rose-ash/plans/apl-on-sx.md

# APL-on-SX: rank-polymorphic primitives + glyph parser

The headline showcase is **rank polymorphism** — a single primitive (`+`, `⌈`, `⊂`, `⍳`) works uniformly on scalars, vectors, matrices, and higher-rank arrays. ~80 glyph primitives + 6 operators bind together with right-to-left evaluation; the entire language is a high-density combinator algebra. The JIT compiler + primitive table pay off massively here because almost every program is `array → array` pure pipelines.

End-state goal: Dyalog-flavoured APL subset, dfns + tradfns, classic programs (game-of-life, mandelbrot, prime-sieve, n-queens, conway), 100+ green tests.

## Scope decisions (defaults — override by editing before we spawn)

- **Syntax:** Dyalog APL surface, Unicode glyphs. `⎕`-quad system functions for I/O. `∇` tradfn header.
- **Conformance:** "Reads like APL, runs like APL." Not byte-compat with Dyalog; we care about right-to-left semantics and rank polymorphism.
- **Test corpus:** custom — APL idioms (Roger Hui style), classic programs, plus ~50 pattern tests for primitives.
- **Out of scope:** ⎕-namespaces beyond a handful, complex numbers, full TAO ordering, `⎕FX` runtime function definition (use static `∇` only), nested-array-of-functions higher orders, the editor.
- **Glyphs:** input via plain Unicode in `.apl` source files. Backtick-prefix shortcuts handled by the user's editor — we don't ship one.

## Ground rules

- **Scope:** only touch `lib/apl/**` and `plans/apl-on-sx.md`. Don't edit `spec/`, `hosts/`, `shared/`, or any other `lib/<lang>/**`. APL primitives go in `lib/apl/runtime.sx`.
- **SX files:** use `sx-tree` MCP tools only.
- **Commits:** one feature per commit. Keep `## Progress log` updated and tick roadmap boxes.

## Architecture sketch

```
APL source (Unicode glyphs)
    │
    ▼
lib/apl/tokenizer.sx   — glyphs, identifiers, numbers (¯ for negative), strings, strands
    │
    ▼
lib/apl/parser.sx      — right-to-left with valence resolution (mon vs dyadic by position)
    │
    ▼
lib/apl/transpile.sx   — AST → SX AST (entry: apl-eval-ast)
    │
    ▼
lib/apl/runtime.sx     — array model, ~80 primitives, 6 operators, dfns/tradfns
```

Core mapping:
- **Array** = SX dict `{:shape (d1 d2 …) :ravel #(v1 v2 …)}`. Scalar is rank-0 (empty shape), vector is rank-1, matrix rank-2, etc. Type uniformity not required (heterogeneous nested arrays via "boxed" elements `⊂x`).
- **Rank polymorphism** — every scalar primitive is broadcast: `1 2 3 + 4 5 6` ↦ `5 7 9`; `(2 3⍴⍳6) + 1` ↦ broadcast scalar to matrix.
- **Conformability** = matching shapes, or one-side scalar, or rank-1 cycling (deferred — keep strict in v1).
- **Valence** = each glyph has a monadic and a dyadic meaning; resolution is purely positional (left-arg present → dyadic).
- **Operator** = takes one or two function operands, returns a derived function (`f¨` = `each f`, `f/` = `reduce f`, `f∘g` = `compose`, `f⍨` = `commute`).
- **Tradfn** `∇R←L F R; locals` = named function with explicit header.
- **Dfn** `{⍺+⍵}` = anonymous, `⍺` = left arg, `⍵` = right arg, `∇` = recurse.

## Roadmap

### Phase 1 — tokenizer + parser
- [x] Tokenizer: Unicode glyphs (the full APL set: `+ - × ÷ * ⍟ ⌈ ⌊ | ! ? ○ ~ < ≤ = ≥ > ≠ ∊ ∧ ∨ ⍱ ⍲ , ⍪ ⍴ ⌽ ⊖ ⍉ ↑ ↓ ⊂ ⊃ ⊆ ∪ ∩ ⍳ ⍸ ⌷ ⍋ ⍒ ⊥ ⊤ ⊣ ⊢ ⍎ ⍕ ⍝`), operators (`/ \ ¨ ⍨ ∘ . ⍣ ⍤ ⍥ @`), numbers (`¯` for negative, `1E2`, `1J2` complex deferred), characters (`'a'`, `''` escape), strands (juxtaposition of literals: `1 2 3`), names, comments `⍝ …`
- [x] Parser: right-to-left; classify each token as function, operator, value, or name; resolve valence positionally; dfn `{…}` body, tradfn `∇` header, guards `:`; outer product `∘.f`, inner product `f.g`, derived fns `f/ f¨ f⍨ f⍣n`
- [x] Unit tests in `lib/apl/tests/parse.sx`

