Chunk range/map/filter to match JVM Clojure

range, map, and filter were fully element-by-element lazy, so
(map f (range 1 50)) realized one element per first/nth where JVM
Clojure realizes a whole 32-element chunk. range is a chunked
LongRange on the JVM and map/filter are chunk-preserving, so the
observable side-effect timing differed.

Following clojure.lang.LongRange, ChunkedCons, ChunkBuffer and
core.clj, this adds a crest field to the cseq record and a
cseq-chunked constructor modeling ChunkedCons (a standalone chunk
pvec, an offset, and the after-chunk seq). The chunk accessors move
to seq.ss next to the representation they read. map/filter/remove
take a chunked branch when the source is chunked, realizing the whole
chunk and chunk-cons'ing it onto a lazy rest, so their output is
itself chunked and chained transforms each batch by 32. Bounded range
is now an eager chunked seq, and the reduce fast path flows through a
ChunkedCons rest. The chunk-buffer/chunk/chunk-cons builder API in
natives-array.ss now produces a real ChunkedCons.

Single-arg (range), multi-coll map, and plain lazy seqs stay
element-by-element, like the JVM.

Adds a lazy / chunking suite to the corpus that observes realization
timing via an atom counter: first over a chunked map realizes 32,
crossing a chunk boundary realizes 49, chained maps batch [32 32],
filter applies the predicate to the whole first block, and a plain
lazy seq still realizes one element at a time. Two cases that
documented the old over-laziness now assert the JVM value of 32 and
were dropped from the allowlist. certify against JVM Clojure 1.12.3
reports 0 new and 0 stale divergences.

host/chez/java/natives-array.ss

@@ -91,51 +91,18 @@
   (let ((s (bitwise-and (exact (floor x)) #xffff))) (if (>= s #x8000) (- s #x10000) s)))
 
 ;; --- chunked seqs -----------------------------------------------------------
-;; A vector's seq is a REAL chunked-seq: (seq v) carries its backing vector +
+;; The chunked-seq accessors (chunked-seq? / chunk-first / chunk-rest / chunk-next)
-;; element index (seq.ss cseq-vec), so chunked-seq? is true and chunk-first hands
+;; live in seq.ss with the cseq core they read; here we only bind them plus the
-;; out a 32-element block (a pvec slice) while chunk-rest is the seq at the next
+;; chunk-builder API (clojure.lang.ChunkBuffer + chunk-cons). chunk-buffer collects
-;; block boundary — the Clojure/CLJS ChunkedSeq contract (chunk-first ++
+;; appended items, chunk seals them into a pvec chunk, and chunk-cons prepends that
-;; chunk-rest == the seq). The eager buffer model (chunk-buffer/chunk-append/
+;; chunk onto a rest seq as a real ChunkedCons (cseq-chunked) — empty chunk == just
-;; chunk) builds a plain cseq; chunk-cons/first/rest fall back to seq ops over it.
+;; the rest, like clojure.core/chunk-cons.
-(define na-chunk-size 32)
 (define-record-type jolt-chunkbuf (fields (mutable items)) (nongenerative jolt-chunkbuf-v1))
 (define (na-chunk-buffer cap) (make-jolt-chunkbuf '()))
 (define (na-chunk-append b x) (jolt-chunkbuf-items-set! b (append (jolt-chunkbuf-items b) (list x))) b)
-(define (na-chunk b) (list->cseq (jolt-chunkbuf-items b)))
+(define (na-chunk b) (make-pvec (list->vector (jolt-chunkbuf-items b))))
-(define (na-chunk-cons chunk rest) (jolt-concat chunk rest))
+(define (na-chunk-cons chunk rest)
-;; backing (vector . end-of-block index) for a vector-seq cell, or #f.
+  (if (fx=? 0 (pvec-count chunk)) rest (cseq-chunked chunk 0 rest)))
-(define (na-vblock s)
-  (and (cseq? s) (cseq-cvec s)
-       (let* ((v (cseq-cvec s)) (i (cseq-ci s)))
-         (cons v (fxmin (fx+ i na-chunk-size) (pvec-count v))))))
-(define (na-chunked-seq? x) (and (na-vblock x) #t))
-;; Copy the block [i, end) straight out of the pvec trie's 32-element leaf node
-;; (pv-chunk-for is O(log n)). na-chunk-size == pv-width and blocks are 32-aligned,
-;; so a block is exactly one leaf; the rare non-aligned window crossing a leaf
-;; boundary falls back to per-index reads. Flattening the whole backing vector
-;; per block (pvec-v) made chunk-first O(n), so walking a vector chunk-by-chunk
-;; was O(n^2).
-(define (na-chunk-first s)
-  (let ((vb (na-vblock s)))
-    (if vb
-        (let* ((pv (car vb)) (i (cseq-ci s)) (end (cdr vb)) (len (fx- end i))
-               (node (pv-chunk-for pv i)) (off (fxand i pv-mask)))
-          (if (fx<=? (fx+ off len) (vector-length node))
-              (make-pvec (vec-copy-range node off (fx+ off len)))
-              (let ((out (make-vector len)))
-                (let loop ((j 0))
-                  (if (fx<? j len)
-                      (begin (vector-set! out j (pvec-nth-d pv (fx+ i j) jolt-nil)) (loop (fx+ j 1)))
-                      (make-pvec out))))))
-        (jolt-first s))))               ; eager-buffer fallback
-(define (na-chunk-rest s)
-  (let ((vb (na-vblock s)))
-    (if vb (if (fx>=? (cdr vb) (pvec-count (car vb))) jolt-empty-list (vec->seq (car vb) (cdr vb)))
-        (jolt-rest s))))
-(define (na-chunk-next s)
-  (let ((vb (na-vblock s)))
-    (if vb (if (fx>=? (cdr vb) (pvec-count (car vb))) jolt-nil (vec->seq (car vb) (cdr vb)))
-        (jolt-next s))))
 
