Conflicts resolved: state.rs keeps both the sweep fields (miner-eff) and the race fields (miner-perf); bench-log.md
keeps both appended entries; publish-manifest.sh keeps master's --override implementation (8082576, the "every
height switch" rule, --verify-only, --tries, the retrying live check) and adds miner-perf's --tuning / --no-tuning
with the carry-over of consensus.override and tuning from the current manifest. One --override case, one parser.
The first Windows update over the air (0.3.3 to 0.3.4) sat on 'the installer is starting' with no helper log on
PC 2. The helper launched by a remote job with the same arguments ran at once. Fix in 0.3.5; the Windows machines
take 0.3.5 by hand once more.
Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
Header: site/partials/nav.html, head.html, footer.html, injected by site/build.mjs between markers in every page (the bench template too) with the active link marked; the master mark in its black square, the wordmark, Litepaper, Live devnet, Engineering log, Evidence, GitHub and the miner button, in that order everywhere; one menu script (closes on a link, a tap outside, Escape with focus back on the button, a resize past 900 px); a skip link; one focus ring and one reduced-motion rule.
Litepaper: the hash routed by a click delegate and popstate instead of the browser's fragment jump. Root cause of the tabs fault: show() scrolled to the top of the article, never to the section; in whole-paper mode the fragment jump landed on the section and the smooth scroll then dragged the reader back to the Abstract; in one-section mode a deep link scrolled while the section was still display:none and landed on the cover. Now: every section and all 29 subsection headings reachable by URL in both modes, scroll-margin from the measured bar height (nav, plus the contents row on a phone), focus moved to the heading, a re-aim after a late font batch, mode buttons stacked in the sidebar.
Pages: homepage headline capped at 9cqw so it is two lines at 390 and 1440 (was four with "fire." alone), journey inlined by the build (no fetch, no shift), dates as "4 Oct 2026"; live page legend no longer forces 420 px of width at 390, the hash-rate unit as small mono, idle redraw at 30/s with a 1.5x backing store (0.7 to 1.2% of one core idle in Chrome for Testing, taken under load), the proving line wraps on a phone; evidence sort headings are buttons, chips carry aria-pressed, a scroll hint under 1,140 px; bench contents in a sticky scroll box, og.png?v=3; 404.html; sitemap with /evidence; vercel.json caches fonts a year and images a day.
Fonts: eight latin woff2 files in site/fonts (139 KB for all, 118 KB a typical page), preloaded first-paint faces, fallback faces with size-adjust and ascent overrides from the font tables; every page lost its render-blocking fonts.googleapis.com stylesheet and the gstatic origin.
HTML per page before to after (gzip): / 17,265 to 21,810 B; /litepaper 20,556 to 24,957; /live 15,640 to 17,777; /evidence 18,899 to 22,493; /bench 94,723 to 98,289. Third-party requests before to after: 3 to 1 (jsdelivr, light client), 1 to 0, 3 to 0, 1 to 0, 1 to 0. Lighthouse before and after is queued under the measure lock (held by reliability runs since 19:16Z, load average 258) and recorded in docs/design/site-polish-2026-10-04.md when it runs.
Checks: node site/build.mjs, tools/ci/link-check.mjs (246 links, 0 broken), identity grep, and a Chrome for Testing run over every page at 390, 768 and 1440 (no overflow, no console errors, menu, router in both modes, deep links on reload, keyboard).
Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
src/sweep.rs (new): the cap steps 100% to 50% in 10% steps clamped to the card's limits, per-step rows (mean draw
from nvidia-smi power.draw, mean worker interval rate), the choice (best MH/W, ties to the higher rate then the lower
cap), the nvidia-smi power parser, a state machine on an explicit clock (15 s settle, 60 s hold, 30 s cap readback
limit), the elevated helper scripts (one administrator prompt per sweep: a command file polled by one elevated
process, self-restoring after 20 idle minutes), the unsupported reasons (Apple silicon, AMD). 9 unit tests with
PC 1's recorded RTX 5090 numbers (575 W default, 460 W cap, 290 W draw, memory temperature [N/A]).
Engine: scheduler (once after install, then weekly; one card at a time; only while the card mines, after 120 s
steady, never under a remote job hold, a pause, or inside 600 s of the hour boundary), the cap-mode probe (direct
when the engine runs elevated, else the helper), abort on any fault (card leaves mining, worker error, GPU 90 C,
job, pause, quit) with the cap restored, the chosen cap held and recorded, SWEEP table lines in the app log,
--sweep mode (sweep every supported card, print the table on stdout, leave the caps, quit). Cap floor 50% (was 60).
A readback that matches the asked cap now counts as applied (PC 1 showed "cap NOT applied" for hours at 460 W).
Dashboard: live eff MH/W on each tile, the sweep line (phase, last result, or why unsupported), Sweep now / Stop /
Unpin, "pinned" and "chosen by the sweep" on the cap line, the Settings toggle, the cards-page note, slider min 50.
A cap moved by hand pins the card: the sweep records but does not change it.