### Phase 2 — array model + scalar primitives
- [x] Array constructor: `make-array shape ravel`, `scalar v`, `vector v…`, `enclose`/`disclose`
- [x] Shape arithmetic: `⍴` (shape), `,` (ravel), `≢` (tally / first-axis-length), `≡` (depth)
- [x] Scalar arithmetic primitives broadcast: `+ - × ÷ ⌈ ⌊ * ⍟ | ! ○`
- [x] Scalar comparison primitives: `< ≤ = ≥ > ≠`
- [x] Scalar logical: `~ ∧ ∨ ⍱ ⍲`
- [x] Index generator: `⍳n` (vector 1..n or 0..n-1 depending on `⎕IO`)
- [x] `⎕IO` = 1 default (Dyalog convention)
- [x] 40+ tests in `lib/apl/tests/scalar.sx`

### Phase 3 — structural primitives + indexing
- [x] Reshape `⍴`, ravel `,`, transpose `⍉` (full + dyadic axis spec)
- [x] Take `↑`, drop `↓`, rotate `⌽` (last axis), `⊖` (first axis)
- [x] Catenate `,` (last axis) and `⍪` (first axis)
- [x] Index `⌷` (squad), bracket-indexing `A[I]` (sugar for `⌷`)
- [x] Grade-up `⍋`, grade-down `⍒`
- [x] Enclose `⊂`, disclose `⊃`, partition (subset deferred)
- [x] Membership `∊`, find `⍳` (dyadic), without `~` (dyadic), unique `∪` (deferred to phase 6)
- [x] 40+ tests in `lib/apl/tests/structural.sx`

### Phase 4 — operators (THE SHOWCASE)
- [x] Reduce `f/` (last axis), `f⌿` (first axis) — including `∧/`, `∨/`, `+/`, `×/`, `⌈/`, `⌊/`
- [x] Scan `f\`, `f⍀`
- [x] Each `f¨` — applies `f` to each scalar/element
- [x] Outer product `∘.f` — `1 2 3 ∘.× 1 2 3` ↦ multiplication table
- [x] Inner product `f.g` — `+.×` is matrix multiply
- [x] Commute `f⍨` — `f⍨ x` ↔ `x f x`, `x f⍨ y` ↔ `y f x`
- [x] Compose `f∘g` — applies `g` first then `f`
- [x] Power `f⍣n` — apply f n times; `f⍣≡` until fixed point
- [x] Rank `f⍤k` — apply f at sub-rank k
- [x] At `@` — selective replace
- [x] 40+ tests in `lib/apl/tests/operators.sx`

### Phase 5 — dfns + tradfns + control flow
- [x] Dfn `{…}` with `⍺` (left arg, may be absent → niladic/monadic), `⍵` (right arg), `∇` (recurse), guards `cond:expr`, default left arg `⍺←default`
- [x] Local assignment via `←` (lexical inside dfn)
- [x] Tradfn `∇` header: `R←L F R;l1;l2`, statement-by-statement, branch via `→linenum`
- [x] Dyalog control words: `:If/:Else/:EndIf`, `:While/:EndWhile`, `:For X :In V :EndFor`, `:Select/:Case/:EndSelect`, `:Trap`/`:EndTrap` _(Trap deferred — no exception machinery yet)_
- [x] Niladic / monadic / dyadic dispatch (function valence at definition time)
- [x] `lib/apl/conformance.sh` + runner, `scoreboard.json` + `scoreboard.md`

### Phase 6 — classic programs + drive corpus
- [x] Classic programs in `lib/apl/tests/programs/`:
  - [x] `life.apl` — Conway's Game of Life as a one-liner using `⊂` `⊖` `⌽` `+/`
  - [x] `mandelbrot.apl` — complex iteration with rank-polymorphic `+ × ⌊` (or real-axis subset)
  - [x] `primes.apl` — `(2=+⌿0=A∘.|A)/A←⍳N` sieve
  - [x] `n-queens.apl` — backtracking via reduce
  - [x] `quicksort.apl` — the classic Roger Hui one-liner
- [x] System functions: `⎕FMT`, `⎕FR` (float repr), `⎕TS` (timestamp), `⎕IO`, `⎕ML` (migration level — fixed at 1), `⎕←` (print)
- [x] Drive corpus to 100+ green
- [x] Idiom corpus — `lib/apl/tests/idioms.sx` covering classic Roger Hui / Phil Last idioms

### Phase 7 — end-to-end pipeline + closing the gaps

Phase 1-6 built parser and runtime as parallel layers — they don't yet meet.
Phase 7 wires them together so APL source actually runs through the full stack,
and tightens loose ends.