 ;; --- extend the collection dispatchers to see a jolt-array ------------------
 (define %na-count jolt-count)

host/chez/natives-meta.ss

@@ -50,7 +50,7 @@
     ;; original mutated it, so (with-meta xs {:k xs}) built a self-referential
     ;; cycle that loops *print-meta* printing.
     ((cseq? x) (make-cseq (cseq-head x) (cseq-tail x) (cseq-forced? x)
-                          (cseq-list? x) (cseq-cvec x) (cseq-ci x)))
+                          (cseq-list? x) (cseq-cvec x) (cseq-ci x) (cseq-crest x)))
     ((jolt-lazyseq? x) (make-jolt-lazyseq (jolt-lazyseq-thunk x) (jolt-lazyseq-val x)
                                           (jolt-lazyseq-realized? x)))
     (else x)))                          ; procedure

host/chez/seq.ss

@@ -31,11 +31,29 @@
 ;; cvec is #f for every other seq; stored as two fields (not a cons) so a vector
 ;; seq cell costs no extra allocation. The rest of the seq layer ignores them, so
 ;; first/rest/count/printing are unchanged.
-(define-record-type cseq (fields head (mutable tail) (mutable forced?) list? cvec ci) (nongenerative chez-cseq-v3))
+;; crest: the ChunkedCons case — cvec holds a STANDALONE chunk pvec (<=32 already-
-(define (cseq-realized head tail) (make-cseq head tail #t #f #f 0))   ; tail already a seq
+;; realized elements), ci the offset within it, and crest the seq AFTER the whole
-(define (cseq-lazy head tail-thunk) (make-cseq head tail-thunk #f #f #f 0))
+;; chunk (the clojure.lang.ChunkedCons _more). This is what map/filter/range emit
-(define (cseq-list head tail) (make-cseq head tail #t #t #f 0))       ; a PersistentList node
+;; so their result is itself a chunked-seq (chained chunked transforms each batch
-(define (cseq-vec head tail-thunk v i) (make-cseq head tail-thunk #f #f v i)) ; vector-backed
+;; by 32, like the JVM). crest is #f for a plain vector-backed seq (whose "rest"
+;; is the next 32-block of the SAME cvec) and for every non-chunked cell.
+(define-record-type cseq (fields head (mutable tail) (mutable forced?) list? cvec ci crest) (nongenerative chez-cseq-v4))
+(define (cseq-realized head tail) (make-cseq head tail #t #f #f 0 #f))   ; tail already a seq
+(define (cseq-lazy head tail-thunk) (make-cseq head tail-thunk #f #f #f 0 #f))
+(define (cseq-list head tail) (make-cseq head tail #t #t #f 0 #f))       ; a PersistentList node
+(define (cseq-vec head tail-thunk v i) (make-cseq head tail-thunk #f #f v i #f)) ; vector-backed
+;; A ChunkedCons cell over a standalone chunk pvec: head is chunk[i], walking
+;; (seq-more) advances within the chunk and then continues into `rest`. `rest` is
+;; the already-coerced after-chunk seq (cseq | jolt-nil | a jolt-lazyseq), held in
+;; crest for chunk-rest/chunk-next and forced lazily by the tail thunk at the chunk
+;; boundary so a chunked map over an infinite chunked source stays productive.
+(define (cseq-chunked chunk i rest)
+  (make-cseq (pvec-nth-d chunk i jolt-nil)
+             (lambda () (let ((i1 (fx+ i 1)))
+                          (if (fx<? i1 (pvec-count chunk))
+                              (cseq-chunked chunk i1 rest)
+                              (jolt-seq rest))))
+             #f #f chunk i rest))
 (define (seq-first s) (cseq-head s))
 (define (seq-more s)                  ; force the tail; returns a seq (cseq | jolt-nil)
   (if (cseq-forced? s) (cseq-tail s)
@@ -246,6 +264,58 @@
                         (string-append (guard (e (#t "value")) (jolt-pr-str f))
                                        " cannot be cast to clojure.lang.IFn"))))))
 
