43 new pages added to src/research-projects/ (follow-up tree +8 chains,
assumptions hunt +7 audits, connection atlas +7 connections, why-not handbook
+7 comparisons, failure casebook +6 incident case histories, misconceptions
+8 audited claims), taking the part to 95 pages / ~220K words. New casebook
incidents (Ariane 5, Heartbleed, GitLab 2017, CrowdStrike 2024, Fastly 2021,
AWS us-east-1 2021) each reconstructed from primary sources with CONFIRMED vs
INTERPRETATION separation. Part description line removed from SUMMARY.md per
user feedback. Scoped validation: 0 broken links, 0 fence problems, Mermaid
5,563/5,563 pass, SUMMARY navigation OK (4,306/4,307, only SUMMARY itself
unreferenced). Per-project Coverage Ledgers updated with sources and gaps.
280 new pages under src/linux/binaries/ (one page per binary across 16
collections, hubs included), wired as the new top-level SUMMARY part
“Linux Userland Binaries”. Page counts: 3,105 → 3,385 content pages.
Words ~7.09M → ~7.73M. Unique external URLs 12,035 → 12,370. Mermaid
diagrams unchanged at 5,527 (new pages favor house-style ASCII diagrams);
all pass the real mermaid@11 parser. Validation: 0 broken links, 0
duplicate SUMMARY destinations, MathJax clean. Reference URLs verified
live the day of writing (285 per-binary manpages + source browsers).
The research branch (75 commits ahead of main) landed 660 new markdown files (~174K lines) and 66 batch-style deep-dive pages. All counts below are from the research tree @ 9c249de.
Step
Result
Batches 1 → 66
Each batch produced 3–4 heavily-cited pages (Crossref DOIs, RFC verbatim, source-code-verified constants). Topics span schedulers (EEVDF, 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 more.
Summary navigation
SUMMARY.md grew from ~2,116 refs (2026-08-16) → 2,776 refs at 9c249de (0 broken, 0 duplicate destinations).
Mermaid diagrams
4,889 diagrams across 1,328 files, 100% passing the real mermaid@11 parser (not only the heuristic).
MathJax
balanced across 128 math pages.
Demo QA
every CLI / code demo byte-exact against an instrumented reference run; commit messages record the byte-identical stdout.
Citation discipline
Crossref-verified DOIs (Spectre 10.1109/SP.2019.00002, Mu’alem-Feitelson 10.1109/71.932708, Musser SPE 1997, Lindholm 10.1109/MM.2008.31), RFC-verbatim QUIC frame types and header-protection sample sizes, Slurm bf_* parameter names from sched_config.html.
Audit & QA fixes
Deep-read audit (2026-09-02) confirmed 100% coverage of all 1,528 prompt.md topics and all 1,374 index.md bullets. Audit-driven fixes: Meltdown CVE corrected to CVE-2017-5754, GrapheneOS→Graphene/Gramine libOS clarification, GPT-4 175B→GPT-3 175B, mirror-clock formula corrected to 11:60 − H:M, four aptitude worked examples rewritten without leaked AI monologue (calendar, mirror, boat, escalator, ages), four DSA worked examples repaired (ch36 diff-array reconstruction, ch178 Burnside cube, ch121 Lyndon minimal rotation, ch169 min-cost-max-flow), EEVDF benchmark block labelled illustrative, and meta/* refreshed.
references/dsa-competitive-programming.md — 170 entries, 72 education resources (37 basic / 35 advanced), 30 research sources, 402 verified URLs, wired into SUMMARY.md and references/README.md.
Video sections
195 resources across the fourteen indexes (channels, conference recordings and lecture playlists), every channel and playlist fetched and title-checked.
Conference sections
86 sources across the fourteen indexes: conference channels and video archives (USENIX, FOSDEM video.fosdem.org, media.ccc.de, Linux Plumbers, netdev, LCA, PyCon US, LLVM dev meetings) plus proceedings and note archives (SIGCOMM, SIGPLAN, SIGMOD, VLDB, NeurIPS, ICLR, ICML, CVF, EuroSys, IEEE S&P, SNIA, ;login:).
Rejected rather than guessed
~35 bad YouTube handles (wrong identity or 404), 4 dead hosts, 2 playlist IDs that resolved to “undefined”; 3 bot-blocked sites kept and flagged.
CSVs
data/*.csv extended to one row per resource — entries, education, video, conference — including education rows for the four indexes that were entry-only.