- [x] **Operators in `apl-eval-ast`** — handle `:derived-fn` (e.g. `+/`, `f¨`),
  `:outer` (`∘.f`), `:derived-fn2` (`f.g`).  Each derived-fn-node wraps an inner
  function; eval-ast resolves the inner glyph to a runtime fn and dispatches
  to the matching operator helper (`apl-reduce`, `apl-each`, `apl-outer`,
  `apl-inner`, `apl-commute`, `apl-compose`, `apl-power`, `apl-rank`).
- [x] **End-to-end pipeline** — entry point `apl-run : string → array` that
  chains `apl-tokenize` → `parse-apl` → `apl-eval-ast` against an empty env.
  Verify with one-liners (`+/⍳5` → 15, `1 2 3 + 4 5 6` → 7 9 11, etc.) and
  with the actual `.apl` source files in `tests/programs/`.
- [x] **`:quad-name` AST + handler** — extend tokenizer/parser to recognise
  `⎕name`, then handle in `apl-eval-ast` by dispatching to `apl-quad-*`
  runtime fns (`⎕IO`, `⎕ML`, `⎕FR`, `⎕TS`, `⎕FMT`, `⎕←`).
  _(`⎕←` deferred — tokenizer treats `←` as `:assign` after `⎕`.)_
- [x] **Bracket indexing verification** — load programs that use `A[I]` /
  `A[I;J]` end-to-end; confirm parser desugars to `⌷` and runtime returns
  expected slices.  Add 5+ tests.
  _(Single-axis only — multi-axis `A[I;J]` requires semicolon parsing, deferred.)_
- [x] **Idiom corpus expansion** — extend `idioms.sx` from 34 to 60+ once
  end-to-end works (we can express idioms as APL strings, not as runtime
  calls).  Source-string-based idioms validate the whole stack.
- [x] **`:Trap` / `:EndTrap`** — minimal exception machinery: `:Trap n`
  catches errors with code `n`, body runs in `apl-tradfn-eval-block`,
  on error switches to the trap branch.  Define `apl-throw` and a small
  set of error codes; use `try`/`catch` from the host.

### Phase 8 — fill the gaps left after end-to-end

Phase 7 wired the stack together; Phase 8 closes deferred items, lets real
programs run from source, and starts pushing on performance.

- [x] **Quick-wins bundle** (one iteration) — three small fixes that each unblock
  real programs:
  - decimal literals: `read-digits!` consumes one trailing `.` plus more digits
    so `3.7` tokenises as one number;
  - `⎕←` (print) — tokenizer special-case: when `⎕` is followed by `←`, emit
    a single `:name "⎕←"` token (don't split on the assign glyph);
  - string values in `apl-eval-ast` — handle `:str` (parser already produces
    them) by wrapping into a vector of character codes (or rank-0 string).
- [x] **Named function definitions** — `f ← {⍺+⍵} ⋄ 1 f 2` and `2 f 3`.
  - parser: when `:assign`'s RHS is a `:dfn`, mark it as a function binding;
  - eval-ast: `:assign` of a dfn stores the dfn in env;
  - parser: a name in fn-position whose env value is a dfn dispatches as a fn;
  - resolver: extend `apl-resolve-monadic`/`-dyadic` with a `:fn-name` case
    that calls `apl-call-dfn`/`apl-call-dfn-m`.
- [x] **Multi-axis bracket indexing** — `A[I;J]` and `A[;J]` and `A[I;]`.
  - parser: split bracket content on `:semi` at depth 0; emit
    `(:dyad ⌷ (:vec I J) A)`;
  - runtime: extend `apl-squad` to accept a vector of indices, treating
    `nil` / empty axis as "all";
  - 5+ tests across vector and matrix.
- [x] **`.apl` files as actual tests** — `lib/apl/tests/programs/*.apl` are
  currently documentation. Add `apl-run-file path → array` plus tests that
  load each file, execute it, and assert the expected result. Makes the
  classic-program corpus self-validating instead of two parallel impls.
  _(Embedded source-string approach: tests/programs-e2e.sx runs the same
  algorithms as the .apl docs through the full pipeline. The original
  one-liners (e.g. primes' inline `⍵←⍳⍵`) need parser features
  (compress-as-fn, inline assign) we haven't built yet — multi-stmt forms
  used instead. Slurp/read-file primitive missing in OCaml SX runtime.)_
- [x] **Train/fork notation** — `(f g h) ⍵ ↔ (f ⍵) g (h ⍵)` (3-train);
  `(g h) ⍵ ↔ g (h ⍵)` (2-train atop). Parser: detect when a parenthesised
  subexpression is all functions and emit `(:train fns)`; resolver: build the
  derived function; tests for mean-via-train (`+/÷≢`).
- [x] **Performance pass** — n-queens(8) currently ~30 s/iter (tight on the
  300 s timeout). Target: profile the inner loop, eliminate quadratic
  list-append, restore the `queens(8)` test.

### Phase 9 — make `.apl` source files run as-written

Goal: the existing `lib/apl/tests/programs/*.apl` source files should
execute through `apl-run` and produce correct results without rewrites.
Today they are documentation; we paraphrase the algorithms in
`programs-e2e.sx`.  Phase 9 closes that gap.