+;; ============================================================================
+;; chunked-seq accessors — the host side of the Clojure IChunkedSeq contract
+;; (chunk-first ++ chunk-rest == the seq). Two chunked shapes share the cseq
+;; record: a vector-backed seq (cvec = whole pvec, ci = absolute index, crest #f,
+;; rest = next 32-block of cvec) and a ChunkedCons (cvec = standalone chunk pvec,
+;; crest = the after-chunk seq). natives-array.ss binds these into clojure.core and
+;; the chunk-buffer/chunk/chunk-cons builder API on top of them.
+;; ============================================================================
+(define seq-chunk-size 32)
+;; (chunk-pvec . end-index) for a chunked cell, else #f. A ChunkedCons block is the
+;; whole remaining chunk (crest carries what comes after); a vector seq block is the
+;; next <=32 elements within cvec.
+(define (na-vblock s)
+  (and (cseq? s) (cseq-cvec s)
+       (let ((v (cseq-cvec s)) (i (cseq-ci s)))
+         (cons v (if (cseq-crest s) (pvec-count v) (fxmin (fx+ i seq-chunk-size) (pvec-count v)))))))
+(define (na-chunked-seq? x) (and (na-vblock x) #t))
+;; Copy the block [i, end) straight out of the pvec trie's 32-element leaf node
+;; (pv-chunk-for is O(log n)). seq-chunk-size == pv-width and vector-seq blocks are
+;; 32-aligned, so a block is exactly one leaf; the rare non-aligned window crossing
+;; a leaf boundary falls back to per-index reads. Flattening the whole backing
+;; vector per block (pvec-v) made chunk-first O(n), so walking chunk-by-chunk was
+;; O(n^2). A ChunkedCons chunk is a small tail-only pvec, so the leaf IS the chunk.
+(define (na-chunk-first s)
+  (let ((vb (na-vblock s)))
+    (if vb
+        (let* ((pv (car vb)) (i (cseq-ci s)) (end (cdr vb)) (len (fx- end i))
+               (node (pv-chunk-for pv i)) (off (fxand i pv-mask)))
+          (if (fx<=? (fx+ off len) (vector-length node))
+              (make-pvec (vec-copy-range node off (fx+ off len)))
+              (let ((out (make-vector len)))
+                (let loop ((j 0))
+                  (if (fx<? j len)
+                      (begin (vector-set! out j (pvec-nth-d pv (fx+ i j) jolt-nil)) (loop (fx+ j 1)))
+                      (make-pvec out))))))
+        (jolt-first s))))               ; eager-buffer fallback
+;; chunk-rest / chunk-next: drop the whole current chunk. For a ChunkedCons that is
+;; crest (the after-chunk seq); for a vector seq it is the seq at the next block.
+(define (na-chunk-rest s)
+  (cond
+    ((and (cseq? s) (cseq-crest s))
+     (let ((r (jolt-seq (cseq-crest s)))) (if (jolt-nil? r) jolt-empty-list r)))
+    ((na-vblock s) => (lambda (vb)
+       (if (fx>=? (cdr vb) (pvec-count (car vb))) jolt-empty-list (vec->seq (car vb) (cdr vb)))))
+    (else (jolt-rest s))))
+(define (na-chunk-next s)
+  (cond
+    ((and (cseq? s) (cseq-crest s)) (jolt-seq (cseq-crest s)))
+    ((na-vblock s) => (lambda (vb)
+       (if (fx>=? (cdr vb) (pvec-count (car vb))) jolt-nil (vec->seq (car vb) (cdr vb)))))
+    (else (jolt-next s))))
+
 ;; ============================================================================
 ;; map / filter / reduce / into / remove + range / take / concat / apply
 ;; ============================================================================
@@ -254,9 +324,22 @@
 ;; an empty seq, so (= () (map f [])) is true and (nil? (map f [])) is false.
 ;; jolt-empty-list seqs back to nil, so it stays a valid lazy-tail terminator for
 ;; the non-empty case (printing / seq= / reduce all walk via jolt-seq).
+;; Single-coll map (core.clj's [f coll] arity). Chunk-preserving: when the source
+;; seq is chunked, realize the WHOLE first chunk — apply f to every element eagerly
+;; into a fresh chunk — and chunk-cons it onto a lazy map of chunk-rest, so the
+;; result is itself a chunked-seq. A non-chunked source maps one element at a time.
 (define (map-seq f s)
-  (if (jolt-nil? s) jolt-empty-list
+  (cond
-      (cseq-lazy (jolt-invoke f (seq-first s)) (lambda () (map-seq f (jolt-seq (seq-more s)))))))
+    ((jolt-nil? s) jolt-empty-list)
+    ((na-chunked-seq? s)
+     (let* ((c (na-chunk-first s)) (n (pvec-count c)) (out (make-vector n)))
+       (let loop ((i 0))
+         (if (fx<? i n)
+             (begin (vector-set! out i (jolt-invoke f (pvec-nth-d c i jolt-nil))) (loop (fx+ i 1)))
+             (cseq-chunked (make-pvec out) 0
+                           (jolt-make-lazy-seq (lambda () (jolt-seq (map-seq f (jolt-seq (na-chunk-rest s)))))))))))
+    (else
+     (cseq-lazy (jolt-invoke f (seq-first s)) (lambda () (map-seq f (jolt-seq (seq-more s))))))))
 (define (map-seq* f seqs)              ; multi-collection map; stops at the shortest
   (if (any-nil? seqs) jolt-empty-list
       (cseq-lazy (apply jolt-invoke f (map seq-first seqs))
@@ -272,12 +355,32 @@
       (jolt-make-lazy-seq (lambda () (jolt-seq (map-seq f (jolt-seq (car colls))))))
       (jolt-make-lazy-seq (lambda () (jolt-seq (map-seq* f (map jolt-seq colls)))))))
 
