WARS: Bee + TRI arm, a real paired campaign, a readable arena; ROADMAP level II; blog posts

.github/workflows/claude-code-review.yml

@@ -34,11 +34,44 @@ jobs:
         with:
           fetch-depth: 1
 
+      # The review runs on z.ai (the owner, 2026-09-27: "замени ANTHROPIC_API_KEY
+      # на z.ai"). Every run before this failed two seconds in: the OAuth token
+      # was refused and ANTHROPIC_API_KEY was empty. z.ai serves the Anthropic
+      # Messages API at /api/anthropic, and Claude Code reaches it through
+      # ANTHROPIC_BASE_URL and ANTHROPIC_AUTH_TOKEN (docs.z.ai, the Claude Code
+      # guide). The key is the ZAI_API_KEY secret; a PR without it (a fork, or
+      # before the secret is set) is told so and not reviewed.
+      - name: Check for the z.ai key
+        id: zai
+        env:
+          ZAI_API_KEY: ${{ secrets.ZAI_API_KEY }}
+        run: |
+          if [ -n "$ZAI_API_KEY" ]; then
+            echo "ready=true" >> "$GITHUB_OUTPUT"
+          else
+            echo "ready=false" >> "$GITHUB_OUTPUT"
+            echo "::notice title=Review skipped::The ZAI_API_KEY secret is not set, so this PR is not reviewed. Add it under Settings > Secrets and variables > Actions."
+          fi
+
       - name: Run Claude Code Review
         id: claude-review
+        if: steps.zai.outputs.ready == 'true'
         uses: anthropics/claude-code-action@v1
+        env:
+          ANTHROPIC_BASE_URL: https://api.z.ai/api/anthropic
+          ANTHROPIC_AUTH_TOKEN: ${{ secrets.ZAI_API_KEY }}
+          # z.ai's guide raises Claude Code's request timeout for GLM.
+          API_TIMEOUT_MS: "3000000"
+          # Background and subagent calls ask for a Haiku, Sonnet or Opus by
+          # name; these send them to GLM instead. ZAI_MODEL (a repository
+          # variable) picks the model; glm-4.6 is the default because it is the
+          # z.ai id the Queen's own model catalogue lists.
+          ANTHROPIC_DEFAULT_OPUS_MODEL: ${{ vars.ZAI_MODEL || 'glm-4.6' }}
+          ANTHROPIC_DEFAULT_SONNET_MODEL: ${{ vars.ZAI_MODEL || 'glm-4.6' }}
+          ANTHROPIC_DEFAULT_HAIKU_MODEL: glm-4.5-air
         with:
-          claude_code_oauth_token: ${{ secrets.CLAUDE_CODE_OAUTH_TOKEN }}
+          # The action requires a key input; it is the same z.ai key.
+          anthropic_api_key: ${{ secrets.ZAI_API_KEY }}
           # Note: the log suggests `show_full_output: true`, but that is not an
           # input this action version accepts -- its inputs are trigger_phrase,
           # prompt, settings, claude_args and the auth/provider set. Extra
@@ -60,5 +93,5 @@ jobs:
 
           # See https://github.com/anthropics/claude-code-action/blob/main/docs/usage.md
           # or https://code.claude.com/docs/en/cli-reference for available options
-          claude_args: '--verbose --allowed-tools "Bash(gh issue view:*),Bash(gh search:*),Bash(gh issue list:*),Bash(gh pr comment:*),Bash(gh pr diff:*),Bash(gh pr view:*),Bash(gh pr list:*)"'
+          claude_args: '--model ${{ vars.ZAI_MODEL || ''glm-4.6'' }} --verbose --allowed-tools "Bash(gh issue view:*),Bash(gh search:*),Bash(gh issue list:*),Bash(gh pr comment:*),Bash(gh pr diff:*),Bash(gh pr view:*),Bash(gh pr list:*)"'
 

.github/workflows/website-checks.yml

@@ -271,6 +271,12 @@ jobs:
       - name: Queen WARS experiment ledger
         run: npm run check:wars && npm run test:wars-spec
 
