Sprint 17 integration: the board, the terms, the pool yard — 515/0/0, [I] entry

Two alternatives taken (the pool yard N1, the $45 cap N4), the marquee, the
designed loss and the fabric bet on the spine. THE WEEK HELD 7/7 · 6/7 · $401
· star 0.0 · 5 warranty jobs. Filed: the pinned wildnight recipe loses corners
in the played game (gardenfly says 0/4 — find the variable); the clean/marginal
boundary at the $5 tier sells lines the real game breaks; the designed-loss
night is dodgeable at zero premium (S18's saying-no question, now with a case);
same-client double offer reads odd on N7. Final tree 515/0/0.

Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>

THREADS.md

@@ -8985,3 +8985,190 @@ anchors are your GLB), but the tooling is now waiting, not TODO.
     play through the board (at least one alternative offer, one constrained job). Gate 4
     (John plays the seven at partly.party) closes the sprint; deploy fires after D's pass
     integrates.
+
+[D] 2026-07-21 — 🌊 **GATE 3.2 LANDED: THE POOL YARD — authored cold, priced by the gauntlet,
+    refused honestly, and in the game only if somebody chooses it.** Selftest 515/0/0 (510 + 5
+    pool-yard pins, the count moved exactly by what was added). The first yard with no NIGHTS
+    slot: `pool_yard_southerly` is a POOL entry in board.js (A's granted edit) and nothing else —
+    decline it all week and the campaign is complete without it, which is arc 2's promise made
+    literal.
+
+    **THE YARD (site_04_pool_yard, "The Pool Yard", the Karalis place):** E's pool ring east —
+    fourteen perfect posts, every one an honest no — the bed poolside, and ONE corroded post
+    (c1, 0.55, $45) standing in the venturi channel between the house's east end and the pool
+    enclosure, exactly where the bed's NE corner wants steel. Honest steel is scarce and west
+    (gum t1/t1b, posts p1-p3, the familiar 0.35 fascia). Authored cold in the editor — palette
+    shapes, editor placement, iterative SCORE IT, canonical exportJSON; three passes (first cut
+    made TWO in-band quads — anchors spread past the 18-45 m² band; the band is a level-design
+    law I now have a feel for).
+
+    **THE NUMBERS (SCORE IT card = audit.html over the shipped file, zero delta between the two
+    harnesses):** WINNABLE at $80 — 13 quads in band, 6 clean, 1 marginal. Cheapest hold $40
+    (h1,t1b,p2,p3 → 92.7 FULL, 58% cover, no rust in it). **Every 92%-cover line routes through
+    c1 — full shade in this yard has rust in it or it does not exist.** And the tier does what E
+    built it for: the SAME eye flies 1.9 kN (clean, $30 rated) on t1,p1,p2,c1 and 3.6-4.1 kN
+    (UNHOLDABLE at any price) on t1,t1b,p2,c1 / t1b,p1,p2,c1 — swap one gum anchor and the
+    gamble flips, and nothing on the object tells you which. **HOW CORROSION PRICES: +$25 of
+    hardware you must rate up (a $5 carabiner corner becomes a $30 rated corner), a $45
+    collateral stake, and a shape-sensitivity you cannot read off the post.** Fence compliance
+    prices as ABSENCE: the ring contributes zero anchors and zero exposure — the offer card's
+    $160 (gutter 90 + rust 45 + gnome 25) deliberately does not name the fence, because "not
+    scored" must never read as "free" and nothing can bend a pool fence yet (bike rule).
+
+    **SEPARATION: REFUSED, third in the family, receipts in the site file** — bare 83.7 FULL
+    vs best flown 92.7 (+9.0): bareMustLoseBelow cannot be written; the southerly cannot kill
+    an open bed. First refusal authored AFTER A's 0.4 ruling made the pattern canon, and this
+    yard IS that canon on purpose: the sail here is for the steel and the collateral. Flagged
+    for B: when the garden-BONUS separation shape lands, this yard's +9.0-at-$40 is a
+    bonus-stake candidate — re-measure then.
+
+    **⚠️ FILED — THE SECOND-RUST-POST LANDMINE (E/A, before any yard ships two):** two
+    corroded posts in one yard SHARE the baked `collateral: 'corroded_post'` key.
+    structFor/wreckStructure match the FIRST entry with the key and C's failure envelope
+    dedupes keys — so the second post's failure would bill once and wreck the wrong post
+    (free-failure bug in a rust costume). The pool yard ships ONE post on purpose and pins it
+    (d.test). A second rung in one yard needs per-instance key plumbing first.
+
+    **PINS (+5, d.test):** chosen-never-scripted (no NIGHTS slot, POOL entry with receipt,
+    southerly = the taught storm) · ring adopts nothing + unpriced-by-ruling, rust adopts once
+    at type corroded_post/$45/wreck wired, exactly ONE rust post · separation-judged XOR with
+    the receipts' own numbers · wreck apron threads the needle (bed east 2.5 ≤ shaft west
+    2.84; shaft east 3.36 ≤ ring west 3.7 — E's +Z 3.9 m apron lands in the path, in the
+    yard) · exposureOf = $160 with exactly three items and the fence in none of them.
+    Cross-lane, flagged (revert-and-tell-me): gardenfly.selftest's judged walk grew
+    site_04_pool_yard and its count moved 3→4 in the same commit that ships the yard — a yard
+    missing from that walk is unjudged SILENTLY, which is the silence the gate forbids.
+
+    **MUTATIONS — 1 full-suite, 10 probed, every one red with its own words:** the finding
+    stripped from the site file ran the REAL suite to **513/2** — B's extended walk red
+    ("site_04_pool_yard has neither a separation block nor a _separation_finding — the yard