+;; Chunk-preserving, like core.clj filter: a chunked source has pred applied to the
+;; whole chunk, the kept elements packed into a fresh (possibly smaller) chunk, and
+;; that chunk-cons'd onto a lazy filter of chunk-rest. An all-rejected chunk emits
+;; no empty cell — it recurses straight into chunk-rest (chunk-cons of an empty
+;; chunk == its rest). A non-chunked source filters one element at a time.
 (define (filter-seq pred s keep)
-  (let loop ((s s))
+  (cond
-    (cond ((jolt-nil? s) jolt-empty-list)   ; empty result is () (see map-seq)
+    ((jolt-nil? s) jolt-empty-list)         ; empty result is () (see map-seq)
-          ((eq? keep (jolt-truthy? (jolt-invoke pred (seq-first s))))
+    ((na-chunked-seq? s)
-           (cseq-lazy (seq-first s) (lambda () (filter-seq pred (jolt-seq (seq-more s)) keep))))
+     (let* ((c (na-chunk-first s)) (n (pvec-count c)))
-          (else (loop (jolt-seq (seq-more s)))))))
+       (let loop ((i 0) (acc '()))
+         (if (fx<? i n)
+             (let ((x (pvec-nth-d c i jolt-nil)))
+               (loop (fx+ i 1) (if (eq? keep (jolt-truthy? (jolt-invoke pred x))) (cons x acc) acc)))
+             (let ((kept (reverse acc)))
+               (if (null? kept)
+                   (filter-seq pred (jolt-seq (na-chunk-rest s)) keep)
+                   (cseq-chunked (make-pvec (list->vector kept)) 0
+                                 (jolt-make-lazy-seq
+                                  (lambda () (jolt-seq (filter-seq pred (jolt-seq (na-chunk-rest s)) keep)))))))))))
+    (else
+     (let walk ((s s))
+       (cond ((jolt-nil? s) jolt-empty-list)
+             ((eq? keep (jolt-truthy? (jolt-invoke pred (seq-first s))))
+              (cseq-lazy (seq-first s) (lambda () (filter-seq pred (jolt-seq (seq-more s)) keep))))
+             (else (walk (jolt-seq (seq-more s)))))))))
 ;; filter/remove are fully lazy (LazySeq): defer the predicate and the source seq
 ;; until forced, like Clojure. (lazy-seq* = a 0-arg lazy node coercing to cseq|nil.)
 (define (jolt-filter pred coll)
@@ -290,18 +393,27 @@
 ;; by the last step is unwrapped on the next turn before the seq is consulted.
 ;; reduce a vector's backing store directly by index from element i — no per-
 ;; element seq cells. Honors `reduced`. The chunked-seq fast path.
+;; Reduce a chunk pvec from index i. Returns the accumulator RAW — a `reduced` box
+;; is returned unwrapped-by reduce-seq, not here — so a ChunkedCons continuation can
+;; see early termination instead of folding it back into the running value.
 (define (vec-reduce f acc v i)
   (let ((n (pvec-count v)) (raw (pvec-v v)))
     (let loop ((i i) (acc acc))
-      (cond ((jolt-reduced? acc) (jolt-reduced-val acc))
+      (cond ((jolt-reduced? acc) acc)
             ((fx>=? i n) acc)
             (else (loop (fx+ i 1) (jolt-invoke f acc (vector-ref raw i))))))))
 (define (reduce-seq f acc s)
   (cond
     ((jolt-reduced? acc) (jolt-reduced-val acc))
     ((jolt-nil? s) acc)
-    ;; a vector-backed (chunked) seq reduces its vector directly, in a tight loop.
+    ;; a chunked seq reduces its chunk pvec directly, in a tight loop. A vector seq
-    ((and (cseq? s) (cseq-cvec s)) (vec-reduce f acc (cseq-cvec s) (cseq-ci s)))
+    ;; (crest #f) reduces the whole backing vector and is then done; a ChunkedCons
+    ;; reduces this chunk and continues into its after-chunk rest.
+    ((and (cseq? s) (cseq-cvec s))
+     (let ((acc2 (vec-reduce f acc (cseq-cvec s) (cseq-ci s))))
+       (cond ((jolt-reduced? acc2) (jolt-reduced-val acc2))
+             ((cseq-crest s) (reduce-seq f acc2 (jolt-seq (cseq-crest s))))
+             (else acc2))))
     (else (reduce-seq f (jolt-invoke f acc (seq-first s)) (jolt-seq (seq-more s))))))
 (define jolt-reduce
   (case-lambda
@@ -328,21 +440,31 @@
     (jolt-persistent! (reduce-seq (lambda (t x) (jolt-conj! t x)) (jolt-transient-new to) (jolt-seq from)))))
 
