CORE / COMPOSITE / VOCAB — the cart action-verb partition

../../runtime/deckrunner-engine-architecture.md §10 — the three verb tiers (this doc partitions Tier 2) and §7 (carts contribute via the DeckRunner interface, never to the player’s hands)
../design-bibles/verb-taxonomy-map.md — which Tier-2 spine each cart contributes (orthogonal to this doc’s which bucket each verb is)
../design-bibles/mission-type-catalog.md — the Tier-3 type layer (don’t confuse a mission type with an action verb)
../modules/ice-breaker.md §“THE TOOLKIT” — the CONS matrix this generalizes
../../gameplay-framework.md — fills gap B of its ledger
ADRs to consult: ADR-0005 (NoshAPI — what a CORE binds), ADR-0006 (cart grammar / declarations), ADR-0009 (legal-form filter), ADR-0028 (mission-contributions)

1. What this partitions (and what it doesn’t)

This partitions Tier-2 verb-tools — the cart-contributed actions an operator invokes during a capability (the right-hand context palette, DeckRunner §10). It does not touch:

Tier 1 — the Lisp primitive keys (CAR/CDR/CONS/EVAL/…), fixed firmware.
Tier 3 — mission types (PENETRATION / EXTRACTION / …); those are the mission-type catalog.

Every Tier-2 verb a cart contributes falls into exactly one of three buckets:

Bucket	What it is	On the deck
CORE	an atomic verb-tool — one irreducible change to deck/run state (≈ one NoshAPI call)	declared in the cart grammar; the archetypes you select + `EVAL`
COMPOSITE	a verb-tool that orchestrates ≥2 cores via `CONS`	built by pressing `CONS` and pairing; the matrix tools
VOCAB-only	no Fe surface, no state change — CIPHER lexicon / T9 ranking / action-bar labels	words the operator and the voice can use; never a callable form

2. The cleavage criterion — state-mutation footprint

The bucket is decided by what the verb does to the world, not by how it’s implemented or how it’s discovered. The operational test:

Does it make one irreducible change to deck/run state? → CORE Does it orchestrate two or more cores? → COMPOSITE Does it change nothing but the voice/UI? → VOCAB-only

Why this axis (and not FFI-dependency or discoverability): it is semantic and universal. It reads the same for every cart, and it makes cost and discoverability per-cart properties, not the definition (§6) — so ICE BREAKER’s consumable, surprise-combo model is one instance of the pattern, and a cart with no scarcity (NeonGrid) partitions exactly the same way. FFI-dependency corroborates (a CORE usually binds one NoshAPI primitive) but does not define; discoverability and cost are downstream.

3. The three buckets in detail

CORE

An atomic Fe verb-tool that performs one irreducible deck/run-state mutation — reveal a node, move position, defeat a lock, plant a payload, erase a trace. Declared in the cart’s grammar; available when its palette is live. A CORE typically maps to a single NoshAPI call. CORES are the archetypes — the alphabet the rest is spelled in.

COMPOSITE

A verb-tool that orchestrates two or more cores. Composition is binary CONS — you pair exactly two operands — and composites nest: a composite may itself be an operand of a further CONS, giving N-ary depth without a new operator (Lisp-native). Deeper combos cost more, because each CONS consumes its operands under the cart’s cost model (§6). Composites are declared in the cart grammar, with an optional :discoverable flag that hides a combo until the operator finds it (the ICE BREAKER surprise-combo pattern, §7).

VOCAB-only

A term with no Fe form and no state change — it flavors the CIPHER voice, seeds T9 token-prediction, or labels the action bar. It gives the operator and the voice a lexicon without inflating the Fe grammar. VOCAB-only is a lexicon layer, not a slice of the action arc — see the §8 finding.

4. Composition mechanics

Binary CONS. Each combination pairs exactly two operands. This preserves ICE BREAKER’s “pick two” tension and keeps the matrix legible.
Composites nest. A composite is itself a valid operand for a further CONS. So LOOP = CONS(GHOST, timing-window) where GHOST = CONS(MASK, JUMP) — depth-2 nesting, no new syntax. Effective N-ary combos emerge from nesting.
Cost escalates with depth. Each CONS consumes its operands per the cart’s cost model; a nested composite consumes a composite (which already cost its constituents) plus another operand. Deep combos are powerful and expensive.
Cross-cart composites are out of scope (v0.x). A composite combines cores within one cart’s set. Cross-cart combos (an ICE BREAKER core × a Black Ledger core) are deferred — Hot Swap and the opaque load-capability boundary make them a separate problem.

5. Declaration, discoverability & the script view

Declared by default. Composites are declared in the cart grammar so the engine and the legal-form filter know them.
:discoverable opt-in. A cart may flag a composite hidden-until-found; it surfaces only after the operator stumbles on the CONS pairing (ICE BREAKER’s REWRITE / SHADOW / LAUNDER). Default off — carts without surprise design don’t pay for it.
The script view. In nEmacs / the Mission Runner, a composite is a named token the operator can invoke; under the hood it expands to its CONS expression. Per the legal-form filter (ADR-0009), a composite inherits its constituents’ legality — it needs no separate filter entry, and a :discoverable composite stays out of token-prediction until found.
Privilege. Carts contribute all of this through the DeckRunner interface (DeckRunner §7) — never binding into the player’s hands directly.

6. Cost model — per-cart

The partition is universal; the cost of using a verb is the cart’s choice. ICE BREAKER consumes both operands of a CONS (scarcity → tempo tension). Black Ledger runs a move-economy. NeonGrid has no resource scarcity. None of that changes what is CORE vs COMPOSITE — it’s a property layered on top, consistent with the §2 cleavage. Per-cart cost numbers/feel are deferred to gameplay tuning.