PC 1 measurement: relay/playbooks/sweep-5090.ps1 (a run job, elevated, miners stopped; a second engine with --sweep
in a scratch data folder, RESULT SWEEP lines) and docs/plans/miner-eff.md with the publish command. Not published.
Untested on a card.
Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
Under emulation (or --race with no other name) the race no longer times base alone: emu/test.sh passes again
(9 source checks, the serve protocol with prepare, swap and self-heal, 17 sampled hashes equal to igneum-pow).
Mac table: g256 (256 threads per threadgroup) +17.3% and +21.2% over the shipped 32-thread groups on two
programs, with the live app's worker sharing the GPU; the serve check found the unfair mutex (fixed in 123ee1a).
Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
Log drawer: the Logs button shows its open state (Close logs, chevron); a Newest button appears when the user scrolls
up and follow re-engages at the end; a time ruler with error and swap marks, an HH:MM:SS jump box, last error and last
swap buttons; substring search with a match count and highlights; copy the lines in view; the height drags from a grip
(140 px to 70% of the window, remembered); a window under 700 px tall opens a 180 px drawer. The list is virtualised
(19 px rows, only the rows in view in the DOM; up to 20,000 lines kept; wrap mode under 5,000 lines). Polling runs only
while the drawer is open and the window is visible.
Screens: one type scale and spacing scale in app.css, tabular figures everywhere, the bottom bar in pieces (machine name
truncates first, address always short, uptime dropped under 1100 px), tile captions carry their full text as a title,
syncing shows an ETA from the measured block rate, the engine-away state names itself on the node tile, short-window
rules, the key sheet scrolls on a short window, compact settings card rows on two lines, only a block that just arrived
flashes (not the whole strip on first paint).
Performance: the minute grid is drawn once into an offscreen layer; the strip redraws twice a second and at the display
rate only during a flash; nothing draws and the state poll drops to every 5 s while document.hidden. ?debug=1 prints a
budget line every 5 s and exposes window.__igneumBench.
tools/ui-mock: a stand-in engine (node tools/ui-mock/server.mjs) replaying recorded, scrubbed API responses with
scenarios (syncing, nodedown, nocards, integrated, clock, paused, biglog, fresh, job, nvidia), so UI work never
touches a running miner.
Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
The engine turns a worker's race line into one TUNING {json} line in the app log (card model as the worker names it,
driver, arch, program class loads and wide loads, every variant's MH/s, winner, gain, the card's power cap and
draw, MH per watt), which the existing intake receives; the card state carries the variant for the dashboard and
one event per race. tools/tuning.mjs aggregates the records from miner_logs per card model (median MH/s or MH per
watt, at least 3 samples, de-duplicated per race) and writes tuning.json; publish-manifest.sh --tuning puts it in
the signed manifest (and now takes --override for consensus.override; both are carried over from the current
manifest when not given, --no-tuning drops it); manifest.rs parses it; ota.rs writes <app data>/tuning.json and
removes it when the manifest drops it; procs::spawn takes an environment and every miner starts with
IGNEUM_TUNING_FILE, which its worker reads at every prepare. Dry run of the publisher against a scratch folder:
tuning and override written, carried over, dropped, signature verified.
docs/plans/miner-perf.md: the signed jobs for PC 1 (fetch the race build of the NVRTC worker, then
relay/playbooks/race-5090.ps1 with the miners stopped: 17 variants, 3 rounds, twice) with the exact publish
commands for the main session; not published by the agent.
Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
the project lead, 4 Oct 2026 evening: "we need to make our miner better than anything else can be". The compile-ahead pipeline
built one kernel per program; it now builds a catalogue (unroll 2 or 8; the dataset load path __ldg, __ldcg, __ldcs
on NVIDIA; a register budget by --maxrregcount or __launch_bounds__, max_total_threads_per_threadgroup on Apple;
2, 4, 8 warps per block or 64 to 256 threads per threadgroup; combinations), self-tests each against the pack's
vector warps (NVIDIA) or the base kernel over 2^16 nonces (Metal), bit for bit or out, and times each for about two
seconds with the job loop paused (one mutex, mining resumes between variants). Base is the pack's text as shipped,
always first, never discarded; a race has a budget (default 120 s against the 600-DAA lead) and keeps the best so
far when it runs out, so the swap is never delayed. One line per race: variants, MH/s each, winner, gain, time.
A tuning file (--tuning, IGNEUM_TUNING_FILE) pins a variant or orders the candidates per card model.
NVIDIA: textual rewrites on the pack's own kernel_bound.cu with exact anchors from igneum-pow's emitter, so the
pack format, the miner and igneum-pow are untouched and old packs race. --race --pack <dir> runs the race alone.
Under IGNEUM_EMU the race is off (the stand-in checks the handed-over text is the pack's). Metal: the MSL hooks in
generateMSL, a lock-protected kernel slot per program, a pair compiled inline races after its first job,
--race-test --seed --day runs the race alone with a table. OpenCL is not raced yet. docs/design/miner-tuning.md.
Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