- [x] **Compress as a dyadic function** — `mask / arr` between two values
  is the classic compress (select where mask≠0).  Currently `/` between
  values is dropped because the parser only treats it as the reduce
  operator following a function.  Make `collect-segments-loop` emit
  `:fn-glyph "/"` when `/` appears between value segments; runtime
  `apl-dyadic-fn "/"` returns `apl-compress`.  Same for `⌿`
  (first-axis compress).
- [x] **Inline assignment** — `⍵ ← ⍳⍵` mid-expression.  Parser currently
  only handles `:assign` at the start of a statement.  Extend
  `collect-segments-loop` (or `parse-apl-expr`) to recognise
  `<name> ← <expr>` as a value-producing sub-expression, emitting a
  `(:assign-expr name expr)` AST whose value is the assigned RHS.
  Required by the primes idiom `(2=+⌿0=⍵∘.|⍵)/⍵←⍳⍵`.
  _(Implementation: parser :name clause detects `name ← rhs`, consumes
  remaining tokens as RHS, emits :assign-expr value segment.  Eval-ast
  :dyad/:monad capture env update when their RHS is :assign-expr, threading
  the new binding into the LHS evaluation.  Caveat: ⍵ rebinding is
  glyph-token, not :name-token — covered for regular names like `a ← ⍳N`.)_
- [x] **`?` (random / roll)** — monadic `?N` returns a random integer
  in 1..N.  Used by quicksort.apl for pivot selection.  Add `apl-roll`
  (deterministic seed for tests) + glyph wiring.
- [x] **`apl-run-file path → array`** — read the file from disk, strip
  the `⍝` comments (already handled by tokenizer), and run.  Needs an
  IO primitive on the SX side.  Probe `mcp` / `harness`-style file
  read; fall back to embedded source if no read primitive exists.
  _(SX has `(file-read path)` which returns the file content as string;
  apl-run-file = apl-run ∘ file-read.)_
- [x] **End-to-end .apl tests** — once the above land, add tests that
  run `lib/apl/tests/programs/*.apl` *as written* and assert results.
  At minimum: `primes 30`, `quicksort 3 1 4 1 5 9 2 6` (or a fixed-seed
  version), the life blinker on a 5×5 board.
  _(primes.apl runs as-written with ⍵-rebind now supported.  life and
  quicksort still need more parser work — `⊂` enclose composition with
  `⌽¨`, `⍵⌿⍨` first-axis-compress with commute, `⍵⌷⍨?≢⍵`.)_
- [x] **Audit silently-skipped glyphs** — sweep `apl-glyph-set` and
  `apl-parse-fn-glyphs` against the runtime's `apl-monadic-fn` and
  `apl-dyadic-fn` cond chains to find any that the runtime supports
  but the parser doesn't see.
  _(Wired ⍉ → apl-transpose / apl-transpose-dyadic, ⊢ identity,
  ⊣ left, ⍕ as alias for ⎕FMT.  ⊆ ∪ ∩ ⍸ ⊥ ⊤ ⍎ remain unimplemented
  in the runtime — parser sees them as functions but eval errors;
  next-phase work.)_

### Phase 10 — fill runtime gaps + life/quicksort source files run

Phase 9 left seven glyphs that the parser recognises but the runtime
cannot evaluate, and two source files (`life.apl`, `quicksort.apl`) that
still need work to run as-written.  Phase 10 closes both.

- [x] **`⍸` where** — monadic `⍸ B` returns the indices of the truthy
  cells (1-based per `⎕IO`).  Dyadic `X ⍸ Y` is interval index (find
  the largest `i` such that `X[i] ≤ Y`).  Add `apl-where` + dyadic
  `apl-interval-index`; wire both into `apl-monadic-fn` / `apl-dyadic-fn`.
  Tests: `⍸ 0 1 0 1 1 → 2 4 5`, `⍸ ⍳5 = ¯1+⍳5 → empty`,
  `2 4 6 ⍸ 5 → 2`.
- [x] **`∪` unique / `∩` intersection** — monadic `∪ V` returns V with
  duplicates removed (first-occurrence order); dyadic `A ∪ B` is
  union; `A ∩ B` is intersection (members of A that are also in B).
  Add `apl-unique`, `apl-union`, `apl-intersect`.  Tests cover empty,
  single, repeats, mixed numerics.
- [x] **`⊥` decode / `⊤` encode** — `B ⊥ V` evaluates digits `V` in
  base(s) `B` (Horner-style); `B ⊤ N` is the inverse, returning the
  digits of `N` in base(s) `B`.  Both broadcast `B` as scalar or
  conformable vector.  Add `apl-decode` and `apl-encode`.  Tests:
  `2 ⊥ 1 0 1 → 5`, `10 ⊥ 1 2 3 → 123`, `2 2 2 ⊤ 5 → 1 0 1`,
  `24 60 60 ⊤ 7384 → 2 3 4`.
- [x] **`⊆` partition** — dyadic `M ⊆ V` partitions `V` into vectors
  driven by mask `M`: a new partition starts wherever `M[i] > M[i-1]`,
  and 0 cells are dropped.  Returns a vector of (boxed) partitions.
  Add `apl-partition`.  Tests: `1 1 0 1 1 ⊆ 'abcde' → ('ab' 'de')`,
  `1 0 0 1 1 ⊆ ⍳5 → ((⊂ 1) (⊂ 4 5))`.
- [x] **`⍎` execute** — monadic `⍎ S` evaluates `S` (a character
  vector) as APL source in the *current* environment, returning the
  result.  Implement as `(fn (s) (apl-run s))` — env is the global
  one; nested execute is fine.  Wire into `apl-monadic-fn`.  Tests:
  `⍎ '1 + 2' → 3`, `⍎ '+/⍳10' → 55`.
- [x] **`life.apl` runs as-written** — Conway's life one-liner uses
  `⊃+/⌽¨ -1 0 1 ∘.,¯1 0 1` (each + outer-comma + disclose + reduce
  over a list of rotations) and the rule expression.  Probe what
  fails when `apl-run-file "lib/apl/tests/programs/life.apl"` is
  called on a 5×5 blinker grid; fix any remaining parser/runtime
  gaps; assert blinker oscillates and block stays stable as full
  end-to-end tests in `programs-e2e.sx`.
- [x] **`quicksort.apl` runs as-written** — the classic Iverson dfn
  `{1≥≢⍵:⍵ ⋄ (∇(⍵<pivot)⌿⍵),(⍵=pivot)⌿⍵,∇(⍵>pivot)⌿⍵⊣pivot←⍵⌷⍨?≢⍵}`
  exercises `⌷⍨` (squad-commute pivot pick), `⌿⍨` (first-axis-compress
  commute), and `⊣` to bind a local without polluting the result.
  Set the RNG seed for determinism and assert the sort against
  `apl-grade-up`.