11 of 12 source-limited rows upgraded with fetched upstream docs (NetworkManager, thermald, UPower, libinput, libdrm, rdma-core, ndctl, procps, dmidecode) plus local NSS/kernel sources; XR-112 stays source-limited (msr-tools docs unreachable)
Validation
every new page ASCII-only, balanced fences, relative links resolve; statuses/notes flipped in all four ledgers; SUMMARY wired (282 glibc entries, 119 class entries)
633 rows inventoried: 623 documented / 10 source-limited / 0 remaining; zero duplicate IDs; rollup numbers match ledger contents exactly
G3 reference policy
C/C++/Rust/comparison chapters cited sources by name only (Rust had zero URLs); 209 URLs live-verified, 144 hyperlinks added -> all chapters carry direct source links
295 authored-but-unwired pages joined SUMMARY.md: binaries flatlands rebuilt (overview -> 16 collection hubs -> pages), sysfs Class Encyclopedia + crossref maps added, glibc hub moved to block top (it split 282 pages under the musl ledger), 14 pseudo-part titles promoted, Section G ledgers became real nav children (fixing 8 live-site 404s)
Summary state
4,116 nav entries, 0 duplicate destinations, 0 dead entries; real mdBook v0.5.4 stub build with create-missing=false exits 0 and renders all 4,119 pages
CI fix
duplicate languages/comparison.md entry that broke the deploy replaced with a real language-chapters hub page
Warning cleanup
130 mdBook “unclosed HTML tag” warnings eliminated: bare <token> placeholders (ttyS<x>, SIG<NAME>, <PCI BDF>) wrapped in inline code across 33 files; negative control mini-build 130 -> 0 warnings, 0 errors
Link check
check-links.py reports 217 broken links/anchors before AND after (pre-existing repo-wide debt, none in touched files; stash-baseline verified)
Fence check
check-fences.py clean across 4,120 files
Branch sync
research -> dev -> main via explicit --no-ff merges, then all three branches fast-forwarded to the same commit (989c5c3)
+7 comparisons: REST vs gRPC, ORM vs raw SQL, TCP vs UDP, sync vs async I/O, ID generation (auto-increment/UUID/Snowflake), row vs column stores, password hashing choices
Failure Casebook
+6 verified cases: Cloudflare control plane (Nov 2025), AWS DynamoDB us-east-1 (2015), Mars Climate Orbiter (1999), Toyota unintended acceleration (2009-2011), UK Post Office Horizon (1999-2025), Boeing 737 MAX MCAS (2018-2020)
39 new chapters across the six research collections (expansion round 18)
Follow-up Tree
+7 chains: load balancing, replication lag, HTTP protocol evolution, API gateway, virtual threads vs async, zero-downtime migrations, SLOs and error budgets
Hidden Assumptions Hunt
+7 audits: rate-limit headers, certificate rotation, locale and formatting, regex input, logging and PII, configuration, git workflows
Concept Connection Atlas
+7 maps: feature flags/deploy, migrations/deploy, ML inference pipeline, config management, payments double-entry ledger, build caches and reproducibility, webhook delivery
Why-Not Handbook
+7 comparisons: GraphQL vs REST, batching vs streaming, ETL vs ELT, monorepo vs multirepo, canary vs blue-green, virtual DOM vs signals, normalization vs denormalization
+7 chains: JWT authentication, HTTP caching, Kubernetes scheduling, distributed tracing, search autocomplete, LSM trees and compaction, consistent hashing
Hidden Assumptions Hunt
+7 audits: environment parity, filesystem paths, integer overflow, process shutdown, idempotency, secrets handling, service discovery
Concept Connection Atlas
+7 maps: serialization/schema evolution, media optimization and delivery, time synchronization infrastructure, secrets management pipeline, batch orchestration lineage, GraphQL federation, browser security model
Why-Not Handbook
+7 comparisons: Kafka vs RabbitMQ, Elasticsearch vs database full-text search, JWT vs sessions, VMs vs containers, Lamport vs vector clocks, optimistic vs pessimistic locking, sagas vs two-phase commit
Failure Casebook
+4 verified cases: Cloudflare leap second 2017 (first-party post fetched in full), Google Cloud networking 2019 (Google incident summary fetched), Okta 2022 (three first-party statements fetched), Microsoft Exchange HAFNIUM 2021 (Microsoft blog + CISA AA21-062A; 403-to-curl recorded)
Misconception Elimination
+7 audits: cross-service transactions, base64 vs encryption, UUID guessability, ORM vs injection, localhost vs loopback, queue ordering, regex vs HTML parsing