packfile.h reads IGNEUM_STATE_LEAVES, IGNEUM_STATE_LEAVES_FNV64, the file name, root and block (generator 5 = class v5; a generator 5 pack without the count, or another class with one, is refused) and pf_load_leaves reads and checks leaves.bin; the loader test carries the class v5 cases known-failed first (no count, no file, a short file, one flipped bit, a v4 pack with a count) and the checked-in v5-dn3-epoch0 pack. The CUDA driver gains cuMemcpyHtoD; the emulation takes both igneum_build shapes (weak overloads) and refuses a shape mismatch. The OpenCL host sets the build arguments in one place for both shapes, uploads the leaves in --bench-pack, --serve and the compiled-in bench (the host reference derives words from the same leaves). The Metal worker's servePackDataset binds buffers 2 and 3 as packbench.swift does and self-tests the dataset head and last word against vectors.json. tools/class-v5: kits-remote.sh (the kit build and tests on igneum-build-1, the zip to /srv/artefacts/packs), fleet-cuda-v5-bench.sh (the fleet card), pc1-amd-v5-bench.ps1 (PC 1's 9070 XT through the coordinator).
Measured on the Mac under the measure lock, 7 October 2026: v5-dn3-epoch0 fingerprint 82b19cbde8557ea5 on Metal (packbench, M5 Max, 18:34:47Z, 14.163 MH/s GPU time) and on Apple OpenCL (--bench-pack, 18:34:59Z, 15.265 MH/s wall; the compiled-in bench's dataset self-test PASS at 18:34:54Z), cache FNV 7334fa46e5d972eb PASS, vectors 3 of 3 on both.
Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
Home row 03, the litepaper's chip section at /litepaper#chip-model (paragraph and table; the class v3 row last, "never the launch
state"), the miner page's line and evidence.md row 17 in its eight columns: 2.1x to 3.9x under class v4 from the first block,
class v5 removing the stored-dataset chip as a category, the 5x to 9x only as the class v3 baseline, the devnet's activation
height as a devnet fact. The litepaper's abstract carried the same claim in the old form ("5x to 9x today; class v4, now on
the vote") and now reads launch-first too. tools/ci/ledger-text-check.mjs follows the sentences (X35 on the home page and the
litepaper). No disclosure prize anywhere.
Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
the project lead: the ember band on /wallet had no vertical padding (the heading on the band's top edge, the button on its bottom), then
"run a full site sweep for padding and visuals".
- The band: .band .wrap carried the padding and the three bands use .container; both selectors now, 56 px (40 on phones), and
24 px at the sides on phones.
- tools/ci/padding-check.mjs: renders every served page at 390, 768, 1024, 1440 and 1600 in both themes and flags TIGHT (text or
a control under 24 px from the edge of the filled band, card or section that holds it; rows of radius 12 and under are the
app's compact rows and keep their own padding; small filled things such as buttons, pills, badges, tabs, cells and the bar are
never containers), SCALE (a section's vertical padding not 0 and not a multiple of 4 from 24), TOUCH (controls under 6 px
apart outside segmented groups), OVERRUN (an image or canvas past its frame) and UNDER-BAR (a heading under the sticky bar on
load, an anchor the bar would cover). --self-test builds the fixture from the site's own stylesheet and bar with the exact
/wallet band with its padding lost, a touching pair, an image over its frame and a heading under the bar; each must fail,
the clean twin must pass. A gate line runs it (self-test, then the built site); where no browser exists it says so.
- Findings fixed (first run 2,740, then 1,040, then 48, now 0 at 230 renders): page heroes 64/48 (were 65/45), the
observatory head and the 404 the same, the footer's bottom 24, doc sections and the litepaper's 40 (were 42), .sec 72 at
tablet (was 70), the 404 card 36, /live's motion block 24 (was 25), the scenes lab 48 (was 50); every card at 24 inside
(download cards, home tiles, the live tiles and stats, the evidence labels, the wallet state cards, the journey phases, the
live inspector and toolbar, the download card on phones).
