{
    "ok": true,
    "keep": 3,
    "count": 3,
    "builds": [
        {
            "version": "v0.1.1",
            "status": "NOMINAL",
            "filename": "Nominal_v0.1.1.zip",
            "size": 574833236,
            "size_human": "548.2 MB",
            "sha256": "1809a860a40dd501b256903356fd25902a1828215511a7844df45c832b82b3cf",
            "uploaded_at": 1786858509,
            "uploaded_iso": "2026-08-15T22:35:09-07:00",
            "downloads": 7,
            "notes": "STATUS: NOMINAL - Build v0.1.1\n\nEVERY PUZZLE IS NOW DEALT IN MISSIONS\n\nFor a dozen versions a mission handed out the same five puzzles while\neverything built since sat on the bench waiting for somebody to play it.\nYou have, so the whole catalogue is live: all 41.\n\nMISSION TIMERS: 30 SECONDS TO 120\n\nThat one number decides how long every mission puzzle runs. A puzzle's\nown time limit never applied inside a mission, which is why some of them\nfelt far tighter in play than they did on the bench - Archive asks for\nninety seconds and was getting thirty.\n\nThirty is fine for a puzzle you have played fifty times and hostile to\none you are meeting for the first time. Right now every puzzle is one you\nare meeting for the first time.\n\nBE WARNED WHAT THIS IS: generous, not balanced. The set has been\nMEASURED - every difficulty curve is monotone, every generator proves its\nboard can actually be finished - but none of it has been tuned against a\nmission's pacing, and 41 puzzles at two minutes each will make missions\nlong. That is deliberate. It is far easier to find what genuinely goes\nwrong by playing it than by guessing at it, and both halves reverse\ncheaply: one bool per puzzle, one number for the clock.\n\nSo: expect it to be loose. Tell me where it drags, where a puzzle\noutstays its welcome, and which ones you would rather never see again.\n\nRELAY PATH IS DELETED\n\nCut from the pool a while back for being genuinely unpleasant, and it has\nbeen sitting as a dead class since. Gone properly now.\n\nWHAT WOULD HELP MOST\n\n1. A full run. That is the whole point of this build - 36 puzzles that\n   have never been dealt under a clock with detection running and other\n   tabs competing for your attention.\n2. Pacing. Two minutes is a guess. Tell me if it should be ninety, or\n   sixty, or per-puzzle.\n3. The solo task cap from v0.1.0 - a node should hand you at most two\n   tasks. Verified by reading the code, never by playing it.\n4. Anything that outstays its welcome. A puzzle can be provably fair and\n   still be one you do not want to see again, and that is the judgement\n   no amount of measuring here can make.",
            "url": "download.php?b=build_20260815_223509.zip"
        },
        {
            "version": "v0.1.0",
            "status": "NOMINAL",
            "filename": "Nominal_v0.1.0.zip",
            "size": 574846613,
            "size_human": "548.2 MB",
            "sha256": "e9ceb94c5bdd6ec8b7152afc37768f36712a73e46180e6990e27beddf9743d1b",
            "uploaded_at": 1786855719,
            "uploaded_iso": "2026-08-15T21:48:39-07:00",
            "downloads": 2,
            "notes": "STATUS: NOMINAL - Build v0.1.0\n\nTwo things you asked for, both from playing v0.0.99.\n\nCACHE SIFT SHOWS THE RECORD UNDER THE TOP ONE\n\nEach card now carries a small colour pip for whatever is beneath it.\n\nThe old rule was that nothing below the top was ever visible, and that\nwas written when stacks were two deep. It stopped being fair once\ndifficulty started buying DEPTH: at forty-two records over ten cells you\nwere committing to pulls with four hidden layers behind them, and a jam\nyou could not have seen coming reads as the puzzle cheating rather than\nas your mistake.\n\nOne layer, deliberately - enough to plan the pull after this one, not\nenough to plan the whole cache.\n\nA NODE NO LONGER DUMPS FIVE TASKS ON ONE PERSON\n\nA node deals a bundle and shares it round the crew, which reads fine at\nfour players: five tasks, nobody holding more than two. Solo, all five\nlanded on you at once, four of them ticking down while you worked the\nfirst, and the console turned into a queue you could not clear rather\nthan work you chose between. The bundle was sized for a full crew and\nnothing scaled it when there was not one.\n\nCapped at TWO PER PERSON rather than special-cased for solo - two players\nsharing five has the same problem in miniature. The node waits for what\nwas actually dealt, so a smaller bundle still completes it normally.\n\nThis one is worth a specific look on a solo run: I could verify it by\nreading the code but not by playing it, because the test harness here\ncannot run a mission.