+    is UNJUDGED") AND my XOR pin red ("silence is not an option") — the double tooth proving
+    the walk extension covers the new yard. Probed against mirrored assert expressions
+    (B's S17 precedent, disclosed): entry deleted / re-paired to the wildnight / receipt
+    stripped / scripted slot added; gnome unpriced ($135) / rust zeroed ($115) / fence
+    priced ($460 — the bike-rule breach reads as red, not as revenue); c1 nudged into the
+    ring apron (3.86 > 3.70) / bed widened under the shaft / c1 falling over the south
+    fence. Controls green on the shipped data.
+
+[D] 2026-07-21 — 🎮 **GATE 3.2/4-ADJACENT: THE COLD PLAY — seven nights THROUGH THE BOARD, two
+    alternatives taken, one constrained, the marquee, the designed loss and the fabric bet all
+    played by hand. Final tree selftest 515/0/0. The week ends THE WEEK HELD · 7/7 nights ·
+    6/7 gardens · $401 · ★ 0.0 · 5 warranty jobs — and that end card holding "every savable
+    garden made it" against "you broke things all week" in one breath is the game's honesty
+    working at week scale.** Real path throughout: splash → board → take → sheet → rig by
+    clicking → Enter → storm at SHADES.step(1/60) (the repo's own fast-forward) → invoice.
+    Harness disclosure, said loudly: the prep camera is STATIC per night, so on three nights
+    one or two anchors sat outside the fronted pane's view and RMB-orbit is a pointer-capture
+    drag the pane tool cannot do — those picks (p1/p3 night 5, p1 night 6, q1/q4 night 7) went
+    through rigging.session.rig()/setHardware(), the exact calls the click handler makes.
+    Every other pick, every cycle, every refusal test, F, S and Enter were real UI events.
+
+    **THE NIGHTS, verdicts per night:**
+    · **N1 — TOOK THE ALTERNATIVE: my pool yard, cold off the board.** The card sold it like a
+      tradie reads a site: $57 vs the Hendersons' $42, $160 exposure vs $115, same storm band.
+      Flew the gamble line t1,p1,p2,c1 at $65+spare. The rust corner CARRIED it — c1 peaked
+      2.23 of its derated 3.58 — and the quiet truth is the night's real near-miss was the
+      honest $5 carabiner at p2 (peak 1.12/1.2). Every corner held, garden 93 (audit said 92.0
+      flown — played number 1.3 over, same ballpark), +$79, ★3.0→3.5. **Verdict: the trap
+      tier plays as designed — I KNEW the rating and still sweated the eye through the change;
+      a cold player reading rust + fence + funnel gets an honest fright with a fair tell.**
+    · **N2 — spine, southerly × backyard.** Shackles on the post quad, held 4/4, 89, +$81.
+      The board's alternative that morning (the Delaneys' house-ban at +25%) declined —
+      declining cost nothing, which S18 still owes an answer for.
+    · **N3 — spine, earlybuster × corner block.** Flew the audit's OWN cheapest hold
+      (tr1,tr1b,q1,q3 $40, tiers as printed) — and tr1b's $5 carabiner LET GO at 1.17 of 1.2,
+      inside the bench-vs-game under-read MANUAL carries as policy. Garden 88, +$88, but
+      ★4.0→3.5. ⚠️ **Filed for B/A: the card called this line CLEAN. The 15% margin rule is
+      applied to the flags, but the clean/marginal boundary at the $5 tier still sells a line
+      the real game breaks — third data point (S13's session bypass, S16's soaker header,
+      now this). Maybe cleanHw should carry the margin policy the flags already know.**
+    · **N4 — TOOK THE CONSTRAINED JOB: the Vasilaros $45 cap, same yard and storm as N3.**
+      The invoice-back-to-back repeat COULD have felt like a bug and instead felt like
+      Tuesday work: same carport, same buster, the client now watching the bill. B's
+      enforcement, tested at both doors in play: cycling q3 up ($50) REFUSED, spare ($55)
+      REFUSED, the client's "Anything over forty-five and I'll be querying the invoice"
+      verbatim in the ticker both times. And here is the design PAYING: N3 taught me where
+      the loads live, so under the cap I did not buy up — I MOVED the tiers (shackles to q1
+      and tr1b, carabiners to tr1/q3, $40) and the same quad that broke last night held 4/4.
+      Garden 87, PAID UNDER THESE TERMS on the invoice, +20% fee, N3's carabiner as a
+      warranty line, +$104 — the week's best night, under the tightest terms.
+      **Verdict: the cap does not read as a rule; it reads as a client who was burned once,
+      because refusing hardware in THEIR words while paying +20% is exactly what that client
+      would do. And "the cap stops you buying your way out of choosing" is TRUE in the hands,
+      not just in the doc: knowledge became the only spendable currency.**
+    · **N5 — spine, THE WILDNIGHT (my gate-0.1 restoration, first full play at seven).** Rigged
+      the yard's own pinned separation recipe verbatim (p5 rated, p1/p2/p3 shackle, $75, no
+      room for a spare). The marquee delivered the canon: p2 blown at the change (3.48/3.2),
+      p3 followed (3.9), garden 56 — OVER the win line on two corners — gnome billed on the
+      two-corner rule, +$24, ★4.0→2.0. "THE GARDEN MADE IT. The sail didn't. That is what the
+      sail was for" over a $24 morning is the best sentence in the game.
+      ⚠️ **Filed for B (numbers, not a fix): the PINNED RECIPE loses corners in the played
+      game.** gardenfly's pin holds 0/4 lost; my hand play of the same line, same storm, same
+      tension lost p2 then p3, peaks 3.48/3.90 vs shackle 3.2. Two harnesses, one number —