 (define (range-from n) (cseq-lazy n (lambda () (range-from (+ n 1)))))
+;; A bounded range is a real chunked-seq, like clojure.lang.LongRange: eager, with
+;; chunk-first handing out a block of up to 32 consecutive values. Each block is
+;; materialized into a pvec and chunk-cons'd onto a lazy continuation, so a chunked
+;; map/filter over a range batches by 32 (the JVM's observable realization), while a
+;; huge range still produces its tail one block at a time.
 ;; An empty range is () (jolt-empty-list), NOT nil — (range 0) and (range 5 5) are
-;; empty seqs in Clojure, so (= () (range 0)) holds. The same () terminates the
+;; empty seqs in Clojure, so (= () (range 0)) holds, and () seqs back to nil so it
-;; lazy tail of a non-empty range (jolt-empty-list seqs back to nil, see jolt-take).
+;; also terminates the chunked tail (see jolt-take).
-(define (range-bounded n end step)
+(define (range-chunked n end step)
   (if (if (> step 0.0) (< n end) (> n end))
-      (cseq-lazy n (lambda () (range-bounded (+ n step) end step)))
+      (let loop ((i 0) (v n) (acc '()))
+        (if (and (fx<? i seq-chunk-size) (if (> step 0.0) (< v end) (> v end)))
+            (loop (fx+ i 1) (+ v step) (cons v acc))
+            (cseq-chunked (make-pvec (list->vector (reverse acc))) 0
+                          (jolt-make-lazy-seq (lambda () (jolt-seq (range-chunked v end step)))))))
       jolt-empty-list))
 ;; numeric tower: exact 0/1 defaults so (range 3) yields exact ints
 ;; (= JVM longs); flonum args still produce flonums (Scheme arithmetic preserves).
+;; (range) with no bound is the lazy, NON-chunked (iterate inc' 0) form.
 (define jolt-range
   (case-lambda
     (() (range-from 0))
-    ((end) (range-bounded 0 end 1))
+    ((end) (range-chunked 0 end 1))
-    ((start end) (range-bounded start end 1))
+    ((start end) (range-chunked start end 1))
-    ((start end step) (range-bounded start end step))))
+    ((start end step) (range-chunked start end step))))
 
 ;; An empty take result is () (jolt-empty-list), NOT nil — (take 0 coll) and
 ;; (take n []) are empty seqs in Clojure, so (= () (take 0 [:a])) and printing

