Project Status

Status snapshot: 2026-10-17 — all three branches (research, dev, main) synced at 989c5c3 (link-debt repair round landed on research), validation passing. The October rounds rebuilt the Linux Interface Encyclopedia per-function, added the process-exit-signal part and the Section G eight-language process/runtime part, wired all 295 orphaned pages into navigation, and cleared 130 mdBook HTML-tag warnings. See “October 2026 rounds” below.

Link-debt repair round — 2026-10-17

One round closed the oldest known-debt item: all 217 broken internal links/anchors (scripts/check-links.py reports 0 on the merged tree). The repair created the 92 missing pages the links pointed at — 91 per-function syscall pages across the seven encyclopedia domains (process 6, fs-mounts-namespace 8, signals-time 11, sched-misc 65, net-ipc 1) plus the musl catalog page the musl ledger references — fixed two wrong relative paths (process-exit-signal-coverage.md to SUMMARY.md; the Section G language index to src/languages/process-runtime/), and repaired a 188-byte null-byte corruption inside sched-misc/overview.md’s domain table (present since the wave-3 rebuild; verified byte-identical back to commit 4cab223, clean at 89274e6). Entries were written against primary sources fetched for this round: the man-pages 6.19 HTML set on man7.org (53 pages fetched, cached), torvalds/linux kernel sources for the arch-specific and 6.13+ syscalls (fchroot, listns, file_getattr/file_setattr, the xattrat family, rseq/rseq_slice_yield, ARM/alpha/sparc/m68k/openrisc privates), and local nm -D on glibc 2.41 for every wrapper claim. 45 ledger rows flipped source-limited → documented (syscalls now 433 documented / 46 researched / 56 source-limited of 535), the domain overview tables and the hub counts were reconciled against the ledger (fixing pre-existing staleness in the process/fs-mount/net-ipc rows), and all 92 pages are wired into SUMMARY.md (4,211 content pages, 0 duplicates, 0 dead entries).

October 2026 rounds — 2026-10-12 → 10-13

Four back-to-back rounds moved the book from 3,385 to 4,119 content pages and closed out the navigation and warning debt:

• Linux Interface Encyclopedia per-function rebuild (2026-10-12, waves 1–7). The old per-domain mega-chapters were replaced by one page per syscall, libc function, sysfs class and cross-layer interface: 207 files under src/linux/syscalls/, 292 under src/linux/libc/ (glibc + musl), 121 under src/linux/sysfs/, 5 under src/linux/crossref/. Coverage ledgers are regenerated from sources and status-honest: syscalls 388 documented / 46 researched / 101 source-limited of 535 (0 not-started); glibc 1,047 documented / 10 source-limited of the 1,881 documented-API rows; sysfs 106 documented / 14 source-limited of 120 classes; crossref 37 documented / 1 source-limited of 38.
• Process-exit-signal knowledge base (2026-10-12). New part src/linux/process-exit-signal/ (43 pages): unified coverage ledger (174 requirement IDs), signals inventory + interfaces, exit-status lifecycle, shell chapters (bash/dash/zsh), service managers (systemd/OpenRC/runit/s6), containers (PID 1 semantics, Docker, Kubernetes), and a troubleshooting decision tree.
• Section G — eight-language process/runtime semantics (2026-10-13). src/languages/process-runtime/ (18 pages): one chapter + one coverage ledger per language (C, C++, Python, Go, Java, Rust, Node.js, OCaml — chapters 8.5k–10.4k words each), part hub, and a 9-dimension cross-language comparison. G1–G6 audited: 633 ledger rows (623 documented / 10 source-limited / 0 remaining), zero duplicate IDs; a G3 reference-policy fix added 144 verified hyperlinks to the C/C++/Rust/comparison chapters; all 388 intra-book links resolve; 290 of 291 external URLs reachable (en.cppreference.com = documented Cloudflare 403).
• Navigation integrity & warning cleanup (2026-10-13). Wired the last 295 orphaned pages into SUMMARY.md (binaries overview + coreutils/util-linux flatlands rebuilt into a designed hierarchy, sysfs Class Encyclopedia, musl catalog, crossref maps; Section G ledgers became real nav children, fixing 8 live-site 404s; 14 bogus part titles removed; the glibc hub repositioned — it had been splitting 282 pages under the musl ledger). SUMMARY.md now has 4,116 nav entries, 0 duplicates, 0 dead entries, verified by a real mdBook v0.5.4 build (create-missing=false) over 4,119 pages. Then all 130 mdBook “unclosed HTML tag” warnings were eliminated by wrapping bare <token> placeholders in inline code across 33 files (negative control: mini-build 130 warnings before → 0 after).