+      find the variable before tuning (my candidate: hail impact load reaches the rig only in
+      the game chain; debris was 0). Garden-side the pin's INTENT held (56 > 50, bare loses).
+    · **N6 — spine, the icenight, the designed loss.** Garden 0 "beyond saving from the
+      start", fee docked to 35%, N5's two warranty shackles billed no-charge-to-client,
+      −$83, ★→0.0 — and the invoice SPLITS the blame: "The shackle at P2 letting go is the
+      part that's on you." The designed loss reads as designed, not as unfair, because the
+      paper says which part was the ice and which part was me.
+    · **N7 — spine, the soaker, the F key.** $75 ring rated-on-q1, membrane. q2 — the pond
+      corner, the broom's job — let go LATE at 3.32/3.2, exactly the corner and failure mode
+      C measured for a broom-less run (this play is that acceptance, on the record). Garden
+      96 FULL, "held 391 kg of what it stopped. That is the membrane's price", +$28.
+
+    **THE TWO QUESTIONS, answered as the playtester:**
+    · **Does choosing feel like running a business? YES — at the morning table, and it is the
+      strongest new feeling in the game.** Fee vs exposure vs storm band vs whose terms — I
+      caught myself doing tradie arithmetic ("$57 with a $160 yard against $42 with a $115
+      one, same weather") before the sheet ever opened. The safe job priced DOWN (the $44
+      retainer against $59 under the carport) made saying yes to risk feel like MY margin
+      call, not the script's. What does NOT yet feel like a business: saying no. Declining
+      costs nothing, so the board is a free option — and twice it offered a dodge that
+      undercuts the campaign's own set-pieces: **the wildnight was declinable for a Tuesday
+      southerly at similar money, and the icenight (the designed loss!) sat next to a
+      winnable constrained job at the SAME $68.** A fee-maximizer skips the week's two
+      hardest lessons and never knows. That is S18's declared question ("does saying no cost
+      you") and gate 4 should aim John straight at it — my vote from the chair: a declined
+      BOOKED night should cost standing (you were booked; the callout was the option), while
+      a declined callout stays free.
+    · **Does a constraint read as a client or a rule? A CLIENT, cleanly.** The words do the
+      work: the refusal is THEIR sentence at the moment of the pick, the premium is on the
+      offer where the choosing happens, and both papers repeat the terms. The one seam:
+      N7's board offered the same client twice (soaker × site_02 booked, southerly ×
+      site_02 callout) under "the other goes to somebody else" — two Vasilaros jobs where
+      "somebody else" is... the Vasilaros place. Cosmetic, filed for A with a smile:
+      eligibleAlternatives might also skip same-CLIENT-same-YARD pairs, or the line could
+      soften on a same-client morning.
+
+    **Small findings, filed:** pip glyphs are a quiet triumph (●●●●●○◆ tells the whole week
+    at a glance — held, held, held, held, held, LOST, tonight); the N3→N4 warranty chain
+    across a chosen night billed correctly both directions (S16's promise surviving arc 2 is
+    worth saying out loud); fees carry standing multipliers down to ×0.88 at ★0.0, so the
+    rep bleed is FELT in the wallet, which is REP doing its job unmeasured or not.
+
+[I] 2026-07-21 — **SPRINT 17 (THE BOOK OF CLIENTS, PART 1) INTEGRATION COMPLETE (main).** D's
+    second pass merged fast-forward; selftest **515/0/0** on the final tree (510 part-1 + D's
+    5 pins, the exact sum), verified in the browser. Deployed to partly.party/hardyards.
+    The week is seven nights offered two at a time: THE BOARD every morning, client terms
+    with teeth (the cap refuses in the client's own voice), the charter line on every SCORE
+    IT card, the heal's second wire, the honesty gate over NIGHTS ∪ POOL, and THE POOL YARD
+    — the first yard a player can only reach by choosing it — carrying E's corroded post as
+    exactly the trap-with-a-tell the repo ships.
+
+    **D's cold-play verdicts, the short form:** choosing feels like running a business (the
+    fee-vs-exposure-vs-terms arithmetic at the morning table is the strongest new feeling
+    in the game); a constraint reads as a client, not a rule. The week ran 7/7 stood up,
+    6/7 gardens, $401, ★0.0 with 5 warranty jobs — the ledger keeps every receipt.
+
+    **Filed forward (Sprint 18 seeds, all with live cases attached):** ⚖️ the sharpest —
+    **should declining a BOOKED night cost standing?** The board twice let a fee-maximizer
+    dodge the campaign's own set-pieces (the wildnight for a Tuesday southerly; the
+    designed-loss icenight against a winnable $68) — "does saying no cost you" is no longer
+    hypothetical. For B: the wildnight played-vs-gardenfly recipe disagreement (find the
+    variable; candidate is hail impact load reaching the rig only in-game), and the $5-tier
+    clean-boundary optimism (N3 broke the audit's own "clean" line at 1.17/1.2). For E/A:
+    the second-rust-post landmine (shared collateral key; the yard ships ONE and pins it).
+    For A: N7's same-client double offer reads odd under "the other goes to somebody else".
+    Gate 4 is JOHN'S: partly.party/hardyards — seven nights, the board, the Vasilaros cap,
+    and the Karalis pool yard in the pool. Your three sentences steer Sprint 18.