+      # LEVEL II on the ROADMAP tab: the raid, honey, bosses, cracks and the
+      # round's pulse. The raid list is pinned here and in gHashTag/t27's
+      # feed_roadmap.py, so the page and the feeder raid the same sector.
+      - name: Roadmap game rules
+        run: npm run check:roadmap-game
+
       - name: Build
         run: npx vite build
 

apps/website/package.json

@@ -97,6 +97,7 @@
     "check:onboarding": "node scripts/onboarding-from-spec.mjs --check",
     "check:wars": "node scripts/queen-wars-from-spec.mjs --check && node --experimental-strip-types qa/queen-wars-contract.mjs",
     "test:wars-spec": "node --test scripts/queen-wars-from-spec.test.mjs",
+    "check:roadmap-game": "node --experimental-strip-types qa/roadmap-game-contract.mjs",
     "test:viewport-spec": "node --test scripts/viewport-from-spec.test.mjs",
     "check:explorer-viewport": "node qa/explorer-viewport-contract.mjs",
     "check:passport": "node --experimental-strip-types qa/passport-figures.mjs"

apps/website/public/queen/runs/campaign-20260926/base-4613.txt

@@ -0,0 +1,41 @@
+# Queen WARS acceptance transcript: gHashTag/t27#4613
+# worktree: /home/user/ghashtag/t27
+# head: afe2186cfa9572c8c2d519506aa85a8c9447eae5
+# judge: t27c 0.4.0; zig 0.16.0
+# at: 2026-09-26T18:59:37Z
+# diff --numstat: (no change)
+
+## criterion 1: FAIL (expected 1)
+$ t27c coverage specs/base/ternary_encoding.t27 2>&1 | grep -cE '^Untested: +0$'
+0
+
+## criterion 2: PASS (expected 13)
+$ grep -cE '^[[:space:]]*(pub[[:space:]]+)?fn[[:space:]]' specs/base/ternary_encoding.t27
+13
+
+## criterion 3: FAIL (expected >= 11)
+$ grep -cE '^[[:space:]]*test[[:space:]]+("|[A-Za-z_])' specs/base/ternary_encoding.t27
+10
+
+## criterion 4: PASS (expected IMPLEMENTED)
+$ t27c spec-status specs/base/ternary_encoding.t27
+IMPLEMENTED
+
+## review FR-002 parse
+$ t27c parse specs/base/ternary_encoding.t27 >/dev/null 2>/tmp/t27c_parse_err; echo exit=$?; head -3 /tmp/t27c_parse_err
+exit=0
+
+## review vacuity
+$ t27c validate-vacuity --specs-dir specs/base --top 500 2>&1 | grep -F ternary_encoding.t27 || echo "ternary_encoding.t27: not listed as vacuous"
+ternary_encoding.t27: not listed as vacuous
+
+## review test-report
+$ t27c test-report specs/base/ternary_encoding.t27 2>&1 | tail -15
+--- test report: specs/base/ternary_encoding.t27 ---
+  BLOCKED  does not compile: /tmp/t27c-test-report-ternary_encoding/spec.zig:8:9: error: assertion failed
+           /tmp/t27c-test-report-ternary_encoding/spec.zig:280:47: note: called at comptime here
+
+  A blocked spec is not a failing one. It never produced a
+  binary, so it has no per-test result to report.
+
+# ACCEPTANCE: FAILED

apps/website/public/queen/runs/campaign-20260926/base-4614.txt

@@ -0,0 +1,46 @@
+# Queen WARS acceptance transcript: gHashTag/t27#4614
+# worktree: /home/user/ghashtag/t27
+# head: afe2186cfa9572c8c2d519506aa85a8c9447eae5
+# judge: t27c 0.4.0; zig 0.16.0
+# at: 2026-09-26T18:51:51Z
+# diff --numstat: (no change)
+
+## criterion 1: FAIL (expected 1)
+$ t27c coverage specs/boards/arty_a7.t27 2>&1 | grep -cE '^Untested: +0$'
+0
+
+## criterion 2: PASS (expected 5)
+$ grep -cE '^[[:space:]]*(pub[[:space:]]+)?fn[[:space:]]' specs/boards/arty_a7.t27
+5
+
+## criterion 3: FAIL (expected >= 18)
+$ grep -cE '^[[:space:]]*test[[:space:]]+("|[A-Za-z_])' specs/boards/arty_a7.t27
+17
+
+## criterion 4: PASS (expected IMPLEMENTED)
+$ t27c spec-status specs/boards/arty_a7.t27
+IMPLEMENTED
+
+## criterion 5: PASS (expected 0)
+$ t27c test-report specs/boards/arty_a7.t27 2>&1 | grep -c BLOCKED
+0
+
+## review FR-002 parse
+$ t27c parse specs/boards/arty_a7.t27 >/dev/null 2>/tmp/t27c_parse_err; echo exit=$?; head -3 /tmp/t27c_parse_err
+exit=0
+
+## review vacuity
+$ t27c validate-vacuity --specs-dir specs/boards --top 500 2>&1 | grep -F arty_a7.t27 || echo "arty_a7.t27: not listed as vacuous"
+specs/boards/arty_a7.t27                                        0        0    0.0%      0      0
+
+## review test-report
+$ t27c test-report specs/boards/arty_a7.t27 2>&1 | tail -15
+--- test report: specs/boards/arty_a7.t27 ---
+
+  tests       17
+  pass        17
+  FAIL        0
+  invariants  11   proved -- comptime, so compiling IS the check
+  rate    100.0%
+
+# ACCEPTANCE: FAILED

