How to read this graph
Draft 2, October 2026. Built to be revised.
Nodes are skills, not topics. The test of a well-formed node is that you can imagine a question that proves you have it or proves you don’t. “Participles” is a topic; “parse a participle in context and identify its syntactic function” is a skill.
Edges are prerequisites, and they run one way: downward. Everything a skill depends on sits above it. You cannot build D18 until the skills feeding into it are solid. Not “have been taught”; solid, at the speed where they no longer cost working memory.
The graph is useful in two directions:
- Forwards, as a map: what am I entitled to attempt next? Skills with a coloured ring are ready: every prerequisite is marked mastered.
- Backwards, as a diagnostic: this sentence defeated me; which node failed? Select a skill and the graph lights up every prerequisite chain above it. When you can’t read a sentence, don’t “review the passage.” Trace the edges upward until you hit the node that’s actually soft, and drill that.
Eleven strands, lettered A to K. They are not sequential. Several run in parallel from early on, and the interesting nodes are the ones where strands converge.
Using the page
- Click a skill to toggle it mastered. Its details open in the side panel, including a button to mark all of its prerequisites at once.
- Hover or select a skill to trace its lineage: solid dark lines lead up to its prerequisites, dashed lines lead down to what it unlocks.
- Zoomed out, the graph shows the strand overview: each strand’s name sits over its skills, with arrows showing which strands feed which. Zoom in and it gives way to the skill names.
- Pick a strand or a milestone in the panel to fade out everything else.
- Progress is saved in this browser. “Copy progress link” makes a link that carries your marks to another device or person.
- Only the direct edges are drawn. If A → B → C, the redundant A → C line is left out, though the detail panel lists every prerequisite as written.
What changed from draft 1
- Fixed a cycle. Chunking (G4 → now G5) required idiom (F7), which required unseen reading (G6 → G8), which required chunking. Chunking now rests on core vocabulary (F2) instead.
- Fixed a milestone contradiction. Reading in Greek word order was required for glossed Xenophon but depended on E20, which wasn’t due until the next milestone. Reading in Greek order (G4) now rests on case, article and preposition syntax, and word order and emphasis (E21) builds on it.
- Expanded strand-level prerequisites. Edges like “C, D”, “E”, “I”, “C1–C6”, “E1–E19” and “C-early” are now specific nodes, so the tree can be drawn and traced.
- Connected dead ends. Uppercase letters feed the Leiden conventions; iota subscript feeds the first declension; elision feeds parsing in context; numerals, adverbs and the accent nodes feed parsing and composition; aspect (D1) now feeds the present system directly; −μι verbs and verbal adjectives feed unseen reading and Attic prose.
- Filled missing edges in the verb system. The future, subjunctive, infinitives and participles now also require the middle/passive endings and the first aorist; principal parts require the aorists and the future they summarize.
- Moved particles earlier. μέν/δέ and γάρ appear in the first sentences you read, so E19 now rests on agreement and core vocabulary rather than on subordinate clauses.
- Renumbered. E19a (ἄν) is now E20 and word order is E21. Strand G gained two nodes: reading a glossed text (G3) and extensive reading (G7), so G runs G1–G10.
- Added effort weights. Each node is marked as hours, weeks or months, answering the “no weights” objection below in a rough way.
- Look a word up (A16) now needs only the alphabet, breathings and accent marks, matching the first milestone.
Where this graph is wrong
- The edges are cleaner than reality. Real learning is spiral: you meet participles, half-learn them, come back. The graph implies a crispness that doesn’t exist.
- The weights are guesses. D18 and C12 are no longer drawn as equals, but “weeks” and “months” are rough. A real version would measure them.
- Strand G is still under-specified. Fluency probably deserves twenty nodes and has ten, because it’s the least well described part of language pedagogy generally.
- The dialect strand pretends Attic is the root. Historically backwards; Homer is older. But pedagogically correct, and this is a learning graph, not a linguistic one.
- No forgetting. Nothing here models decay, which is the actual reason most people who “took Greek” can’t read Greek. Every node needs a maintenance schedule, not just an acquisition event. Unmark a node when it goes soft.
- F6 and I8 aren’t really skills. They’re more like slow changes in perception. They’re on the graph because leaving them off would misrepresent the destination.
Revise freely. The graph is a tool, not a claim.