7. Relationship to the ICE BREAKER CONS matrix

ICE BREAKER is the precedent, now generalized — it is instance #1 of this pattern:

ICE BREAKER	This partition
the 6 archetypes (CRACK / MIRROR / SPIKE / CLOAK / SIPHON / GHOST)	CORES
the 6×6 matrix tools (QUICKCOPY / PHANTOM / GRIND)	COMPOSITES (declared)
the surprise combos (REWRITE / SHADOW / LAUNDER)	COMPOSITES with `:discoverable`
”using a combo consumes both archetypes”	ICE BREAKER’s per-cart cost model (§6)

Nothing about ICE BREAKER changes; the doc just names the general shape its matrix was a specific case of.

8. The 30-verb arc, partitioned (the test set)

Applying the §2 test to the netrun runner arc. Finding: the action arc is almost entirely CORE + COMPOSITE; none of the 30 is VOCAB-only — capturing, erasing, and revealing are all mutations, so they are CORES. This refines the earlier chat proposal (which had SNIFF / PURGE / TRACE-back as vocab-only): under the state-footprint test they are CORES. VOCAB-only is reserved for the lexicon (“ghosted,” “burned,” “flatlined” as voice words), not for these actions.

#	Verb	Bucket	Composed of (if composite)	Home cart	Note
1	SCAN	CORE	—	ICEB	reveal surface state
2	PROBE	CORE	—	ICEB	active reveal + attention
3	MAP	CORE	—	ICEB	records topology (borderline — could be repeated-SCAN)
4	SNIFF	CORE	—	ICEB	deposits captured traffic (was “vocab” in chat)
5	JACK	CORE	—	framework	open the link / enter the grid
6	MASK	CORE	—	ICEB	scrub signature
7	SPOOF	CORE	—	ICEB	present forged creds → gate state
8	BURN	COMPOSITE	MASK + one-time token	ICEB	authenticate clean by spending a token
9	JUMP	CORE	—	framework	move to adjacent node
10	PIVOT	COMPOSITE	JUMP + trust-edge	ICEB	lateral move via a trusted relationship
11	GHOST	COMPOSITE	MASK + JUMP	ICEB	traverse below the logging threshold
12	LOOP	COMPOSITE	GHOST + timing-window	ICEB	nested composite — the worked nesting example
13	CRACK	CORE	—	ICEB	defeat an encryption layer
14	SLICE	CORE	—	ICEB	cut a path through ICE (folds into ICEB break-ice)
15	EXPLOIT	CORE	—	ICEB	fire a known vulnerability
16	INJECT	CORE	—	ICEB	drop a payload into a process
17	FETCH	CORE	—	ICEB	pull a specific asset
18	PARSE	CORE	—	ICEB/BL	interpret a pulled asset → buffer state
19	FORGE	CORE	—	BL	write false records
20	ECHO	COMPOSITE	SNIFF/MIRROR + replay	ICEB	replay captured creds/signals
21	SHIELD	CORE	—	ICEB	raise countermeasures
22	DECOY	CORE	—	ICEB	spawn a false target
23	TRACE-back	CORE	—	BL/DC	reveal a hostile process’s origin (no satisfier yet — gap)
24	FLIP	COMPOSITE	capture + redeploy	ICEB	turn an attacker’s tool against them
25	STASH	CORE	—	ICEB	cache data in a dead-drop node
26	DUMP	CORE	—	ICEB	commit the haul to safe storage
27	PURGE	CORE	—	ICEB	erase fingerprints (was “vocab” in chat)
28	PATCH	CORE	—	ICEB	close the holes you came in through
29	EJECT	CORE	—	ICEB	clean, sequenced unjack
30	FLATLINE	CORE	—	ICEB	emergency dump + burn the deck

Tally: 24 CORE · 6 COMPOSITE · 0 VOCAB-only (from this arc). The 6 composites — BURN, PIVOT, GHOST, LOOP, ECHO, FLIP — each genuinely orchestrate ≥2 cores; LOOP demonstrates nesting. Grammar surface stays tight: a cart ships ~20 cores and a handful of declared composites, while the operator’s mental vocabulary covers the whole arc.

9. ADRs to spawn / update (flagged, not written)

Cart-grammar ADR — the Tier-2 verb-tool declaration form (how a cart declares its CORES + COMPOSITES + :discoverable flags), and the CONS-nesting rule. Likely extends ADR-0006 (cart grammar) and the mission-contributions form (ADR-0028).
NoshAPI note — CORES bind NoshAPI primitives; confirm the surface is sufficient or enumerate additions (ADR-0005).
Legal-form filter — confirm composites inherit constituents’ legality with no new filter entries (ADR-0009).

These become downstream tasks; this design doc does not write them.

10. Open / deferred

Borderline verbs — MAP (record vs. repeated-SCAN), SLICE (folds into break-ice), BURN/GHOST (composite vs. atomic). A cart may fold a borderline CORE into a sibling core or promote it to a composite — the criterion gives the default; per-cart authoring makes the final call.
Per-cart cost numbers — consumable vs. move-economy vs. free, and the depth-cost curve.
Cross-cart composites — deferred (v0.x out of scope, §4).
VOCAB-only registry — whether the lexicon layer needs its own declaration surface (it overlaps the CIPHER vocabulary + T9 pools already contributed by carts).
TRACE-back has no cart satisfier yet — a real coverage gap (candidate: Black Ledger or Depthcharge signature analysis).