NC Standard Course of Study · Reading Foundational Skills

Mapping the Alphabet Code

What it would take to grow a two-minute letter-sound check into something that covers the standards, knows what depends on what, and tells a teacher what to do on Tuesday.

01The short answer

The code is the easy third. What exists now covers two clarifying objectives out of forty-two kindergarten standards, and it took a day. Covering the rest of the foundational skills strand is a few weeks of evenings. The adaptive engine is a couple of weeks after that.

The hard part is not software. It is building an item bank that is actually valid, and writing the decision rules that connect a result to a move. That work is measured in months, it is teacher work rather than developer work, and skipping it produces a tool that generates numbers nobody acts on.

~1 week

Finish the strand

Letter names, phonological awareness, blending, high-frequency words. Same shell, new item types. The one genuinely new cost is that oral items have no picture — the adult has to say the sound, so the script matters more than the screen.

~2 weeks

Make it adaptive

Item scheduling, mastery states, the dependency gate, and the stopping rules. This is the interesting engineering and it is where the tool stops being a worksheet.

~3 weeks

Off one device

A server, teacher accounts, a class roster, reporting across a grade level. Entirely understood problems. Do it last, because everything above teaches you what to store.

Ongoing

Validate the bank

Does a child who passes your probe actually read? Until you have checked that against something external, the number on the screen is a guess with a decimal point.

02Segmenting the standards

Kindergarten has 42 standards across six strands and twenty clusters. The counts below come from the DPI clarifications document rather than from the Common Core, which matters: North Carolina inserts a Handwriting cluster that Common Core has no equivalent for, and every foundational code after it shifts by one.

Strand	Code	K standards	Clusters	Suits a tool like this?
Reading: Literature	`RL.K`	10	Key ideas · Craft · Integration · Range	No
Reading: Informational	`RI.K`	10	Key ideas · Craft · Integration · Range	No
Reading: Foundational	`RF.K`	5	Print concepts · Handwriting · Phonological awareness · Phonics · Fluency	Built for it
Writing	`W.K`	6	Text types & publishing · Research	No
Speaking & Listening	`SL.K`	5	Collaboration · Presentation	No
Language	`L.K`	6	Conventions · Knowledge of language · Vocabulary	Parts

The single most useful thing to notice: five of the forty-two standards are a good fit for a two-minute probe, and they are all in one strand. Foundational skills are discrete, finite, and right-or-wrong — there are about a hundred separable facts in the alphabet code and a child either produces one or does not. Comprehension, writing and discussion are none of those things. Building a tool that claims to score RL.K.3 is how ed-tech earns its reputation.

The foundational strand, in full

Code	Cluster	What the child does	Items
`RF.K.1`	Print Concepts	Directionality, word boundaries, letter–word relationship, and naming all 52 letterforms (1.d)	~56
`RF.K.2`	Handwriting NC only	Print upper- and lowercase letters	52
`RF.K.3`	Phonological Awareness	Rhyme, syllables, onset–rime, then isolating and blending phonemes in CVC words	~40
`RF.K.4`	Phonics & Word Recognition	Consonant sounds, long and short vowels, high-frequency words, minimal pairs	~70
`RF.K.5`	Fluency	Read emergent-reader texts with purpose and understanding	Not item-based

That is roughly 220 separable items for one child in one year. At three seconds an item you cannot test them all, which is the whole reason adaptivity matters: the job is choosing the twelve items that tell you the most.

03What depends on what

Standards are written as a list because documents are linear. Learning is not. The foundational skills form a directed graph, and the two most expensive mistakes in early reading both come from ignoring it: teaching above a gap, and assessing something the child has not been taught.

Amber marks the two gates. Everything in the bottom half is unreachable until phoneme blending works, and blending is oral — it has nothing to do with letters. Green is what the tool already covers, which is a narrow slice sitting on top of a dependency almost nobody tests.

Three dependencies worth designing around

Blending is the hidden gate. A child can know all twenty-six sounds and still not read a word, because turning /c/ /a/ /t/ into cat is a separate skill that lives in the ear. If you test only sounds, this child looks fine and keeps failing.
Names and sounds are not the same dependency. Letter names help but do not gate. A child who knows no names can still learn sounds. Do not hold back sound instruction waiting for the alphabet song.
Digraphs depend on their parts. Teaching sh before s and h are secure wastes the lesson. The tool should refuse to schedule a digraph whose components are not yet secure.