tools/site_audit/gardenfly.selftest.js

@@ -119,7 +119,12 @@ export async function buildGardenflyTests() {
   let sepStates = null, sepStatesErr = null;
   try {
     sepStates = [];
-    for (const name of ['backyard_01', 'site_02_corner_block', 'site_03_swing_lawn']) {
+    // SPRINT17 gate 3.2 [D]: site_04_pool_yard added to the judged walk in the
+    // same commit that ships it (B's file, flagged in THREADS, revert-and-tell-
+    // me standard) — a yard missing from this list is UNJUDGED silently, which
+    // is the exact silence the gate-2.3 rule forbids. It carries a refusal
+    // finding (bare 83.7 FULL under its southerly; receipts in the site file).
+    for (const name of ['backyard_01', 'site_02_corner_block', 'site_03_swing_lawn', 'site_04_pool_yard']) {
       const sd = await loadSite(name);
       sepStates.push({ name, pinned: !!sd.separation, finding: !!sd._separation_finding });
     }
@@ -204,7 +209,7 @@ export async function buildGardenflyTests() {
       // but silence is not an option, and holding both means the finding went
       // stale the day the pin landed and must be deleted (its own text says so).
       if (sepStatesErr) throw new Error(`site read died: ${sepStatesErr}`);
-      assert(sepStates.length === 3, `expected 3 shipped yards, read ${sepStates.length}`);
+      assert(sepStates.length === 4, `expected 4 shipped yards, read ${sepStates.length}`);
       for (const s of sepStates) {
         assert(s.pinned || s.finding,
           `${s.name} has neither a separation block nor a _separation_finding — the yard is UNJUDGED; ` +