test/chez/corpus.edn

@@ -2728,7 +2728,7 @@
   {:suite "untested / unchecked-* are plain ops" :label "unchecked-byte" :expected "65" :actual "(unchecked-byte 65)"}
   {:suite "untested / unchecked-* are plain ops" :label "unchecked-char" :expected "\\a" :actual "(unchecked-char 97)"}
   {:suite "untested / unchecked-* are plain ops" :label "unchecked-short" :expected "5" :actual "(unchecked-short 5)"}
-  {:suite "untested / chunk family (eager equivalents) + cat" :label "chunk round-trip" :expected "1" :actual "(let [cb (chunk-buffer 4)] (chunk-append cb 1) (chunk-first (chunk-cons (chunk cb) nil)))"}
+  {:suite "untested / chunk family (eager equivalents) + cat" :label "chunk round-trip" :expected "[1]" :actual "(let [cb (chunk-buffer 4)] (chunk-append cb 1) (chunk-first (chunk-cons (chunk cb) nil)))"}
   {:suite "untested / chunk family (eager equivalents) + cat" :label "cat transducer" :expected "[1 2 3]" :actual "(into [] cat [[1] [2 3]])"}
   {:suite "untested / chunk family (eager equivalents) + cat" :label "ensure-reduced wraps" :expected "true" :actual "(reduced? (ensure-reduced 5))"}
   {:suite "untested / chunk family (eager equivalents) + cat" :label "ensure-reduced keeps reduced" :expected "true" :actual "(reduced? (ensure-reduced (reduced 5)))"}
@@ -2755,11 +2755,14 @@
   {:suite "seq-type-model / specialized seq classes collapse" :label "sort" :expected "clojure.lang.PersistentList" :actual "(class (sort [3 1 2]))"}
   {:suite "seq-type-model / specialized seq classes collapse" :label "subvec" :expected "clojure.lang.PersistentVector" :actual "(class (subvec [1 2 3] 1))"}
   {:suite "seq-type-model / specialized seq classes collapse" :label "drop over a vector" :expected "clojure.lang.LazySeq" :actual "(class (drop 1 [1 2 3]))"}
-  ;; --- chunking-model divergences (allowlisted): jolt is unchunked, so forcing
+  ;; --- chunking-model realization granularity (JVM-matching): a vector's seq is
-  ;; one element realizes 1 (JVM realizes a 32-element chunk), and mapcat/dedupe
+  ;; chunked, so forcing one element of (map f a-vector) realizes the whole first
-  ;; realize 0 at construction (JVM forces the first chunk). jolt-mm6v.
+  ;; 32-block, like the JVM.
-  {:suite "chunking-model / unchunked realization granularity" :label "first over a vector realizes one, not a chunk" :expected "1" :actual "(let [a (atom 0)] (first (map (fn [x] (swap! a inc) x) (vec (range 100)))) @a)"}
+  {:suite "chunking-model / realization granularity" :label "first over a chunked vector realizes the whole 32-block" :expected "32" :actual "(let [a (atom 0)] (first (map (fn [x] (swap! a inc) x) (vec (range 100)))) @a)"}
-  {:suite "chunking-model / unchunked realization granularity" :label "nth 0 over a vector realizes one" :expected "1" :actual "(let [a (atom 0)] (nth (map (fn [x] (swap! a inc) x) (vec (range 100))) 0) @a)"}