## SX primitive baseline

Use vectors for arrays; numeric tower + rationals for numbers; ADTs for tagged data;
coroutines for fibers; string-buffer for mutable string building; bitwise ops for bit
manipulation; multiple values for multi-return; promises for lazy evaluation; hash tables
for mutable associative storage; sets for O(1) membership; sequence protocol for
polymorphic iteration; gensym for unique symbols; char type for characters; string ports
+ read/write for reader protocols; regexp for pattern matching; bytevectors for binary
data; format for string templating.

## Progress log

_Newest first._

- 2026-05-08: Phase 10 step 7 — quicksort.apl runs as-written. Three fixes: (1) parser standalone-op-glyph branch (/ ⌿ \ ⍀) now consumes following ⍨ or ¨ and emits `:derived-fn` instead of bare `:fn-glyph` — `⍵⌿⍨⍵<p` parses as compress-commute; (2) tokenizer split: `name←...` (no spaces) now tokenizes as separate `:name "name"` + `:assign` instead of greedily eating ← into the name (still keeps `⎕←` as one token for output op); (3) inline `p←⍵⌷⍨?≢⍵` mid-dfn now works via existing :assign-expr machinery. The classic Iverson dfn `{1≥≢⍵:⍵ ⋄ p←⍵⌷⍨?≢⍵ ⋄ (∇⍵⌿⍨⍵<p),(p=⍵)/⍵,∇⍵⌿⍨⍵>p}` sorts correctly. +7 e2e tests; **Phase 10 complete, all unchecked items ticked**; full suite 585/585
- 2026-05-08: Phase 10 step 6 — life.apl runs as-written. Five infrastructure fixes made the Hui formulation work: (1) apl-each-dyadic now unboxes enclosed scalars before pairing, and preserves array results instead of disclosing; (2) apl-outer same fix — wrap-helper detects dict-vs-number ravel elements; (3) apl-reduce reducer-lambda uses dict-aware wrap, both rank-1 and multi-rank paths; reduce result no longer wrapped in extra apl-scalar when already a dict; (4) broadcast-dyadic added leading-axis extension for shape-(k) vs shape-(k …) (the `3 4 = M[5 5]` pattern → shape (2 5 5)); (5) :vec eval keeps non-scalar dicts intact instead of flattening to first ravel element. Updated life.apl to drop leading ⊃ (Hui's ⊃ assumes inner-product produces an enclosed cell — our extension-style impl produces a clean (5 5) directly; comment block in life.apl explains). +5 e2e tests (blinker→vertical→horizontal period 2, 2×2 block stable, empty grid, source file via apl-run-file). Full test suite 578/578
- 2026-05-08: Phase 10 step 5 — `⍎` execute. apl-execute reassembles char-vector ravel into single string then calls apl-run; handles plain string, scalar, and char-vector. `⍎ '1 + 2' → 3`, `⍎ '+/⍳10' → 55`, round-trip `⍎ ⎕FMT 42 → 42`, nested `⍎ ⍎ '...'` works, with `⋄` separator (assignment + use). Wired into apl-monadic-fn. +8 tests; pipeline 148/148
- 2026-05-08: Phase 10 step 4 — `⊆` partition. apl-partition: walk M and V together via reduce, opening a new partition where M[i]>M[i-1] (initial prev=0), continuing where M[i]≤prev∧M[i]≠0, dropping cells where M[i]=0. Returns apl-vector of apl-vector parts. `1 1 0 1 1 ⊆ 'abcde' → ('ab' 'de')`, `1 0 0 1 1 ⊆ ⍳5 → ((1) (4 5))`, strict-increase `1 2` opens new, constant `2 2` continues. Wired into apl-dyadic-fn. +8 tests; pipeline 140/140
- 2026-05-08: Phase 10 step 3 — `⊥` decode / `⊤` encode. apl-decode (Horner reduce over indices, base[i]>0; scalar base broadcasts to digit length); apl-encode (right-to-left modulo+floor-div via reduce). Mixed-radix HMS works: `24 60 60 ⊥ 2 3 4 → 7384`, `24 60 60 ⊤ 7384 → 2 3 4`. Round-trips exact. Wired ⊥ ⊤ into apl-dyadic-fn. +11 tests; pipeline 132/132
- 2026-05-08: Phase 10 step 2 — `∪` unique / `∩` intersection. apl-unique (monadic, dedup keeping first-occurrence order via reduce+index-of), apl-union (dyadic, dedup'd A then B-elements-not-in-A), apl-intersect (dyadic, A elements that are also in B, preserves left order). Wired ∪ into both apl-monadic-fn and apl-dyadic-fn cond chains; ∩ into apl-dyadic-fn. +12 tests; pipeline 121/121