- The eye pass over every page at 1440 and 390 dark: the journey's phases are cards now, so the old timeline rail and marker
(drawn for an unboxed list, sitting on the card's top border) go and the when column keeps clear of the body; tables keep
words whole (break-word, not anywhere: "regenerat es" on the litepaper) and give the label column 14ch (one word per line on
/randomx); the litepaper's body paragraphs share one measure; the ledger's count-table chips stay on one line; orphan words
go through text-wrap: pretty on body text (copy law 18, "measurement." alone on /claims); the dev-fee page lost its doubled
gutter (a lifted section keeps its own .container) and its table clears the text above; /faucet and /metamask keep a gap
between headline and lead, line their two cards up at the top and end 72 px above the footer; a heading after a card grid
gets 40 px.
- The overlap sweep at the same five widths and both themes: 0 overlaps at 290 renders.
Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
Suite at c79f6d96 on box 2: 103 passed, 0 failed (lib 140.8 s: every class v4 draw in the tests pays the 2^20 ratio pass). CI 37639406567 success. Census: 4,099 programs, 0 lossy sites of 65,584, 0 exhaustions, mean attempt 2.086 (c709afef 1.998), max 17; the ratio refuses about 4 percent of candidates that pass every other test. tools/ca3-v4-uniform draws the seeds across the available cores (the serial run became a four-hour job under the ratio pass); the analysis and the row order are unchanged.
Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
The first 16 GB card and the first Thunderbolt card. The app's first poll listed it (vendor nvidia, 16,311 MiB,
driver 610.47 shared with the 5090, kind discrete: the app does not know it is an eGPU) and started a miner on
it by itself. Job run-5060ti-bench-20261007-b (tools/bench-5060ti/pc2-5060ti-bench.ps1, the card alone under
the runner's --cards-off, the 5090 mining, no power limit written: 180 W default read back unchanged, PC 2 lost
power twice that day): G1 PASS with 867dbc45cfb36b4d (sub-version 1) and 90f794dd556f7a3b equal to the Mac's;
the window 1,105 dispatches of 2^24 in 602 s at 30.882 MH/s, 114.8 W mean at utilisation 90 percent and over,
0.269 MH/W, PCIe 4.0 x4 through the enclosure (the hash is latency-bound, the link costs nothing measurable).
Owed: the efficient point (the app's Ember Tune; PC 2's Power Helper refuses every request since the restart)
and the prove-beside read (the shipped WSL2 host has no IGNEUM_CUDA_DEVICE selector; the floor server is on
PC 2; the prover-floor host is the cut). The kit (tools/bench-5060ti/make-kit.sh) takes the sub-version 1 packs
from ca3-v4-amend aa31c396, since master's packs-ca3-v4 is the 6 October stream.
Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
the project lead's three on the live site, plus the /live key's last item:
- Mobile nav (URGENT): the sheet was position:fixed inside the bar, and the bar's backdrop-filter is a containing block for
fixed descendants, so top and bottom resolved against the 60 px bar and the sheet opened with zero height while
aria-expanded said true. It is now absolute under the sticky bar, a viewport-tall box (100dvh minus the bar), display block
over the package's grid rule. tools/site/sheet-test.mjs opens every page at 390 px, clicks the burger and asserts the first
link is on screen and on top (elementFromPoint at its centre returns the link); --self-test forces a zero-height sheet first
and must catch it; a gate line runs it where a browser exists (the box, CI) and says so where none does. On build-2: the
zero-height sheet is caught, 23 pages pass.
- Home at 375: the renderer's "Waiting for observer data" was drawn on the fold's canvas under the lead. The fold now carries
.pending until the feed is live or stale and keeps the canvas at opacity 0 meanwhile; the status pill is its own row under
the buttons on phones. ?feed=off holds the pending state for a render; the overlap sweep (whose server answers /api with
503) renders the pending state by construction and ran at 375, 390 and 1440 in both themes: 0 overlaps.
- /miner and /wallet: the light-theme screenshot variants are gone (dark only), captions unwrapped, "light and dark" dropped
from the gallery line; one gallery of frames with one caption each.
- /live: "Your block" leaves the key, the "Your blocks" filter and tile and the Mark-my-key prompt leave the page (the site
has no "you"); the stats row is three tiles; the key reads Included, Excluded, Selected chain, Proven, Locked checkpoint.
- Captures: ?motion=off (head partial) marks data-motion=off so every .reveal is drawn in a headless render; capture.sh passes it.
tools/site/serve.mjs survives a refused api origin (502, never a crash).
Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