+  {:suite "chunking-model / realization granularity" :label "nth 0 over a chunked vector realizes the whole 32-block" :expected "32" :actual "(let [a (atom 0)] (nth (map (fn [x] (swap! a inc) x) (vec (range 100))) 0) @a)"}
+  ;; mapcat/dedupe stay lazier than the JVM at construction: jolt's (apply concat …)
+  ;; and sequence transformer don't force the first chunk just to build the seq
+  ;; (the JVM forces 5 here). Allowlisted in known-divergences.edn (jolt-mm6v).
   {:suite "chunking-model / unchunked realization granularity" :label "mapcat is fully lazy at construction" :expected "0" :actual "(let [a (atom 0)] (mapcat (fn [x] (swap! a inc) [x]) (range 5)) @a)"}
   {:suite "chunking-model / unchunked realization granularity" :label "dedupe is fully lazy at construction" :expected "0" :actual "(let [a (atom 0)] (dedupe (map (fn [x] (swap! a inc) x) (range 5))) @a)"}
   ;; --- integer-box-model: narrow int values behave as integers (certified) ---
@@ -3483,4 +3486,10 @@
   {:suite "host-interop / a fn reports its munged class" :label "(class a-def'd-fn) is ns$munged-name, with the IFn hierarchy as ancestors" :expected "[\"clojure.core$odd_QMARK_\" true]" :actual "[(.getName (class odd?)) (boolean (some #{java.lang.Runnable} (ancestors (class odd?))))]"}
   {:suite "host-interop / MultiFn methods" :label ".getMethod / .dispatchFn on a multimethod" :expected "[true true true]" :actual "(do (defmulti mmc identity) (defmethod mmc :a [_] 1) [(some? (.getMethod mmc :a)) (nil? (.getMethod mmc :z)) (ifn? (.dispatchFn mmc))])"}
   {:suite "reader / namespaced map literal" :label "#:ns{...} qualifies bare keys; :_/x stays unqualified" :expected "[:search 2 nil]" :actual "[(:event/type #:event{:type :search :id 5}) (count #:x{:a 1 :b 2}) (:k #:_{:k 9})]"}
+  {:suite "lazy / chunking" :label "(seq (range 1 50)) is a chunked-seq" :expected "true" :actual "(chunked-seq? (seq (range 1 50)))"}
+  {:suite "lazy / chunking" :label "map over a chunked range realizes a whole 32-block on first" :expected "32" :actual "(let [c (atom 0) s (map (fn [x] (swap! c inc) x) (range 1 50))] (first s) @c)"}
+  {:suite "lazy / chunking" :label "crossing a chunk boundary realizes the next 32-block" :expected "49" :actual "(let [c (atom 0) s (map (fn [x] (swap! c inc) x) (range 1 50))] (nth s 32) @c)"}
+  {:suite "lazy / chunking" :label "chained maps each batch by 32 (result is itself chunked)" :expected "[32 32]" :actual "(let [fc (atom 0) gc (atom 0) s (map (fn [x] (swap! gc inc) x) (map (fn [x] (swap! fc inc) x) (range 1 50)))] (first s) [@fc @gc])"}
+  {:suite "lazy / chunking" :label "filter over a chunked range applies pred to the whole first block" :expected "32" :actual "(let [c (atom 0) s (filter (fn [x] (swap! c inc) (even? x)) (range 1 50))] (first s) @c)"}
+  {:suite "lazy / chunking" :label "a plain lazy seq (not chunked) realizes one element at a time" :expected "1" :actual "(let [c (atom 0) lz (fn lz [n] (lazy-seq (cons n (lz (inc n))))) s (map (fn [x] (swap! c inc) x) (lz 0))] (first s) @c)"}
 ]