Userland binaries round — 2026-10-09

• New part: src/linux/binaries/ — 280 pages, 16 collections, one page per binary (coreutils 104, util-linux 96, shadow 24, login 8, procps 18, findutils, diffutils, grep, sed, gawk, tar, acl, hostname, bsdutils, busybox, uutils), each with a package-level hub.
• Validation after the round: 3,385 pages reachable via SUMMARY (0 duplicates), 0 broken links/anchors, MathJax clean, 5,527/5,527 Mermaid diagrams pass the real mermaid@11 parser.
• References: 285 per-binary manpage URLs verified live (manpages.debian.org, incl. the post-split Debian package mapping), sources.debian.org, GitHub mirrors, BusyBox manual, POSIX 2018 spec.

Current status

All branches synced and repaired. research, dev and main point at the same commit (989c5c3); the repository contains 4211 content pages under src/ (+SUMMARY.md), all reachable from SUMMARY.md (4208 nav entries) with 0 duplicate destinations and 0 dead entries — the 2026-08-era orphan problem is closed.

A full external review on 2026-09-19 re-ran every validator from scratch, added a real mermaid@11 parse, probed all ~6,500 external URLs, resolved every DOI, and read high-value pages by hand. It found and fixed: a code-fence corruption class that rendered prose as code, wrong worked-example mathematics, a MathJax/link collision, 48 site-root-absolute links, malformed URLs, conflicting kernel sizes, stale counts in this file, and missing CI. See “Deep review & fix pass”. The October 2026 rounds (above) then rebuilt the interface encyclopedia, added the process/exit/signal and Section G parts, wired the last 295 orphaned pages, and cleared 130 mdBook HTML-tag warnings.

Known debt

• 217 broken internal links/anchors (scripts/check-links.py, full run 2026-10-15). Proven pre-existing: a stash-baseline run during the 2026-10-13 warning round produced the identical 217 before and after the change, and none sit in files touched by the recent rounds. They need a per-link decision (re-point, retarget, or drop) and are the next maintenance candidate.
• Source-limited ledger rows (documented, not hidden): 101 syscalls, 10 glibc APIs, 14 sysfs classes, 1 crossref row, 10 Section G rows — each labelled in its ledger with the reason (primary source unreachable from the work environment).

Reference libraries (2026-10-09). Fourteen verified primary-source indexes, one per major topic, each carrying the same five parts: numbered entry categories, a two-track Education section, a research-papers section, a Video courses/channels/talks section, and a Conference videos/notes/archives section. A new fourteenth index (DSA, competitive programming and competitive math) joined the set this week. Per-index counts, the CSV list and the verification dates live in src/references/README.md; the headline totals there (1,257 entries, 2,744 unique URLs) are recomputed from the pages rather than carried forward, because the previous totals did not reconcile with their own per-index column.

Validation commands

Latest full-suite runs on the merged tree (2026-10-15 unless noted):

• scripts/check-summary.py → 4208 nav entries, 0 duplicate destinations, 0 dead entries
• scripts/check-links.py → 0 broken links/anchors (2026-10-17, after the link-debt repair round)
• scripts/check-fences.py → 0 problems across 4,120 files
• scripts/check-mathjax.py → 0 unbalanced delimiters, 0 link-in-math hazards
• scripts/validate-mermaid-heuristic.mjs → 5,546/5,546 pass (100%)
• real mermaid@11 parse → 5,527/5,527 pass on 2026-10-09 (last full run; 19 diagrams added since)
• real mdBook v0.5.4 stub build → exit 0, all 4211 pages render (2026-10-13; re-run before release after the 2026-10-17 round)
• scripts/check-doi.py → 491/491 DOIs resolve (2026-09-19 run)

External URLs are not fully clean. A sweep of all 8,638 external links found 6,112 live and 630 dead (HTTP 404/410); the remainder returned 403/429/bot-challenge and are counted unverifiable, not dead. Six repair passes have fixed 399 dead links — each fetched and title-checked first, or backed by a verified archive snapshot — including all 471 kernel.org links, now 100% live. Of the rest, 63 are intentional placeholders and 395 are genuine broken links catalogued by host in scripts/dead-links-report.txt; a further 116 404ing URLs live only inside fenced code blocks and are correct as written.

Automated resolution is now exhausted, and the report records what was tried and measured: structural URL variants (2,740 candidates → 7 fixes), host-alias rewrites (638 → 0), Crossref metadata matching against 88 dead PDF citations (5 confirmed), and harvesting the sitemaps of 184 live hosts (102,590 URLs → 16 confirmed; a sixth pass added OpenAlex metadata matching, GitHub API checks and archive CDX resolution). The 395 that remain need a human decision each — re-point, substitute, or drop — because no equivalent page exists to point at. substitute, or drop — because no equivalent page exists to point at.

An earlier revision of these figures was overstated: the URL extractor stopped at the first ) and bled across ], so every URL containing a parenthesis was measured truncated and reported dead. A Markdown-aware extractor fixed this and recovered 20 links (14 DOIs, 5 Wikipedia articles, 1 testcontainers page) that were never broken.

The full mdBook build peaks above the memory limit of the environment used for this review, so the review ran a 47-chapter subset build of every changed file (clean, zero warnings). Run the full build locally before release.