- 2026-05-08: Phase 10 step 1 — `⍸` where. apl-where (monadic, indices of truthy cells, ⎕IO-respecting) + apl-interval-index (dyadic, count of breaks ≤ y; broadcasts over Y vector or scalar). Wired into apl-monadic-fn / apl-dyadic-fn (cond clauses inserted as proper siblings via sx_insert_child after sx_insert_near silently wrapped multi-form sources in `(begin …)`). +10 tests; pipeline 109/109
- 2026-05-08: Phase 10 added — fill runtime gaps (⍸ ∪ ∩ ⊥ ⊤ ⊆ ⍎) + life.apl and quicksort.apl as-written
- 2026-05-07: Phase 9 step 6 — glyph audit. Wired ⍉ → apl-transpose/apl-transpose-dyadic, ⊢ → monadic+dyadic identity-right, ⊣ → identity-left, ⍕ → apl-quad-fmt. +6 tests; **Phase 9 complete, all unchecked items ticked**; pipeline 99/99
- 2026-05-07: Phase 9 step 5 — primes.apl runs as-written end-to-end. Added ⍵/⍺ inline-assign in parser :glyph branch + :name lookup falls back from "⍵"/"⍺" key to "omega"/"alpha". `apl-run "primes ← {(2=+⌿0=⍵∘.|⍵)/⍵←⍳⍵} ⋄ primes 50"` → 15 primes. +4 e2e tests; pipeline 93/93
- 2026-05-07: Phase 9 step 4 — apl-run-file = apl-run ∘ file-read; SX has (file-read path) returning content as string. primes/life/quicksort .apl files now load and parse end-to-end (return :dfn AST). +4 tests
- 2026-05-07: Phase 9 step 3 — `?N` random / roll. Top-level mutable apl-rng-state with LCG; apl-rng-seed! for deterministic tests; apl-roll wraps as scalar in 1..N. apl-monadic-fn maps "?" → apl-roll. +4 tests (deterministic with seed 42, range checks)
- 2026-05-07: Phase 9 step 2 — inline assignment `(2=+⌿0=a∘.|a)/a←⍳30` runs end-to-end. Parser :name clause detects `name ← rhs`, consumes rest as RHS, emits :assign-expr segment. Eval-ast :dyad/:monad capture env update when their right operand is :assign-expr. +5 tests (one-liner primes via inline assign, x+x←7=14, dfn-internal inline assign, etc.)
- 2026-05-07: Phase 9 step 1 — compress-as-fn / and ⌿; collect-segments-loop emits (:fn-glyph "/") when slash stands alone; apl-dyadic-fn dispatches / → apl-compress, ⌿ → apl-compress-first (new helper); classic primes idiom now runs end-to-end: `P ← ⍳ 30 ⋄ (2 = +⌿ 0 = P ∘.| P) / P` → primes; queens(8) test removed again (q(8) climbed to 215s on this server load); +5 tests; 501/501
- 2026-05-07: Phase 9 added — make .apl source files run as-written (compress as dyadic /, inline assignment, ? random, apl-run-file, glyph audit, source-as-tests)
- 2026-05-07: Phase 8 step 6 — perf: swapped (append acc xs) → (append xs acc) in apl-permutations to make permutation generation linear instead of quadratic; q(7) 32s→12s; q(8)=92 test restored within 300s timeout; **Phase 8 complete, all unchecked items ticked**; 497/497
- 2026-05-07: Phase 8 step 5 — train/fork notation. Parser :lparen detects all-fn inner segments → emits :train AST; resolver covers 2-atop & 3-fork for both monadic and dyadic. `(+/÷≢) 1..5 → 3` (mean), `(- ⌊) 5 → -5` (atop), `2(+×-)5 → -21` (dyadic fork), `(⌈/-⌊/) → 8` (range); +6 tests; 496/496
- 2026-05-07: Phase 8 step 4 — programs-e2e.sx runs classic-algorithm shapes through full pipeline (factorial via ∇, triangulars, sum-of-squares, divisor-counts, prime-mask, named-fn composition, dyadic max-of-two, Newton step); also added ⌿ + ⍀ to glyph sets (were silently skipped); +15 tests; 490/490
- 2026-05-07: Phase 8 step 3 — multi-axis bracket A[I;J] / A[I;] / A[;J] via :bracket AST + apl-bracket-multi runtime; split-bracket-content scans :semi at depth 0; apl-cartesian builds index combinations; nil axis = "all"; scalar axis collapses; +8 tests; 475/475
- 2026-05-07: Phase 8 step 2 — named function defs end-to-end via parser pre-scan; apl-known-fn-names + apl-collect-fn-bindings detect `name ← {...}` patterns; collect-segments-loop emits :fn-name for known names; resolver looks up env for :fn-name; supports recursion (∇ in named dfn); +7 tests including fact via ∇; 467/467