apps/website/public/queen/runs/campaign-20260926/base-4695.txt

@@ -0,0 +1,46 @@
+# Queen WARS acceptance transcript: gHashTag/t27#4695
+# worktree: /home/user/ghashtag/t27
+# head: afe2186cfa9572c8c2d519506aa85a8c9447eae5
+# judge: t27c 0.4.0; zig 0.16.0
+# at: 2026-09-26T18:59:36Z
+# diff --numstat: (no change)
+
+## criterion 1: FAIL (expected 1)
+$ t27c coverage specs/fpga/testbench/simulator_tb.t27 2>&1 | grep -cE '^Untested: +0$'
+0
+
+## criterion 2: PASS (expected 4)
+$ grep -cE '^[[:space:]]*(pub[[:space:]]+)?fn[[:space:]]' specs/fpga/testbench/simulator_tb.t27
+4
+
+## criterion 3: FAIL (expected >= 7)
+$ grep -cE '^[[:space:]]*test[[:space:]]+("|[A-Za-z_])' specs/fpga/testbench/simulator_tb.t27
+6
+
+## criterion 4: PASS (expected IMPLEMENTED)
+$ t27c spec-status specs/fpga/testbench/simulator_tb.t27
+IMPLEMENTED
+
+## criterion 5: PASS (expected 0)
+$ t27c test-report specs/fpga/testbench/simulator_tb.t27 2>&1 | grep -c BLOCKED
+0
+
+## review FR-002 parse
+$ t27c parse specs/fpga/testbench/simulator_tb.t27 >/dev/null 2>/tmp/t27c_parse_err; echo exit=$?; head -3 /tmp/t27c_parse_err
+exit=0
+
+## review vacuity
+$ t27c validate-vacuity --specs-dir specs/fpga/testbench --top 500 2>&1 | grep -F simulator_tb.t27 || echo "simulator_tb.t27: not listed as vacuous"
+specs/fpga/testbench/simulator_tb.t27                           7        0    0.0%      1      0
+
+## review test-report
+$ t27c test-report specs/fpga/testbench/simulator_tb.t27 2>&1 | tail -15
+--- test report: specs/fpga/testbench/simulator_tb.t27 ---
+
+  tests       6
+  pass        6
+  FAIL        0
+  invariants  1   proved -- comptime, so compiling IS the check
+  rate    100.0%
+
+# ACCEPTANCE: FAILED