Validation is a local/agent step, not a CI step. CI in this repository is reserved for builds and deploys (.github/workflows/deploy.yml); adding validation jobs would consume hosted minutes on every push. The scripts under scripts/ are meant to be run by a human or an agent in a dev environment.

Research-branch additions (2026-08-02 → 2026-09-02)

The research branch added 75 commits and 660 new files. Highlights:

• Batch-style deep-dives (batches 1–66): each batch produced 3–4 high-quality, heavily-cited pages on topics spanning scheduler internals (EEVDF, CFS, EASY/FCFS backfilling), RCU torture testing, transient-execution attacks, Zobrist hashing, pdqsort, NVIDIA MIG, QUIC congestion control, Haystack/SeaweedFS, LMAX Disruptor, GPU warp scheduling, CUDA Graphs, Intel RDT/resctrl, and many more.
• Citation discipline: Crossref-verified DOIs (e.g. Spectre 10.1109/SP.2019.00002, Mu’alem-Feitelson TPDS’01 10.1109/71.932708, Musser SPE 1997, Lindholm IEEE Micro 10.1109/MM.2008.31), RFC-verbatim QUIC frame types and header-protection sample sizes, and Slurm bf_* parameter names quoted verbatim from sched_config.html.
• Demo QA: every CLI / code demo is byte-exact against an instrumented reference run; batch commit messages record the byte-identical stdout.

Original major expansion (2026-08-15)

20 Advanced Topic Sections (A–T)

Build-It-Yourself Track (34 projects)

• OS: kernel, scheduler, allocator, filesystem, shell, TCP stack, debugger, eBPF tracer
• DB: B-tree, LSM tree, WAL, MVCC, query optimizer, vector DB, distributed KV
• Distributed: Raft, Paxos, gossip, consistent hashing, distributed lock, replicated log, scheduler
• Networking: TCP, HTTP/1.1, HTTP/2, DNS resolver, reverse proxy, load balancer
• Compilers: lexer, parser, interpreter, bytecode VM, optimizer, toy JIT

Safety constraints

• Development work is performed on research (and dev before promotion).
• Release promotion to main occurs only after validation.
• Credentials are read only at command time and are not stored in repository files, commits, or documentation.

Deep review & fix pass — 2026-09-19

An independent review of the whole tree (own tooling, real Mermaid parser, live HTTP probing, DOI resolution, manual page reads) produced the following fixes. Each is verified, not asserted.

Correctness

Links

Structure

• Renamed os/virtual-memory/page-rejection.md → page-replacement-overview.md (the filename was a typo for “replacement” and the SUMMARY label hid it).
• Fixed both SUMMARY.md indentation jumps (4-space children among 2-space siblings) that mis-nested the page-replacement and query-processing groups.
• Disambiguated the three page-replacement pages and the two query-optimization pages with distinct nav labels and mutual cross-links; retitled the thin page Page Replacement: Interview Guide since its H1 collided with the deep dive.
• Expanded mobile/README.md from 373 bytes to a full section landing page.

Tooling & process

• New scripts/check-fences.py — catches malformed openers, unclosed fences and nested fences with CommonMark-correct length rules. Wired into validate-all.sh as step 2/7. Positive controls confirm it rejects all three bug classes and accepts 4-backtick nesting.
• Hardened scripts/check-links.py — now flags site-root-absolute links and non-.md relative targets, and strips inline code so C++/Go samples like Sum[int](myInts) are not misread as links. Explicit allowlist for the CI-generated cross-reference-graph-view.html.
• Hardened scripts/check-mathjax.py — detects ]( inside math spans (the ppo.md bug class), and its always-true if dollars == line.count("$$") condition is gone (deduplication already happens downstream).
• CI claims corrected, CI left build-only — the README badge, CONTRIBUTING and scripts/README.md variously implied that links/SUMMARY/MathJax/Mermaid were validated in CI “and weekly”, while the repository’s only workflow is the main-only Pages build+deploy. Rather than adding validation jobs (which would bill hosted minutes on every push), the documents now state plainly that the suite is run locally by a human or an agent, and the badge says so too.
• book.toml edit links now point at research (a strict superset of main), so they resolve for research-only pages instead of 404ing.
• .gitignore no longer hides package.json/Cargo.toml, which had made the Mermaid-parser toolchain unreproducible from a fresh clone; added the real-parser report path.
• scripts/dead-links-report.txt — the 758 dead external URLs grouped by file, as an actionable backlog.

Deliberately not changed

• The 74 colliding-topic page groups (backend/ vs distributed/, os/ vs concurrency/, …). Spot-checks showed the duplicates agree in substance, so merging them is an editorial decision, not a correctness fix.
• IEEE TBD, 2021 in vector-databases.md and rag-advanced.md is not a placeholder: TBD = IEEE Transactions on Big Data.
• The #### sub-headings under ## sections in linux/reference/further-reading.md are a consistent file-local style (all 79 entries), not a stray level skip.