\n\nFIRST BUILD CLICKED BEFORE IT SHIPPED\n\nv0.0.96 and v0.0.97 both went out with every button on the console\nbroken, and both passed every automated check in the project - all of\nwhich drive puzzles through code paths that never touch mouse handling\nat all. There is now a page that puts up real buttons, counts presses\nagainst completed clicks, and scores itself, and it gets run before a\nbuild is published rather than after somebody notices.\n\nYou clicked this one. That is the whole point of it.\n\nWHAT WOULD HELP MOST\n\n1. A solo run, watching how much work a node hands you.\n2. Cache Sift at difficulty 5 now the next layer is visible - is it\n   readable, or is it now too easy?\n3. Bench puzzles at difficulty 4 and 5. Measured, not played.\n4. Badge Swipe, Packet Hop and Sweep - their difficulty is speeds, and no\n   screenshot can check a speed.\n5. Patch Panel at difficulty 5: 81 holes, 7 cables, 60 seconds. I still\n   think that is wrong.",
            "url": "download.php?b=build_20260815_214839.zip"
        },
        {
            "version": "v0.0.99",
            "status": "NOMINAL",
            "filename": "Nominal_v0.0.99.zip",
            "size": 574817251,
            "size_human": "548.2 MB",
            "sha256": "06609ecc7d1b565f21a6595d1c431452c0cfc43c97fc4601d5e643e61bdafc12",
            "uploaded_at": 1786852196,
            "uploaded_iso": "2026-08-15T20:49:56-07:00",
            "downloads": 7,
            "notes": "STATUS: NOMINAL - Build v0.0.99\n\nTwo puzzles you flagged, both fixed rather than struck, plus the bench\nbeing less annoying to test with.\n\nCUT TETHER COULD NOT BE WON ON A LOT OF SEEDS\n\nYou were right, and it was worst at the EASIEST rung. Measured over\ntwenty seeds a difficulty:\n\n  difficulty 1:  9 boards in 20 had NO solution at all\n  difficulty 3:  3 with none, and 7 more solvable only by a pair cut\n  difficulty 5:  none unsolvable, but 9 in 20 needed a pair\n\nUnsolvable boards were dealt anyway. And the \"pair\" ones were arguably\nworse, because they looked fine: the generator proved those by cutting\ntwo ropes AT THE SAME INSTANT, which hands cannot do. You cut one, the\nrig swings, and the second cut happens in a different world.\n\nThe cause was doing it backwards - roll a random target port, then go\nhunting for a cut that reaches it. It now simulates each cut, sees where\nthe payload actually lands, and makes one of those the target. Every\nboard is solvable by construction, always with ONE cut. 150 seeds across\nall five difficulties, zero failures.\n\nTwo-tether boards are gone as well. With two ropes the only cut that\nreleases the payload is cutting both, which is the \"slash everything\"\nanswer this puzzle exists to refuse. Two tethers was not an easy version\nof it, it was a broken one.\n\nAPPROACH WAS UNREADABLE RATHER THAN HARD\n\nEach node you could step to showed \"-26  +40\". Two numbers, no units, and\nnothing on screen saying which was the cost and which the gain - while\nthat pair IS the whole decision. The node's kind was a single initial in\nits own colour on a dark disc, which read as a smudge, and nothing said\nwhat winning was.\n\nNow the numbers are labelled, every legible node names its kind in the\ncolour of its meter in the WATCHING row, and the goal is stated. At\ndifficulty 3 the trade-off reads at a glance: CORE at 26 noise for +40\nprogress, against RELAY at 6 for +6. That was always the puzzle. None of\nit was reaching you.\n\nStill tell me if it earns its place. That part is a play call and I have\nonly made it legible, not proved it is fun.\n\nBACKLOG SHOWS THE NEXT BLOCK\n\nWithout it a falling-block well is a reaction test - you cannot decide\nwhere THIS piece goes without knowing what follows it.\n\nMINEFIELD TILES ARE SQUARE\n\nAt every board size. They were being stretched to fill a fixed rectangle,\nworst at the low difficulties.\n\nTHE BENCH IS LESS TEDIOUS\n\n- Every row starts on its own seed instead of all of them on 1.\n- RND re-rolls on launch and writes the seed it used back into the box,\n  so a board worth reporting can still be reproduced.\n- ? on each row opens HOW TO PLAY for that puzzle without leaving the\n  bench.\n\nWHAT WOULD HELP MOST\n\n1. Cut Tether and Approach, since both were yours. Cut Tether should now\n   always be winnable with a single slash - if you find one that is not,\n   the seed is in the bench box and that is the whole bug report.\n2. Bench puzzles at difficulty 4 and 5. Measured, not played.\n3. Badge Swipe, Packet Hop and Sweep - their difficulty is speeds, and no\n   screenshot can check a speed.\n4. Patch Panel at difficulty 5: 81 holes, 7 cables, 60 seconds. I suspect\n   that is wrong but would rather have your answer than my guess.",
            "url": "download.php?b=build_20260815_204956.zip"
        }
    ]
}