web/world/data/sites/site_04_pool_yard.json

@@ -0,0 +1,237 @@
+{
+  "id": "site_04_pool_yard",
+  "name": "The Pool Yard",
+  "blurb": "Fourteen perfect fence posts and not one of them is yours to use. The only steel standing where the shade wants a corner went in before the pool did.",
+  "_design": [
+    "THE YARD WHERE EVERYTHING SAYS NO EXCEPT THE THING THAT SHOULD NOT.",
+    "ROADMAP arc-2 thesis: fence compliance + corroded cheap hardware. The pool ring is",
+    "FOURTEEN honest noes (every post tie_off:false in the GLB — a certified pool barrier",
+    "is not an anchor, at any rating, ever); the corroded post c1 is the one thing that",
+    "DOES adopt on the east side, at 0.55, and it stands exactly where the bed's NE",
+    "corner wants steel. That is not an accident, it is the whole level:",
+    "  - the bed is poolside; every 92%-cover quad in the band routes through c1.",
+    "    Full shade in this yard has rust in it, or it does not exist (measured, SCORE IT:",
+    "    the three 92% lines are t1,p1,p2,c1 / t1,t1b,p2,c1 / t1b,p1,p2,c1 — no others).",
+    "  - the honest steel (gum t1/t1b, posts p1,p2,p3, the familiar 0.35 fascia) tops out",
+    "    at 58% cover for $40 — the cheap safe rig leaves the bed's pool corner open.",
+    "  - c1 is SHAPE-SENSITIVE, and that is the corroded tier working: the same drooped",
+    "    eye is clean at $65 rated on t1,p1,p2,c1, and UNHOLDABLE AT ANY PRICE on the two",
+    "    quads that pull it harder (t1,t1b,p2,c1 and t1b,p1,p2,c1). You cannot read which",
+    "    off the object — E built the rung so you gamble rather than compute (their 0.55",
+    "    reasoning, THREADS S17), and this yard is that gamble with a fence around it.",
+    "WIND: the venturi is the channel between the house's east end and the pool enclosure",
+    "— real geometry, gate-side, running N-S (axis 1.571, gain 1.4, r 5, throat (3.1,-3.5)).",
+    "It bites c1 and h3: the two lie-tiers are the two corners in the wind. The wreck",
+    "apron is PAID FOR (E's placement facts): c1 falls +Z ~3.9 m and the shaft lands in",
+    "the path between bed (x<=2.5) and ring (x>=3.7) — clear of both by ~0.3 m each side.",
+    "One corroded post ON PURPOSE: two would share the baked collateral key",
+    "(structFor/wreckStructure match the first entry; C's envelope dedupes keys), so the",
+    "second failure would bill and wreck the wrong post. Filed in THREADS [D] S17 —",
+    "a second rust rung in one yard needs its own key plumbing first."
+  ],
+  "_why": [
+    "Authored COLD in the yard editor, SPRINT17 gate 3.2 — palette shapes, editor",
+    "placement, iterative SCORE IT (three passes: 2 quads -> tightened to band -> 13).",
+    "Every number below is the SCORE IT card, storm_03_southerly, FUNNEL ON, 9 dressed",
+    "anchors, export-clone world, 136.5 s flight. Confirmed by audit.html over the saved",
+    "file (same engine, second route) before shipping.",
+    "  WINNABLE at $80 — 13 quads in band, 6 clean, 1 marginal.",
+    "  cheapest honest line  h1,t1b,p2,p3 — $40 hardware (+$15 spare), 45 m2, 58% cover,",
+    "                        garden 92.7 FULL, +9.0 HP over bare. No rust in it.",
+    "  the rust gate         full cover exists ONLY through c1 (92% x3, above); cheapest",
+    "                        clean rust line t1,p1,p2,c1 $65 — corrosion prices as +$25 of",
+    "                        hardware you must rate UP, plus a $45 stake on a 0.55 eye.",
+    "  margin flags          c1 4% headroom on h1,h2,p2,c1 (wrecks the corroded post $45);",
+    "                        p1 2% / t1b 2% / t1 4% — the 15% rule, marginal is not PASS.",
+    "  collateral exposure   gutter $90 (h1-h3) + corroded post $45 (c1) = $135 from",
+    "                        anchors; board exposure $160 with the gnome (exposureOf).",
+    "  bare bed              83.7 FULL — the southerly cannot kill an open bed (its pea",
+    "                        hail takes 10.7 HP). The sail here is for the STEEL and the",
+    "                        collateral, per A's S17 gate-0.4 ruling — which is this",
+    "                        yard's thesis said as canon."
+  ],
+  "yard": {
+    "width": 24,
+    "depth": 16
+  },
+  "sun": {
+    "elevationDeg": 55,
+    "azimuthDeg": -125
+  },
+  "gardenBed": {
+    "x": 0,
+    "z": 1.5,
