation is complete.
When to use
- When reviewing
kb/log.md - When a synthesis or abstraction entry seems like it wants promotion
- When you are unsure whether to create a note or make a smaller change
Steps
1. Read the log entry precisely
Extract:
- the candidate claim or pattern
- the notes or sources named in the entry
- whether the entry is proposing a synthesis, abstraction, update, or gap
Rewrite the proposal in one sentence in your own words. If you cannot restate it clearly, do not create a note yet.
2. Read the full notes index first
Read kb/notes/dir-index.md before opening individual notes.
Scan the entire index, not just keyword matches. Flag notes whose descriptions suggest they may already cover:
- the same mechanism
- the same distinction
- the same failure mode
- the same design implication
This is the primary duplicate-detection step. Do not rely on the log entry's cited notes alone.
3. Load the most relevant existing notes
Read the notes and source ingests most likely to overlap with the proposal.
Prioritise:
- Notes explicitly named in the log entry
- Notes flagged from
kb/notes/dir-index.md - Any directly relevant source ingests
Read enough to answer:
- What does the KB already say?
- Which note currently gives the deepest mechanism?
- Is the proposal actually new, or just a rephrasing of existing material?
Cap at 8 notes and 4 source ingests. Load depth matters more than breadth.
4. Run the novelty tests
Reject the proposal unless it passes all of these tests.
A. Delta test
What would a future reader learn from the proposed note that they would not already get from the strongest existing note?
If the answer is "the same idea, phrased more clearly" or "a cleaner summary," do not create a note.
B. Mechanism test
Does the proposal explain why something works, fails, or matters?
If it only groups cases into phases, levels, categories, or strategies without adding causal structure, do not create a note.
C. Compression test
Could the useful content fit as:
- one paragraph in an existing note
- one added link in an ingest report
- one sharpened open question
- one line kept in
kb/log.md
If yes, prefer that over a new note.
D. Traversal test
Ask: when would a future agent load this proposed note instead of the strongest existing note?
If the existing note is always the better destination, the proposal is redundant.
E. Prediction test
Does the proposal imply at least one non-obvious consequence?
Examples of acceptable consequences:
- a failure mode to watch for
- a design choice that follows from the claim
- a boundary where one approach stops working
- a testable distinction between similar-looking cases
If the proposal does not change what a reader would do or predict, do not create a note.
5. Choose the smallest sufficient outcome
Choose exactly one outcome:
Outcome A: Reject
Use when the proposal is redundant, obvious, or weaker than existing notes.
Action:
- do not create a note
- remove the log entry if it is now resolved or misleading
Outcome B: Fold into existing notes
Use when the proposal contains one useful clarification but not enough for a standalone note.
Action:
- update the strongest existing note, open question, or footer links
Outcome C: Add links in source ingests
Use when the proposal's main value is improved graph connectivity rather than a new claim.
Action:
- update the relevant
.ingest.mdfiles with missing connections and relationship text
Outcome D: Keep in the log
Use when the proposal points at something real but the mechanism is not yet clear.
Action:
- leave the log entry in place
- optionally sharpen its wording so the unresolved question is easier to revisit
Outcome E: Create the new note
Use only when the proposal survives the novelty tests and cannot be compressed into a smaller change.
Action:
- write the note following
kb/notes/COLLECTION.md - connect it to the strongest related notes and indexes
- remove or revise the originating log entry
6. If creating a note, define the claim before writing
Before drafting, write a short note plan:
- one-sentence claim
- what existing note it most directly differs from
- what mechanism it adds
- what consequence or prediction it adds
If you cannot fill in all four, stop and choose a smaller outcome.
Red flags
These usually mean "do not create a note":
- "X, Y, and Z are really phases/levels/strategies" with no new mechanism
- the proposal is justified mainly by combining two or three already-strong notes
- the best summary is "use the right approach at the right time"
- the proposal adds vocabulary but no decision value
- every useful connection is
extendsorexemplifies, but nothing depends on the new note
Verify
Before creating a note, confirm all of the following:
- This note adds a mechanism the KB does not already have.
- This note would be a better traversal target than at least one existing note for a specific future question.
- This note cannot be replaced by a paragraph, link addition, or sharper open question.
- This note changes what a reader would predict, look for, or build.
If any line is false, do not create the note.