A graph starts from a question, and a hunch is a source

.gitignore

@@ -6,3 +6,4 @@ dist/
 .pytest_cache/
 .coverage
 .knoten.lock
+build/

README.md

@@ -188,13 +188,22 @@ your `graph.yaml` says they mean — the core checks `node_types` for membership
 nothing else. A shape that recurs:
 
 ```
-source ──▶ idea ──▶ hypothesis ──▶ experiment ──▶ finding
-             ▲                                       │
-             └────── findings open new ideas ────────┘
+question ─▶ source ─▶ idea ─▶ hypothesis ─▶ experiment ─▶ finding
+                       ▲                                     │
+                       └────── findings open new ideas ──────┘
 
 gate    stands outside the loop: the bar every claim must survive
 ```
 
+A graph begins with **one question, statement or task**, and `knoten init` scaffolds it as
+a node rather than leaving it as prose — so a finding can be traced back to the question it
+serves, and a second sub-question has somewhere to live. From there the usual move is to
+investigate **sources**: papers, blogposts, datasets, searches. Work that starts from your
+own head instead is not an exception — write `source-own-intuition` and cite it the same
+way. One rule, every time: an idea names where it came from. It also makes an
+uncomfortable question answerable — how much of this graph rests on hunches rather than on
+having read anything?
+
 The names above are a convention, not a rule — `knoten viz` lays columns out in that
 order and puts anything it does not recognise after the ones it does. (`method` is
 deliberately unused: the rename in #15 freed the word for "the approach derived from
@@ -239,7 +248,8 @@ Because it invents none, the only place your words can be *defined* is your own
 
 ```yaml
 node_types:
-  source:     external material the work starts from — a paper, dataset or search
+  question:   what this graph exists to answer — a question, a statement or a task
+  source:     where the work came from — a paper, dataset, search, or your own intuition
   idea:       what you took from a source; a direction, not yet a testable claim
   hypothesis: a falsifiable claim derived from an idea
   experiment: the test built to verify or falsify a hypothesis

SKILL.md

@@ -15,12 +15,17 @@ A research graph that remembers what did NOT work. Run `knoten` in the graph dir
 **Read it first** — knoten defines none of these words, so `hypothesis` means whatever
 that graph says it means. A common shape:
 
-    source ──▶ idea ──▶ hypothesis ──▶ experiment ──▶ finding
-                 ▲                                       │
-                 └────── findings open new ideas ────────┘
+    question ─▶ source ─▶ idea ─▶ hypothesis ─▶ experiment ─▶ finding
+                           ▲                                     │
+                           └────── findings open new ideas ──────┘
 
     gate    stands outside the loop: the bar every claim must survive
 
+A graph starts from ONE question, statement or task — `knoten init` scaffolds it, and
+everything else descends from it. Investigate sources first (papers, posts, datasets,
+searches); if the work starts from your own head instead, record that as a source too, so
+an idea always names where it came from.
+
 Those names are the order `knoten viz` lays columns out in; a type it does not know lands
 after the ones it does. Links are a list, so one hypothesis can carry several experiments
 and several findings.

SPEC.md

@@ -77,11 +77,12 @@ rather than thinking.**
 
 ### Node types (conventions, not hardcoded)
 
-`source` · `idea` · `hypothesis` · `experiment` · `finding` · `retraction` · `gate`
+`question` · `source` · `idea` · `hypothesis` · `experiment` · `finding` · `retraction` · `gate`
 
-A convention only — the core checks `node_types` for membership and nothing else. `gate`
-stands outside the loop the others form: it is the bar a claim must survive, not a stage
-it passes through. `method` is deliberately unclaimed; the rename that freed it reserved
+A convention only — the core checks `node_types` for membership and nothing else. A graph
+begins with a `question` and everything descends from it; a `source` is where the work came
+from, including the author's own intuition. `gate` stands outside the loop the others form:
+it is the bar a claim must survive, not a stage it passes through. `method` is deliberately unclaimed; the rename that freed it reserved
 it for "the approach derived from findings that survived", and nothing implements that
 yet.