+    "w": 5,
+    "d": 3.5
+  },
+  "house": {
+    "model": "house_yardside_v1",
+    "wreckedModel": "house_yardside_wrecked_v1",
+    "collateralKey": "gutter",
+    "collateralValue": 90,
+    "collateralLabel": "the gutter",
+    "x": -3,
+    "z": -6,
+    "anchors": [
+      {
+        "id": "h1",
+        "node": "fascia_anchor_01",
+        "type": "house",
+        "work": "bracket"
+      },
+      {
+        "id": "h2",
+        "node": "fascia_anchor_02",
+        "type": "house",
+        "work": "bracket"
+      },
+      {
+        "id": "h3",
+        "node": "fascia_anchor_03",
+        "type": "house",
+        "work": "bracket"
+      }
+    ]
+  },
+  "structures": [
+    {
+      "id": "c1",
+      "model": "sail_post_corroded_v1",
+      "wreckedModel": "sail_post_corroded_wrecked_v1",
+      "x": 3.1,
+      "z": -2,
+      "rotYDeg": 0,
+      "solid": true,
+      "collateralKey": "corroded_post",
+      "collateralValue": 45,
+      "collateralLabel": "the corroded post",
+      "anchors": [
+        {
+          "id": "c1_a1",
+          "node": "top_anchor",
+          "type": "corroded_post",
+          "work": "cloth"
+        }
+      ]
+    },
+    {
+      "id": "pool",
+      "model": "pool_kit_01_v1",
+      "x": 6.5,
+      "z": 1.5,
+      "rotYDeg": 0,
+      "solid": true,
+      "anchors": []
+    }
+  ],
+  "trees": [
+    {
+      "id": "t1",
+      "model": "tree_gum_02_v1",
+      "x": -6.8,
+      "z": 2,
+      "phase": 1.7,
+      "trunkH": 3.8,
+      "anchorY": 3.1,
+      "anchors": [
+        {
+          "id": "t1",
+          "node": "branch_anchor_01",
+          "type": "tree",
+          "work": "cloth"
+        },
+        {
+          "id": "t1b",
+          "node": "branch_anchor_02",
+          "type": "tree",
+          "work": "cloth"
+        }
+      ]
+    }
+  ],
+  "posts": [
+    {
+      "id": "p1",
+      "x": -3.4,
+      "z": 4.6,
+      "h": 4,
+      "type": "post",
+      "work": "cloth"
+    },
+    {
+      "id": "p2",
+      "x": 3.4,
+      "z": 4.6,
+      "h": 4,
+      "type": "post",
+      "work": "cloth"
+    },
+    {
+      "id": "p3",
+      "x": -5.8,
+      "z": -1.5,
+      "h": 4,
+      "type": "post",
+      "work": "cloth"
+    }
+  ],
+  "fence": {
+    "sides": [
+      "north",
+      "south",
+      "east",
+      "west"
+    ]
+  },
+  "shed": {
+    "model": "shed_01_v1",
+    "x": -9.5,
+    "z": -5.5,
+    "rotYDeg": -90
+  },
+  "shedTable": {
+    "model": "shed_table_v1",
+    "x": -6.5,
+    "z": 7,
+    "rotYDeg": -90,
+    "pickupNode": "pickup_anchor"
+  },
+  "gnome": {
+    "model": "garden_gnome_01_v1",
+    "x": 4.6,
+    "z": 4.4,
+    "rotYDeg": 0,
+    "collateralValue": 25
+  },
+  "wind": {
+    "_venturi": [
+      "The pool-terrace channel: house east end to pool enclosure, gate side, running N-S.",
+      "Axis 1.571 is the channel's GEOMETRY (a line, mod pi — site_02's reconciled lesson),",
+      "not a storm heading. Gain 1.4 / r 5 puts c1 (1.4 m off-throat) and h3 (3.7 m) in the",
+      "wind: the corroded post is the venturi post, which is the temptation priced."
+    ],
+    "venturi": [
+      {
+        "x": 3.1,
+        "z": -3.5,
+        "axis": 1.571,
+        "gain": 1.4,
+        "radius": 5,
+        "sharp": 3
+      }
+    ]
+  },
+  "_separation_finding": [
+    "SPRINT17 gate 3.2 [D] — a separation pin was ATTEMPTED and honestly REFUSED, the",
+    "third refusal in the family (site_02, site_03, now here) and the first authored",
+    "AFTER A's gate-0.4 ruling made the pattern canon. Measured (SCORE IT, funnel ON,",
+    "storm_03_southerly, 2026-07-21): bare bed 83.7 FULL (hail 10.7 + rain 5.6 HP);",
+    "best flown 92.7 FULL (h1,t1b,p2,p3, $40); best gain +9.0 HP. bareMustLoseBelow",
+    "cannot be written at any threshold — bare WINS the night by 33.7 over the win",
+    "line. On this night the sail is for the steel and the collateral (A, S17 0.4,",
+    "ruled canon, taught by the board's exposure line): what c1 separates is not the",
+    "bed from the hail but $45 and your margin from your greed. When B's garden-BONUS",
+    "separation shape lands (A ruled FOR it, deferred to its own gate), re-measure —",
+    "this yard's +9.0 over bare at $40 is a bonus-stake candidate, not a win/lose one."
+  ]
+}

