feat(website): blog post -- actors beat the loop on recovery, not everywhere, seven measured changes to the Queen runtime (Closes #1582)

[gHashTag]

Closes #1582

Do not merge yet; this PR is a draft until the coordinator says to merge. The lane 5 numbers (the simulation gate, gHashTag/trios#1726) are now in the post. The production merge, gHashTag/trios#1730, is open, so the post still says that nothing runs in production.

A post on the Queen actor runtime work of 2026-10-09: the competitor study, the seven improvements of gHashTag/trios#1712 with the numbers posted on gHashTag/t27#7851, and what the actor decision card would cost on the bench FPGA. English and Russian bodies. Every number is taken from a merged PR, a benchmark comment or the FPGA provenance in #1582, and each is labelled simulated or measured as its source labels it.

The post (actors-beat-the-loop-on-recovery-not-everywhere, dated 2026-10-10, published: true):

• Result first: the worst recovery falls from 1181 s to 27 s and overload throughput rises 18.7% (simulated, feat(queen): actor runtime MVP, the reviewer as actors behind a flag, and the benchmark (Closes #1697) trios#1698), but the actors do not win everywhere.
• One section per lane, each with its price: adaptive concurrency loses with 3 lanes or a small container; stoppable turns cut overload throughput 78.4 to 60.7 an hour because the old code ran past its bound; waits as rows cut GitHub reads 180 to 24 but notice the end of CI later (83 s against 23 s); the 60-minute turn bound, not the actors, gave most of the dispatcher's gain.
• The node-link chaos table (21/10/5/25/40 to 0), measured on PostgreSQL 16.
• The simulation gate (test(queen): the actor runtime under a seeded deterministic simulation, as a CI gate - simulation.t27 (Closes #1721) trios#1726, spec(queen): simulation.t27 - the actor runtime under a seeded deterministic simulation, as a CI gate (Closes #8279) t27#8292 and #8306): 18 cases of 10,000 steps, each run twice, in 33.5 s on CI with 0 divergences; four defects put back and caught from a seed; the real bug it found first (a turn run for an exited process, fixed with a regression test); its limits.
• The FPGA section: 26,737 LUT and 13 functions that do not synthesize, the jitter's 20,225 LUT, binary integers ahead of GF16/TNF/ternary, about 37,000 actors a board limited by block RAM, and the 12-bit node field that already allows three boards. Synthesis only, labelled so.
• "What this does not show": nothing is in production, which numbers are simulations, the telemetry cost is an estimate, and the FPGA record is assumed.

The widget: a new recording, public/term/t27c-queen-netlink-fence/, in a terminal block in body and ruBody.

• Seven commands against gHashTag/t27 master c6237a7: t27c test-report specs/queen/netlink.t27 passes 9 of 9 with 0 vacuous; a one-line mutation moves the self-fence from 15 s to the 20 s lease; test-report then fails exactly a_node_stops_before_any_peer_can_see_it_down; git checkout restores the same sha256.
• It was made with the exported functions of apps/website/scripts/t27c-mutation-casts.mjs (events, cast text, page, card, gallery row) from a scratch driver that is not committed, so the format matches the published recordings and no recorder code changes. Two differences from that recorder, both visible on the page: the mutation uses perl -pi instead of GNU sed -i (the host is macOS), and the card was rendered with the locally installed Chromium build.
• The t27c 0.5.0 build used predates t27c: test-report exits non-zero on FAIL or BLOCKED (Closes #7370) t27#8152, so test-report still exits 0 on a FAIL; the caption says so.

No og-art triptych. image_edit (GPT Image 2.5 via Kie) answered "Credits insufficient"; FAL and Replicate are locked. Nothing is drawn and nothing is faked. The post stays on the fallback card until art lands in gHashTag/ghashtag.github.io, tracked with the other posts in gHashTag/ghashtag.github.io#45.

After merge: the trinity dispatch to the apex is dead (GHIO_TOKEN), so publish with gh workflow run publish-website.yml --repo gHashTag/ghashtag.github.io.