apps/website/public/queen/runs/campaign-20260926/judge/accept.py

@@ -0,0 +1,72 @@
+#!/usr/bin/env python3
+"""Queen WARS acceptance judge: run an issue's own acceptance commands in a worktree.
+
+Usage: accept.py <issue> <worktree> <transcript-out>
+
+Runs every acceptance criterion exactly as the issue states it, from the
+worktree root, with the judge t27c and zig 0.16.0 on PATH, and writes a
+transcript: command, raw output, expected value, pass/fail. Then two review
+checks the issue's requirements name (FR-002 parse, vacuity) and the spec's
+own test-report summary. Exit 0 iff every acceptance criterion passes.
+"""
+import datetime
+import os
+import subprocess
+import sys
+
+JUDGE_PATH = "/home/user/wars/judge:/home/user/tools/bin"
+
+SPECS = {
+    "4614": ("specs/boards/arty_a7.t27", 5, 18, True),
+    "4695": ("specs/fpga/testbench/simulator_tb.t27", 4, 7, True),
+    "4613": ("specs/base/ternary_encoding.t27", 13, 11, False),
+}
+
+
+def sh(cmd, cwd):
+    env = dict(os.environ, PATH=JUDGE_PATH + ":" + os.environ["PATH"])
+    p = subprocess.run(["bash", "-c", cmd], cwd=cwd, env=env,
+                       capture_output=True, text=True, timeout=900)
+    return (p.stdout + p.stderr).strip()
+
+
+def main():
+    issue, wt, out = sys.argv[1], sys.argv[2], sys.argv[3]
+    spec, n_fn, min_tests, has_c5 = SPECS[issue]
+    crit = [
+        ("1", f"t27c coverage {spec} 2>&1 | grep -cE '^Untested: +0$'", lambda o: o == "1", "1"),
+        ("2", f"grep -cE '^[[:space:]]*(pub[[:space:]]+)?fn[[:space:]]' {spec}",
+         lambda o: o == str(n_fn), str(n_fn)),
+        ("3", f"grep -cE '^[[:space:]]*test[[:space:]]+(\"|[A-Za-z_])' {spec}",
+         lambda o: o.isdigit() and int(o) >= min_tests, f">= {min_tests}"),
+        ("4", f"t27c spec-status {spec}", lambda o: "IMPLEMENTED" in o.split(), "IMPLEMENTED"),
+    ]
+    if has_c5:
+        crit.append(("5", f"t27c test-report {spec} 2>&1 | grep -c BLOCKED",
+                     lambda o: o == "0", "0"))
+    lines = [f"# Queen WARS acceptance transcript: gHashTag/t27#{issue}",
+             f"# worktree: {wt}",
+             f"# head: {sh('git rev-parse HEAD', wt)}",
+             f"# judge: {sh('t27c --version', wt)}; zig {sh('zig version', wt)}",
+             f"# at: {datetime.datetime.now(datetime.timezone.utc).strftime('%Y-%m-%dT%H:%M:%SZ')}",
+             f"# diff --numstat: {sh('git diff --numstat', wt) or '(no change)'}",
+             ""]
+    ok_all = True
+    for n, cmd, ok, want in crit:
+        o = sh(cmd, wt)
+        passed = ok(o)
+        ok_all &= passed
+        lines += [f"## criterion {n}: {'PASS' if passed else 'FAIL'} (expected {want})",
+                  f"$ {cmd}", o, ""]
+    for label, cmd in (("review FR-002 parse", f"t27c parse {spec} >/dev/null 2>/tmp/t27c_parse_err; echo exit=$?; head -3 /tmp/t27c_parse_err"),
+                       ("review vacuity", f"t27c validate-vacuity --specs-dir {os.path.dirname(spec)} --top 500 2>&1 | grep -F {os.path.basename(spec)} || echo \"{os.path.basename(spec)}: not listed as vacuous\""),
+                       ("review test-report", f"t27c test-report {spec} 2>&1 | tail -15")):
+        lines += [f"## {label}", f"$ {cmd}", sh(cmd, wt), ""]
+    lines.append(f"# ACCEPTANCE: {'PASSED' if ok_all else 'FAILED'}")
+    open(out, "w").write("\n".join(lines) + "\n")
+    print(f"#{issue}: {'PASSED' if ok_all else 'FAILED'} -> {out}")
+    return 0 if ok_all else 1
+
+
+if __name__ == "__main__":
+    sys.exit(main())