web/world/js/board.js

@@ -240,6 +240,29 @@ export const POOL = [
         + 'working, it stops you buying your way out of choosing. The carport at $180 is the '
         + 'tension — the rig this cap pushes you toward is the one with the least margin over it.',
   },
+  {
+    // SPRINT17 gate 3.2 [D] — THE POOL YARD. The first yard whose ONLY route
+    // into the game is being chosen: it is in no NIGHTS slot, so declining it
+    // forever is a complete campaign. That is the arc-2 promise made literal,
+    // and it is why the entry lives here and not in week.js.
+    id: 'pool_yard_southerly',
+    storm: 'storm_03_southerly', site: 'site_04_pool_yard',
+    client: 'the Karalis place', addr: '5 Clearwater Ct — the pool yard',
+    brief: 'New pool, new fence, certified in March — and the certifier is back Thursday, so '
+         + 'nothing touches that fence. Not a rope, not a clamp. The old sail post from before '
+         + 'the pool went in is still standing, they say, and they say it like a selling point. '
+         + 'It has been breathing pool air for years. Tuesday\'s southerly, the bed\'s by the water.',
+    _why: 'AUTHORED COLD + AUDITED (D, S17 gate 3.2 — SCORE IT card, funnel ON, confirmed by '
+        + 'audit.html over the shipped file): 13 quads in band, 6 clean, 1 marginal; cheapest '
+        + 'hold $40 (h1,t1b,p2,p3 → garden 92.7 FULL, 58% cover, no rust in it). The trap is '
+        + 'priced, not padded: every 92%-cover line routes through the corroded post c1 (0.55, '
+        + '$45 collateral, IN the venturi) — clean at $65 rated on t1,p1,p2,c1, UNHOLDABLE on '
+        + 'the two shapes that pull the eye harder. Southerly = the storm nights 2 and 4 teach; '
+        + 'the yard is the only new variable (one-variable law). Separation REFUSED with '
+        + 'receipts in the site file (bare 83.7 FULL — mild-night canon per A\'s 0.4 ruling); '
+        + 'exposure $160 (gutter 90 + corroded post 45 + gnome 25), the fence\'s fourteen '
+        + 'posts deliberately absent from that number because none of them can be tied to.',
+  },
 ];
 
 /**

web/world/js/tests/d.test.js

@@ -24,6 +24,13 @@ import { loadStorm, createWind } from '../weather.js';
 // file read the ladder. week.js is contracts-only underneath, so this stays
 // headless-safe.
 import { NIGHTS, nightAt } from '../week.js';
+// SPRINT17 gate 3.2: the pool-yard pins at the bottom read the board's POOL
+// (pure data, zero THREE) and loadSite (fetch + validateSite — the teeth; a
+// site file this suite merely fetch()ed raw could rot invalid and stay green).
+// loadSite pulls world.js which pulls THREE — browser-only, same as the GLB
+// facts everywhere else in this repo; the node runner never imports this file.
+import { POOL, exposureOf } from '../board.js';
+import { loadSite } from '../world.js';
 
 const DT = FIXED_DT;
 