{
  "version": 1,
  "head_sha": "8455bc372b9fe79e83e64d5c78b1e6adf0f63ce0",
  "summary": "A t27.ai post on the Queen actor runtime: on one seeded input the actors cut the worst recovery from 1181 s to 27 s, but seven measured changes show where they lose, and the decision card as written costs 26,737 LUT on an XC7A200T. It carries a new netlink.t27 recording in both languages.",
  "changes": [
    "apps/website/src/data/blog/bodies/actors-beat-the-loop-on-recovery-not-everywhere.ts: English and Russian bodies with 13 sections, 3 tables, 2 lists and a terminal block each; the simulation gate section carries the lane 5 numbers.",
    "apps/website/src/data/blog/index.ts and posts.ts: the index entry with 29 receipts and 6 open questions, and the import and map line.",
    "apps/website/public/term/t27c-queen-netlink-fence/: session.cast, meta.json, index.html and card.png of a new recording of t27c test-report on specs/queen/netlink.t27 and its fence mutant.",
    "apps/website/public/term/index.html: the new recording listed first, 106 recordings."
  ],
  "tests": [
    {
      "command": "python3 -m unittest discover -s scripts -p 'test_pr_blog*.py'",
      "status": "passed",
      "result": "74 tests OK on this head",
      "evidence": "local run on this head, 2026-10-09 UTC"
    },
    {
      "command": "python3 scripts/pr_blog_report.py validate --event <this PR's event from the GitHub API> --files <its changed files> --posts <its changed bodies> --output <tmp>",
      "status": "passed",
      "result": "exit 0: the work report validates and post_widget.t27 finds a widget in body and ruBody",
      "evidence": "local run against this PR's API data right after it was opened, 2026-10-09 UTC"
    },
    {
      "command": "t27c test-report apps/website/specs/policy/post_widget.t27 (t27c 0.5.0)",
      "status": "passed",
      "result": "5 of 5 tests pass",
      "evidence": "local run, 2026-10-09 UTC"
    },
    {
      "command": "bun build apps/website/src/data/blog/posts.ts, then import the bundle and inspect the new post",
      "status": "passed",
      "result": "the post is first of 100, published, with 1 terminal block in each language, the same block kinds in both, and no ragged table row",
      "evidence": "local run on this head, 2026-10-09 UTC"
    },
    {
      "command": "the recording's own checks (problemsOf from scripts/t27c-mutation-casts.mjs)",
      "status": "passed",
      "result": "all 7 commands exit 0, the base passes 9 of 9, the mutant fails exactly a_node_stops_before_any_peer_can_see_it_down, the diff is one line, the hash after git checkout equals the hash before",
      "evidence": "public/term/t27c-queen-netlink-fence/meta.json and session.cast in this PR"
    },
    {
      "command": "gitleaks protect --staged",
      "status": "passed",
      "result": "no leaks found in the staged change",
      "evidence": "local run before the commit, 2026-10-09 UTC"
    },
    {
      "command": "npm run build and the website-checks gate",
      "status": "not_run",
      "result": "not run locally because the sparse clone has no node_modules; the bodies were parsed by bun build instead",
      "evidence": "the website-checks workflow on this pull request runs them"
    }
  ],
  "limitations": [
    "The production merge, trios#1730, is open; the post says nothing runs in production and must be updated if that changes before it is published.",
    "No og-art triptych: every image provider refused, so the post shows the fallback card until art lands in ghashtag.github.io.",
    "Most runtime numbers are simulations with assumed fault rates; the post labels each one, and nothing in it runs in production.",
    "The FPGA numbers are yosys synthesis with an assumed 333-bit actor record; no place-and-route, no clock rate, nothing on silicon.",
    "The recording was made with a t27c 0.5.0 build that predates t27#8152, so its mutant run exits 0 with one FAIL line.",
    "The post is about 3,200 words in English, longer than the blog-post skill's 1,000 to 1,500, because it covers the study, seven lanes and the FPGA study together."
  ],
  "tags": [
    "Queen",
    "Actors",
    "Measurement",
    "FPGA"
  ],
  "blog": {
    "title": "Actors beat the loop on recovery, not everywhere",
    "summary": "On one seeded input the Queen's actor runtime cuts the worst recovery from 1181 s to 27 s, but seven measured changes show two losses, one deliberate throughput price and a dispatcher gain that mostly came from a time limit.",
    "outline": [
      "The Queen moved to actors under a rule: no swap without an MVP, tests and a benchmark on the same input, and a competitor study named three gaps.",
      "Seven changes landed behind flags, each with a t27 spec and a benchmark against the code it would replace, and each with a stated price.",
      "A seeded simulation gate caught every re-created defect and found a real bug first; nothing runs in production yet, and the FPGA study is synthesis only."
    ],
    "cast": "term/t27c-queen-netlink-fence/session.cast",
    "reproduce": [
      "t27c test-report specs/queen/netlink.t27"
    ]
  }
}

🤖 Generated with Claude Code

[github-actions]

X Brain Health Check

Score: 100.0/100
Status: 🟢 HEALTHY
Threshold: 80/100

X Brain is above merge threshold

[github-actions]

🧪 Brain Stress Test Results

Tests: 261/261 passed, 0 leaked
Status: 🟢 PASS
Gate: every test in zig build test-brain-stress passes and none leaks

[tri-claw]

Summary by t27-bees

This PR adds a blog post about the Queen actor runtime improvements with an accompanying terminal recording widget. The post presents seven measured changes showing where actors beat the loop on recovery and where they don't, plus an FPGA synthesis analysis. The terminal recording demonstrates a fencing behavior change in the netlink spec.

Worth a look

• apps/website/src/data/blog/bodies/actors-beat-the-loop-on-recovery-not-everywhere.ts (diff cut): The blog body diff was not shown. Cannot verify that all cited numbers match their sources (merged PRs, benchmark comments, FPGA provenance) or that "simulated" vs "measured" labels are correctly applied throughout.

• apps/website/src/data/blog/index.ts and apps/website/src/data/blog/posts.ts (diffs cut): The blog index and posts list modifications were not shown. Cannot confirm the post is correctly registered, dated 2026-10-10, marked published: true, and appears in the correct order.

• apps/website/public/term/t27c-queen-netlink-fence/card.png (binary): The social card image is a new binary file with no textual diff. If the image is corrupt, missing, or doesn't match the 1200×630 og:image dimensions declared in index.html, social previews will break.

• apps/website/public/term/t27c-queen-netlink-fence/meta.json:14 — "post": "https://t27.ai/#/blog/actors-beat-the-loop-on-recovery-not-everywhere": Hardcodes the blog slug. If the post's slug or routing changes, the "Read the post" link from the recording page will 404.

• apps/website/public/term/index.html:50 — New recording inserted after the 2026-10-10 entry: The grid shows the 2026-10-10 recording first, then this 2026-10-09 recording. Order is correct (newest first), but verify the CSS grid doesn't reorder unexpectedly on mobile/viewport changes.

---

_{Read: 8 files, +1080 -0, the whole diff up to 60000 characters (the rest was not read). Head 8455bc3. Model: openai-compatible/nvidia/nemotron-3-ultra-550b-a55b. Commands: @t27-bees review | summary | help | pause | resume. Free for every repository.}