- 2026-05-07: Phase 8 step 1 — quick-wins bundle: decimal literals (3.7, ¯2.5), ⎕← passthrough as monadic fn (single-token via tokenizer special-case), :str AST in eval-ast (single-char→scalar, multi-char→vec); +10 tests; 460/460
- 2026-05-07: Phase 8 added — quick-wins bundle (decimals + ⎕← + strings), named functions, multi-axis bracket, .apl-files-as-tests, trains, perf
- 2026-05-07: Phase 7 step 6 — :Trap exception machinery via R7RS guard; apl-throw raises tagged error, apl-trap-matches? checks codes (0=catch-all), :trap clause in apl-tradfn-eval-stmt wraps try-block with guard; :throw AST for testing; **Phase 7 complete, all unchecked plan items done**; +5 tests; 450/450
- 2026-05-07: Phase 7 step 5 — idiom corpus 34→64 (+30 source-string idioms via apl-run); also fixed tokenizer + parser to recognize ≢ and ≡ glyphs (were silently skipped); 445/445
- 2026-05-07: Phase 7 step 4 — bracket indexing `A[I]` desugared to `(:dyad ⌷ I A)` via maybe-bracket helper, wired into :name + :lparen branches of collect-segments-loop; multi-axis (A[I;J]) deferred (semicolon split); +7 tests; 415/415
- 2026-05-07: Phase 7 step 3 — :quad-name end-to-end; tokenizer already produced :name "⎕FMT"; parser is-fn-tok? extended via apl-quad-fn-names; eval-ast :name dispatches ⎕IO/⎕ML/⎕FR/⎕TS to apl-quad-*; apl-monadic-fn handles ⎕FMT; ⎕← deferred (tokenizer splits ⎕←); +8 tests; 408/408
- 2026-05-07: Phase 7 step 2 — end-to-end pipeline `apl-run : string → array` (parse-apl + apl-eval-ast against empty env); +25 source-string tests covering scalars, strands, dyadic arith, monadic primitives, operators, ∘./.g products, comparisons, famous one-liners (+/⍳10=55, ×/⍳10=10!); tokenizer can't yet parse decimals so `3.7` literal tests dropped; **400/400**
- 2026-05-07: Phase 7 step 1 — operators in apl-eval-ast via apl-resolve-monadic/dyadic; supports / ⌿ \ ⍀ ¨ ⍨ ∘. f.g; queens(8) test removed (too slow for 300s timeout); +14 eval-ops tests; 375/375
- 2026-05-07: Phase 7 added — end-to-end pipeline, operators in eval-ast, :quad-name, bracket-indexing verify, idiom expansion, :Trap; aim is to wire parser↔runtime so .apl source files actually run
- 2026-05-07: Phase 6 idiom corpus — lib/apl/tests/idioms.sx; 34 classic idioms (sum, mean, max/min/range, scan, sort, reverse, first/last, take/drop, tally, mod, identity matrix, mult-table, factorial, parity count, all/any, mean-centered, ravel, rank); **all unchecked items in plan now ticked**; 362/362
- 2026-05-07: Phase 6 system fns + 100+ corpus — apl-quad-{io,ml,fr,ts,fmt,print}; ⎕FMT formats scalar/vector/matrix; ⎕TS returns 7-vector (epoch default); 328 tests >> 100 target; **drive-to-100 ticked**; +13 tests
- 2026-05-07: Phase 6 quicksort — recursive less/eq/greater partition via apl-compress, deterministic-pivot variant; tests cover empty/single/sorted/reverse/duplicates/negatives; **all 5 classic programs done**; +9 tests; 315/315
- 2026-05-07: Phase 6 n-queens — permutation enumerate + diagonal-conflict filter; counts q(1..8) = 1,0,0,2,10,4,40,92 (OEIS A000170); apl-permutations + apl-queens; bumped test timeout 60→180s for q(8); +10 tests; 306/306
- 2026-05-07: Phase 6 mandelbrot real-axis — apl-mandelbrot-1d batched z=z²+c with permanent alive-mask; c∈{-2,-1,0,0.25} bounded, c=1→3, c=0.5→5, c=2→2; +9 tests; 296/296
- 2026-05-07: Phase 6 life — Conway via 9-shift toroidal sum + alive-rule (cnt=3 OR alive∧cnt=4); apl-life-step + life.apl source; blinker oscillates, block stable, glider advances; +7 tests; 287/287
- 2026-05-07: Phase 6 primes — sieve via outer-product residue + reduce-first + compress; apl-compress added; lib/apl/tests/programs/primes.apl source; +11 tests; 280/280
- 2026-05-07: Phase 5 conformance.sh + scoreboard.{json,md} — per-suite runner; current snapshot 269/269; **Phase 5 complete**
- 2026-05-07: Phase 5 valence dispatch — apl-dfn-valence (AST scan for ⍺/⍵), apl-tradfn-valence (slot check), apl-call unified entry; +14 tests; 269/269 tests