test/conformance/known-divergences.edn

@@ -16,7 +16,7 @@
   :seq-type-model
   "jolt models every seq as PersistentList (eager) or LazySeq (deferred); JVM reifies a specialized class per producer (Cons/Iterate/LongRange/Repeat/Cycle/ChunkedSeq/StringSeq/KeySeq/RSeq/ArraySeq/SubVector). Values and laziness are correct, only (class …) differs. jolt-aei7",
   :chunking-model
-  "jolt seqs are unchunked: forcing one element realizes one (JVM realizes a ~32-element chunk), and mapcat/dedupe realize 0 at construction where JVM forces the first chunk — jolt is a finer-grained lazy superset. jolt-mm6v",
+  "jolt now chunks range/vector seqs through map/filter like the JVM (forcing one element realizes the whole ~32-element chunk). What remains finer-grained: mapcat/dedupe realize 0 at construction where the JVM's (apply concat …) / sequence transformer force the first chunk just to build the seq. jolt-mm6v",
   :integer-box-model
   "jolt unifies every integer as one exact-integer type. A Chez fixnum is an immediate identical to the plain integer (no distinct identity to tag, no metadata on numbers), so (byte/short/int n) report Long, not Byte/Short/Integer. Value/arithmetic/equality are correct; a faithful narrow box would crash raw compiled (+ …) or de-optimize all arithmetic. jolt-k9sw"},
  :entries
@@ -53,8 +53,6 @@
   {:suite "seq-type-model / specialized seq classes collapse", :label "sort", :category :seq-type-model}
   {:suite "seq-type-model / specialized seq classes collapse", :label "subvec", :category :seq-type-model}
   {:suite "seq-type-model / specialized seq classes collapse", :label "drop over a vector", :category :seq-type-model}
-  {:suite "chunking-model / unchunked realization granularity", :label "first over a vector realizes one, not a chunk", :category :chunking-model}
-  {:suite "chunking-model / unchunked realization granularity", :label "nth 0 over a vector realizes one", :category :chunking-model}
   {:suite "chunking-model / unchunked realization granularity", :label "mapcat is fully lazy at construction", :category :chunking-model}
   {:suite "chunking-model / unchunked realization granularity", :label "dedupe is fully lazy at construction", :category :chunking-model}
   {:suite "integer-box-model / narrow int class collapses to Long", :label "byte class", :category :integer-box-model}