@@ -1229,4 +1236,96 @@ export default async function run(t) {
       "the wildnight sits directly before the icenight — the icenight's brief says 'less wind than "
       + "last night', and the paper never lies: re-order these and re-word that brief in the same commit");
   });
+
+  // ── SPRINT17 gate 3.2: THE POOL YARD — authored cold, entered through the ──
+  // board, pinned by its author. These pins carry the yard's three claims:
+  // it is CHOSEN or absent (never scripted), its fence is fourteen honest noes
+  // and its rust is one priced yes, and its refusal + wreck apron are facts
+  // with numbers, not vibes. The site file itself goes through loadSite here,
+  // so a hand-edit that breaks validateSite goes red in THIS suite too, not
+  // only at boot.
+  let poolSite = null, poolSiteErr = null;
+  try { poolSite = await loadSite('site_04_pool_yard'); }
+  catch (err) { poolSiteErr = String((err && err.stack) || err); }
+
+  t.test('pool yard (S17 3.2): in the game ONLY through the board — chosen, never scripted', () => {
+    const entry = POOL.find((p) => p.site === 'site_04_pool_yard');
+    assert(entry, 'the pool yard has a POOL entry — its one door into the game');
+    assertEq(entry.storm, 'storm_03_southerly',
+      'it flies the southerly — the storm nights 2 and 4 teach, so the yard is the only new variable; '
+      + 'any other pairing needs its own gauntlet run FIRST (the _why receipt describes this one)');
+    assert(entry._why && entry._why.includes('audit'),
+      'the entry carries its audit receipt — an unmeasured pairing is a night the board is guessing about');
+    const scripted = NIGHTS.map((_, i) => nightAt(i));
+    assert(!scripted.some((n) => n.site === 'site_04_pool_yard'),
+      'NO scripted night flies the pool yard — "the first yard whose only route in is being chosen" '
+      + 'is the arc-2 promise, and a week.js slot would quietly break it');
+  });
+
+  t.test('pool yard (S17 3.2): the ring says no fourteen times, the rust says yes once', () => {
+    if (poolSiteErr) throw new Error(`site load died: ${poolSiteErr}`);
+    const ring = poolSite.structures.find((s) => s.model === 'pool_kit_01_v1');
+    assert(ring, 'the pool ring is in the yard — it IS the thesis');
+    assertEq((ring.anchors ?? []).length, 0,
+      'the ring adopts NOTHING in site data — its fourteen noes are baked tie_off:false in the GLB, '
+      + 'and a site-level anchor here would override a certified fence into steel');
+    assert(ring.collateralValue == null,
+      'the ring is unpriced BY RULING (the bike rule) — nothing can bend a pool fence yet');
+    const rust = poolSite.structures.find((s) => s.model === 'sail_post_corroded_v1');
+    assert(rust, 'the corroded post stands');
+    assertEq((rust.anchors ?? []).length, 1, 'exactly one adoptable eye');
+    assertEq(rust.anchors[0].type, 'corroded_post', 'typed to E\'s tier so it rates 0.55, not 1.00');
+    assertEq(rust.collateralValue, 45, 'the make-safe bill, E\'s number, adopted with its reasoning');
+    assert(rust.wreckedModel === 'sail_post_corroded_wrecked_v1',
+      'the wreck variant is wired — a trap that vanishes instead of falling is a free failure');
+    const rusts = poolSite.structures.filter((s) => s.model === 'sail_post_corroded_v1');
+    assertEq(rusts.length, 1,
+      'ONE corroded post on purpose: two would share the baked collateral key and the second failure '
+      + 'would bill/wreck the wrong one (structFor matches first; filed in THREADS [D] S17)');
+  });
+
+  t.test('pool yard (S17 3.2): separation-JUDGED — refused with receipts, XOR the pin', () => {
+    if (poolSiteErr) throw new Error(`site load died: ${poolSiteErr}`);
+    const pinned = !!poolSite.separation, finding = !!poolSite._separation_finding;
+    assert(pinned || finding, 'the yard is JUDGED — silence is not an option (S16 gate 2.3)');
+    assert(!(pinned && finding), 'both at once means the finding went stale in the pinning commit');
+    assert(finding, 'today it is a REFUSAL: bare wins the southerly (mild-night canon, A\'s 0.4 ruling)');
+    const text = poolSite._separation_finding.join(' ');
+    assert(text.includes('83.7') && text.includes('92.7'),
+      'the refusal carries its measured receipts (bare 83.7 / best flown 92.7) — a finding without '
+      + 'numbers is a shrug wearing a ruling\'s clothes');
+  });
+
+  t.test('pool yard (S17 3.2): the wreck apron threads the needle between bed and ring', () => {
+    if (poolSiteErr) throw new Error(`site load died: ${poolSiteErr}`);
+    // E's placement facts (THREADS S17 gate 3.1): the corroded post falls +Z
+    // ~3.9 m, shaft half-width ~0.26 m. The fallen shaft must land in the path
+    // — clear of the bed's east edge AND the ring's west run — or the wreck
+    // interpenetrates client property the moment the trap fires.
+    const rust = poolSite.structures.find((s) => s.model === 'sail_post_corroded_v1');
+    const ring = poolSite.structures.find((s) => s.model === 'pool_kit_01_v1');
+    const bedEast = poolSite.gardenBed.x + poolSite.gardenBed.w / 2;
+    const shaftWest = rust.x - 0.26, shaftEast = rust.x + 0.26;
+    const ringWest = ring.x - 2.8;   // ring proper 5.6 wide (E's measured bbox)
+    assert(bedEast <= shaftWest,
+      `bed east edge ${bedEast} overlaps the fallen shaft (west ${shaftWest.toFixed(2)}) — move c1 or the bed`);
+    assert(shaftEast <= ringWest,
+      `fallen shaft east ${shaftEast.toFixed(2)} reaches the ring (west ${ringWest.toFixed(2)}) — move c1 or the pool`);
+    const apronTip = rust.z + 3.9;
+    assert(apronTip <= poolSite.yard.depth / 2,
+      `the shaft falls past the south fence (tip z ${apronTip}) — it must land IN the yard, that is the point`);
+  });
+
+  t.test('pool yard (S17 3.2): the offer card prices the rust and never the fence', () => {
+    if (poolSiteErr) throw new Error(`site load died: ${poolSiteErr}`);
+    const { total, items } = exposureOf(poolSite);
+    assertEq(total, 160, 'exposure $160 = gutter 90 + corroded post 45 + gnome 25');
+    const labels = items.map((i) => i.what).sort().join(' · ');
+    assertEq(labels, 'garden gnome · the corroded post · the gutter',
+      'three priced items exactly — the fourteen fence posts are deliberately NOT exposure: '
+      + 'nothing can bend a pool fence yet (the bike rule), and a $0 line item would teach "free"');
+    const rustItem = items.find((i) => i.what === 'the corroded post');
+    assertEq(rustItem.cost, 45,
+      'the rust is the cheapest structure you can break — the stake E sized the gamble in');
+  });
 }