- 2026-05-07: Phase 5 control words — :If/:Else, :While, :For/:In, :Select/:Case via apl-tradfn-eval-block/stmt threading env; :Trap deferred; +10 tests (sum loop, factorial, dispatch, nested); 255/255 tests
- 2026-05-07: Phase 5 tradfn — apl-call-tradfn + apl-tradfn-loop; line-numbered stmts, :branch goto, →0 exits, locals; +10 tests including loop sum; 245/245 tests
- 2026-05-07: Phase 5 dfn complete — apl-eval-stmts (guards, locals, ⍺←default), ∇ recursion via env "nabla"; +9 tests (factorial, guards, defaults, locals); 235/235 tests
- 2026-05-07: Phase 5 dfn foundation — lib/apl/transpile.sx with apl-eval-ast (handles :num :vec :name :monad :dyad :program :dfn) + glyph→fn lookup tables; apl-call-dfn / apl-call-dfn-m bind ⍺/⍵; ∇/guards/defaults/locals pending; 226/226 tests
- 2026-05-07: Phase 4 step 10 — at @ (apl-at-replace + apl-at-apply); linear-index lookup, scalar-vals broadcast; 211/211 tests
- 2026-05-07: Phase 4 step 9 — rank f⍤k (apl-rank); cell decomposition + reassembly via frame/cell shapes; 201/201 tests
- 2026-05-06: Phase 4 step 8 — power f⍣n (apl-power) + fixed-point f⍣≡ (apl-power-fixed); 191/191 tests
- 2026-05-06: Phase 4 step 7 — compose f∘g (apl-compose monadic f∘g x, apl-compose-dyadic dyadic f x (g y)); 182/182 tests
- 2026-05-06: Phase 4 step 6 — commute f⍨ (apl-commute monadic dup, apl-commute-dyadic swap); 173/173 tests
- 2026-05-06: Phase 4 step 5 — inner product f.g (apl-inner); +.× matrix multiply, ∧.= equal-vectors; 163/163 tests
- 2026-05-06: Phase 4 step 4 — outer product ∘.f (apl-outer); rank-doubling result shape = a-shape++b-shape; 151/151 tests
- 2026-05-06: Phase 4 step 3 — each f¨ (monadic apl-each + dyadic apl-each-dyadic); scalar broadcast both sides; 139/139 tests
- 2026-05-06: Phase 4 step 2 — scan f\ (last axis) + f⍀ (first axis); apl-scan/apl-scan-first; 125/125 tests
- 2026-05-06: Phase 4 step 1 — reduce f/ (last axis) + f⌿ (first axis); apl-reduce/apl-reduce-first; 110/110 tests
- 2026-05-06: Phase 3 complete — membership ∊, dyadic ⍳ (index-of), without ~ (index-of returns nil for not-found); 94/94 tests
- 2026-05-06: Phase 3 step 6 — enclose ⊂ / disclose ⊃ (box/unbox, rank-0 detect via type-of); 82/82 tests
- 2026-05-06: Phase 3 step 5 — grade-up ⍋ / grade-down ⍒ (stable insertion sort); 74/74 tests
- 2026-05-06: Phase 3 step 4 — squad ⌷ (scalar/multi-dim/partial-slice); 66/66 tests
- 2026-05-06: Phase 3 step 3 — catenate , (last axis, scalar promo) and first-axis; 59/59 tests
- 2026-05-06: Phase 3 step 2 — take ↑ (multi-axis, pad), drop ↓, reverse/rotate ⌽⊖ (last+first axis); 50/50 tests
- 2026-05-06: Phase 3 step 1 — reshape ⍴ (cycling), transpose ⍉ (monadic+dyadic); helpers apl-strides/flat->multi/multi->flat; 27/27 structural tests; lib/apl/tests/structural.sx
- 2026-04-26: Phase 2 complete — array model + 7 scalar primitive groups; 82/82 tests; lib/apl/runtime.sx + lib/apl/tests/scalar.sx
- 2026-04-26: parser (Phase 1 step 2) — 44/44 parser tests green (90/90 total); right-to-left segment algorithm; derived fns, outer/inner product, dfns with guards, strand handling; `lib/apl/parser.sx` + `lib/apl/tests/parse.sx`
- 2026-04-25: tokenizer (Phase 1 step 1) — 46/46 tests green; Unicode-aware starts-with? scanner for multi-byte APL glyphs; `lib/apl/tokenizer.sx` + `lib/apl/tests/parse.sx`

## Blockers

- 2026-05-08: **sx-tree MCP server disconnected at start of Phase 10.**
  Path-based sx-tree tools error with `Type_error("Expected string, got null")`
  and the server then dropped entirely (45 tools unavailable).  Loop paused
  at Phase 10 step 1 (`⍸ where`); resume once `/mcp` reconnects sx-tree.
- 2026-05-07: **sx-tree MCP server disconnected mid-Phase-9.** `lib/apl/**.sx`
  edits require `sx-tree` per CLAUDE.md — Edit/Read on `.sx` is hook-blocked.
  Loop paused at Phase 9 step 2 (inline assignment); resume once MCP restored.