04The adaptive engine

Adaptive does not mean harder-when-right. In a finite, discrete domain it means choosing which items to show, and when to show them again. Two mechanisms carry almost all of the value.

Item states, and what moves between them

State	Means	Next review	On success	On failure
Unknown	Never produced	Every session	→ Learning	stays
Learning	Produced once	Next day	→ Review	→ Unknown
Review	Twice, different days	4 days	→ Secure	→ Learning
Secure	3× across ≥2 sessions, ≥3 days apart	Monthly spot-check	stays	→ Learning

The spacing is doing the work, not the repetition. Three correct answers in one sitting is short-term memory; three correct answers across three weeks is learning. Any mastery rule that can be satisfied in a single session will overstate what the child knows.

Choosing the twelve items

A session is roughly twelve items before a five-year-old is done. Spend them like this:

Seven at the edge — Unknown and Learning. This is where instruction gets its information.
Three maintaining — Review items that have come due.
Two already secure — catches decay, and more importantly the child starts and ends on something they can do.

That last one is not sentiment. A probe that is all failure teaches a child that reading is where they are bad at things, and you will pay that back for years.

Where to start, and when to stop

Entry. Start one tier below the last secure skill, not at the standard. A child secure on blending does not need every letter sound tested.
Ceiling. Five consecutive correct in a tier — stop, move up.
Floor. Three consecutive wrong — stop, drop a tier, and probe below. Never keep going to “get a complete picture”. The picture is already complete.
Budget. Cap assessment at about five per cent of instructional minutes. Every minute probing is a minute not teaching.

Record what they said, not just whether it was right. The current tool scores binary, and binary throws away the diagnosis. A child who reads d as /b/ has a different problem from one who shrugs, and a completely different lesson. Confusion pairs — b/d, p/q, m/n, f/v, and the short vowels against each other — are the most valuable data the probe can collect, and cost one extra tap.

05What the data has to decide

This is the test for every measure you add: name the move it changes. If a number does not change what happens on Tuesday morning, it is decoration, and collecting it costs a child instructional time.

What the data shows	What you do
Sounds secure, blending weak	Stop drilling sounds. Blend three phonemes orally, no print in front of them.
A specific confusion pair	Separate them in time. Teach one to secure, then contrast deliberately. Never introduce b and d in the same week.
Names known, sounds not	Sounds only. Names are not the bottleneck and more alphabet song will not help.
Neither names nor sounds, weak phonological awareness	Drop out of print entirely. Rhyme, syllables, onset–rime. Print is premature and will feel like failure.
Fewer than one new item a week for four weeks	Change the method, not the dose. More of what is not working is not the answer. Flag for intervention.
Secure items decaying between sessions	Review interval is too long for this child. Shorten it rather than reteaching.

06Build order

Add blending before anything else. It is the gate, it is cheap to test orally, and it changes the interpretation of every letter-sound result you already have.
Add letter names (RF.K.1.d). Same mechanic as sounds, doubles the coverage, and lets the tool tell names and sounds apart as separate problems.
Add the item state machine. Four states, a due date, the selection split. This is the point where it stops being a worksheet.
Add confusion capture. One extra tap, the largest single gain in diagnostic value.
Add the dependency gate so the scheduler will not offer an item whose prerequisites are not secure.
Then the server, accounts and class reporting — by which point you know exactly what needs storing.

07Six ways this goes wrong

Treating the standards list as the teaching sequence. Standards are a destination. The dependency graph is the route, and they are not the same shape.
Assessing what has not been taught. Produces noise, and teaches the child that the tool is where they fail.
Mistaking a probe for mastery. Ten items sampled from fifty-two is an estimate. Say so on the screen, or someone will put it in a cumulative file.
Over-assessing. The most common failure of progress monitoring is that it eats the instruction it was meant to direct.
Building the backend first. It is the most familiar work and the least valuable, and it locks in a schema before you know the questions.
Letting the tool decide. It surfaces patterns. A teacher who has watched the child decides what they mean. Design every screen so the last judgement is a person's.

Standard codes, cluster names and counts are taken from Kindergarten: ELA Standards, Clarifications and Glossary, NC Department of Public Instruction, adopted 2017 and implemented 2018–19. The adaptive design is ordinary spaced-retrieval practice applied to a finite item set; nothing in it is novel, which is the point.