apps/website/public/queen/runs/campaign-20260926/judge/mutate.py

@@ -0,0 +1,114 @@
+#!/usr/bin/env python3
+"""Queen WARS review: does an arm's NEW test catch a fixed set of mutants?
+
+Usage: mutate.py <issue> <worktree> <new-test-name> <transcript-out>
+
+For each mutant of the function under test (fixed per issue, the same for
+every arm), rewrite the arm's patched spec, run `t27c test-report --verbose`,
+and record whether the arm's new test FAILs (killed), passes (survived) or
+the spec does not build (blocked). The arm's file is restored byte-for-byte
+after every mutant, and the restore is checked.
+"""
+import os
+import subprocess
+import sys
+
+PATH = "/home/user/wars/judge:/home/user/tools/bin:" + os.environ["PATH"]
+
+MUTANTS = {
+    "4614": ("specs/boards/arty_a7.t27", [
+        ("return NUM_BUTTONS + 1", "fn count_buttons() -> usize {\n        return NUM_BUTTONS;",
+         "fn count_buttons() -> usize {\n        return NUM_BUTTONS + 1;"),
+        ("return 0", "fn count_buttons() -> usize {\n        return NUM_BUTTONS;",
+         "fn count_buttons() -> usize {\n        return 0;"),
+        ("return NUM_LEDS - 1", "fn count_buttons() -> usize {\n        return NUM_BUTTONS;",
+         "fn count_buttons() -> usize {\n        return NUM_LEDS - 1;"),
+    ]),
+    "4695": ("specs/fpga/testbench/simulator_tb.t27", [
+        ("cycle += 2", "        sim_cycle = sim_cycle + 1;\n    }\n\n    fn reset",
+         "        sim_cycle = sim_cycle + 2;\n    }\n\n    fn reset"),
+        ("clock ends low", "        clk = false;\n        clk = true;\n        sim_cycle = sim_cycle + 1;",
+         "        clk = true;\n        clk = false;\n        sim_cycle = sim_cycle + 1;"),
+        ("no cycle advance", "        sim_cycle = sim_cycle + 1;\n    }\n\n    fn reset",
+         "        sim_cycle = sim_cycle;\n    }\n\n    fn reset"),
+        ("tick also counts an event", "        sim_cycle = sim_cycle + 1;\n    }\n\n    fn reset",
+         "        sim_cycle = sim_cycle + 1;\n        events_processed = events_processed + 1;\n    }\n\n    fn reset"),
+    ]),
+    "4613": ("specs/base/ternary_encoding.t27", [
+        ("never rejects", "            if (!is_valid_trit(trits[i])) {\n                return false;",
+         "            if (false) {\n                return false;"),
+        ("checks only the first slot", "fn validate_trits(trits: []i32, len: usize) → bool {\n        var i : usize = 0;\n        while (i < len) {",
+         "fn validate_trits(trits: []i32, len: usize) → bool {\n        var i : usize = 0;\n        while (i < 1) {"),
+        ("skips the last slot", "fn validate_trits(trits: []i32, len: usize) → bool {\n        var i : usize = 0;\n        while (i < len) {",
+         "fn validate_trits(trits: []i32, len: usize) → bool {\n        var i : usize = 0;\n        while (i + 1 < len) {"),
+        ("rejects everything", "            i = i + 1;\n        }\n        return true;\n    }\n\n    // ═",
+         "            i = i + 1;\n        }\n        return false;\n    }\n\n    // ═"),
+    ]),
+}
+
+# Pre-existing blockers in ternary_encoding.t27 at afe2186c, outside #4613's
+# boundary: two comptime invariants that fail (balanced encoders, unipolar
+# decoders) and two tests that discard a return value. Removed only in the
+# review copy, so an arm's new test can be run at all.
+DEBLOCK = {"4613": ["invariant byte_trits_roundtrip", "invariant bits_trits_roundtrip",
+                    "test byte_to_trits_roundtrip", "test char_encoding_roundtrip"]}
+
+
+def deblock(text, names):
+    out, skip = [], False
+    for line in text.split("\n"):
+        head = line[4:] if line.startswith("    ") and not line.startswith("     ") else None
+        if head is not None or line.startswith("}"):
+            skip = head is not None and any(head == n or head.startswith(n + " ")
+                                            for n in names)
+        if not skip:
+            out.append(line)
+    return "\n".join(out)
+
+
+def main():
+    issue, wt, test_name, out = sys.argv[1:5]
+    source = sys.argv[5] if len(sys.argv) > 5 else None   # arm file to review in wt
+    spec, mutants = MUTANTS[issue]
+    path = os.path.join(wt, spec)
+    if source:
+        text = deblock(open(source).read(), DEBLOCK[issue])
+        open(path, "w").write(text)
+    original = open(path, "rb").read()
+    lines, killed = [f"# mutation review: gHashTag/t27#{issue}, new test `{test_name}`"
+                     + (f" (de-blocked copy of {source})" if source else ""), ""], 0
+    if source:
+        r = subprocess.run(["bash", "-c", f"t27c test-report --verbose {spec} 2>&1"],
+                           cwd=wt, env=dict(os.environ, PATH=PATH),
+                           capture_output=True, text=True, timeout=900)
+        lines += ["## unmutated de-blocked copy", r.stdout.strip(), ""]
+    for label, old, new in mutants:
+        text = original.decode()
+        if text.count(old) != 1:
+            lines.append(f"## {label}: SKIPPED (anchor not unique: {text.count(old)})")
+            continue
+        try:
+            open(path, "w").write(text.replace(old, new))
+            r = subprocess.run(["bash", "-c", f"t27c test-report --verbose {spec} 2>&1"],
+                               cwd=wt, env=dict(os.environ, PATH=PATH),
+                               capture_output=True, text=True, timeout=900)
+            rep = r.stdout
+        finally:
+            open(path, "wb").write(original)
+        assert open(path, "rb").read() == original, "restore failed"
+        if "BLOCKED" in rep:
+            verdict = "blocked"
+        elif f"FAIL  {test_name}" in rep:
+            verdict, killed = "killed", killed + 1
+        elif f"pass  {test_name}" in rep:
+            verdict = "survived"
+        else:
+            verdict = "not-run"
+        lines += [f"## {label}: {verdict}", rep.strip(), ""]
+    lines.append(f"# MUTANTS KILLED BY THE NEW TEST: {killed}/{len(mutants)}")
+    open(out, "w").write("\n".join(lines) + "\n")
+    print(f"#{issue} {os.path.basename(wt)}: {killed}/{len(mutants)} killed -> {out}")
+
+
+if __name__ == "__main__":
+    main()