We don’t write a syllabus. We compute a path.
Most curricula are ordered by subject — chapter three follows chapter two because someone put it there. Ours is a map of how skills actually rest on each other, so you can see how we decide what to teach, how we train it, and what any one idea is built from.
375 micro-topics · 529 prerequisite links · validated DAG
Node inspector
Pick a concept
Selecting a dot lights everything the idea rests on and prints the written reason for each link. Depth runs top to bottom: a concept sits below whatever holds it up. Nothing here gates you — it maps what a skill is made of.
Concept index
Systems & Tooling57
Engineering Practice30
Web Interface49
Programming Languages61
Data & Storage25
Services & APIs29
AI Systems58
Reliability & Operations66
375
Micro-topics
One teachable idea each.
529
Prerequisite edges
435 required, 94 helpful. Every one carries a written reason.
53
Domains
Grouped into 8 subjects.
16
Deepest path
35 entry points, 170 terminal concepts.
0
Cycles
Validated DAG. Nothing is its own prerequisite.
Four layers, in order
Each layer answers a different question, and most curricula only answer the last one.
The labor market picks the destination
Lightcast skill demand plus O*NET and BLS data set the target concept set. Market moves, graph gets rebuilt behind it.
The dependency graph picks the route
Every concept decomposed to one teachable idea, wired to what it rests on. Sequence falls out of the graph; nobody hand-orders a table of contents.
Gradual Release of Responsibility picks the depth
Each concept encountered three times at decreasing support — demonstrated, practiced with a spotter, performed alone. Veterans know this as crawl, walk, run.
An artifact closes the node
Nothing completes because a video ended. Each node names the thing produced and the criterion it's judged against. Trained to standard, or not trained.
Two kinds of link
Every link says how much weight it carries, and why. That is what makes the map checkable rather than a matter of opinion.
Load-bearing — solid, arrowed
It carries weight
Retrieval has nothing to search without stored vectors. Teach it the other way round and the second idea has nothing to stand on.
Supporting — dashed, open
It smooths the way
You can build it unmeasured, you just won’t know if it works.
All 529 edges carry a written sentence like that. If we can’t write the reason, the edge doesn’t go in.
What one micro-topic carries
Every node in the graph is specified to this level. These fields are what the graph is built from, not prose written after the fact.
Node T20 · sample record
Measure generation faithfulness
- Concept
- Measure generation faithfulness
- Description
- Detects answers unsupported by context
- Subject
- AI Systems
- Domain
- retrieval
- Type
- meta (of: conceptual / procedural / representational / language / meta)
- Exit depth
- scaffolded (of: guided / scaffolded / unassisted)
- Market anchor
- LLM evaluation
- Evidence criterion
- Catches a fabricated answer with a scored check
- Curriculum ref
- M19 §19.4
- Prerequisite depth
- 14 — longest chain of load-bearing links beneath it
The evidence criterion is the whole point of the record. A node does not close when a lesson is watched or a box is checked; it closes when the named artifact exists and meets the criterion. That is what an employer is buying, and it is what a funder can audit.
Type tells an instructor how to teach it. Exit depth tells them how much support to remove. Market anchor is the receipt for why the concept is in the map at all. And the curriculum ref cites the written source it came from, by module and section — Hashflag Stack Curriculum v2.0 (December 2025) — so nothing here is a claim you have to take on trust.
Tap any dot in the map above to read the same record for it.
The graph is grouped, not sequenced
Subjects say what a concept is about. They do not say when you meet it — the prerequisite edges decide that, and a single subject can run the whole length of the graph.
Systems & Tooling
57 topics · 10 domains
Engineering Practice
30 topics · 6 domains
Web Interface
49 topics · 7 domains
Programming Languages
61 topics · 4 domains
Data & Storage
25 topics · 4 domains
Services & APIs
29 topics · 5 domains
AI Systems
58 topics · 7 domains
Reliability & Operations
66 topics · 10 domains
Three questions a syllabus can't answer
A list of modules can tell you what is covered. It cannot tell you how the pieces hold each other up, which is the part you need in order to trust it or build on it.
Question 01
Where does this person actually start?
Whatever they already hold, the map shows what it reaches. A veteran who has run Linux for a decade starts from what that covers, not from week one.
Question 02
What's the shortest path to this specific job?
Name the target concept set from a real posting and the graph returns the minimum set of nodes that reaches it, in order.
Question 03
Why is this taught at all?
Every concept carries its market anchor and its evidence criterion, and every link carries its reason. You can audit the whole thing without taking our word for any of it.
Where the four layers come from
Graph structure
Marble open taxonomy
1,590 topics, 3,221 edges, ODbL 1.0. Different subject, same architecture.
Market layer
Lightcast · O*NET · BLS
Skill demand, occupational profiles and wage data set the target concept set.
Depth model
Gradual Release of Responsibility
Demonstrated, then practiced with support, then performed alone.
None of the four layers is ours alone. Assembling them and aiming the result at what employers pay veterans for is.
Retool. Retrain. Relaunch.
Free · Remote · 17 weeks · EIN 86-2122804