Studio: design system, reports, live views, Genesis as an autonomous scientist, identity file (features 007-020)

.claude/launch.json

@@ -0,0 +1,33 @@
+{
+  "version": "0.0.1",
+  "configurations": [
+    {
+      "name": "studio",
+      "runtimeExecutable": "uv",
+      "runtimeArgs": [
+        "run",
+        "--directory",
+        "monarch-benchmark/workflowbench",
+        "wb",
+        "studio"
+      ],
+      "port": 8765,
+      "autoPort": true
+    },
+    {
+      "name": "studio-fixture",
+      "runtimeExecutable": "uv",
+      "runtimeArgs": [
+        "run",
+        "--directory",
+        "monarch-benchmark/workflowbench",
+        "python",
+        "tests/browser/server.py",
+        "--port",
+        "8766",
+        "--live"
+      ],
+      "port": 8766
+    }
+  ]
+}

.claude/skills/monarch-benchmark/SKILL.md

@@ -15,11 +15,19 @@ names, task ids and command lines exactly as they are — do not translate them.
 
 ## Rules you may not break
 
-- No round beyond smoke scale (10 tasks, 2 repetitions = 20 attempts per
-  competitor) without Carlos approving that specific round AND `approved_by`
-  filled in the plan file. `wb run` enforces this; do not work around it.
-- Always state attempts and a cost band before running. Never run without the
-  explicit "sim".
+- Lucas approves paid rounds (decision D5, 8 Sep 2026). Above smoke scale (20
+  attempts per competitor, retries included) `wb run` by anyone but an approver
+  writes an approval request and waits; after `wb approve <id>` it runs with
+  `wb run ... --request <id>`. `approved_by` in a plan file approves nothing.
+  Every paid launch needs `WB_OPERATOR=<name>` in the environment. Do not work
+  around any of this.
+- The weekly ledger (US$ 300, `research/budget.sqlite3`) is the spending gate:
+  `wb run` refuses a round the week cannot cover and names the shortfall;
+  `wb budget status` shows what is left. Always state attempts and a cost band
+  before running. Never run without the explicit "sim".
+- Monarch competitors, `wb monarch recipes` and `wb doctor --monarch-probe` are
+  refused until milestone M5 verifies an instance; Claude Code until M7. Say
+  so; do not look for another path.
 - Pre-registration: never edit a task's prompt, starting data or approval rule.
   Never edit prices. If the user asks for an edit after seeing results, say it
   needs Lucas's sign-off and that it makes old rows non-regradable.
@@ -72,8 +80,8 @@ names, ask the user to pick. Never invent a plan.
 
 If the user describes a round no plan file matches, draft a new file under
 `config/plans/` — copy the shape of `config/plans/railway-round-001.yaml`, with
-`approved_by: null`, `audience: internal` and a `cost_ceiling_usd` — and show it
-before continuing. Field tables: `config/README.md`,
+`audience: internal`, a `cost_ceiling_usd` and, when the default US$ 3.00 is not
+right, an `attempt_cap_usd` — and show it before continuing. Field tables: `config/README.md`,
 `specs/001-declarative-benchmark-config/contracts/config-files.md`,
 `specs/002-monarch-create-run/contracts/config-files.md`.
 
@@ -118,8 +126,9 @@ has no rows, say **"sem dados anteriores"** for it — never guess.
 ```
 Rodada: <plan> · produto sob teste: <product> (<kind>, <N> apps) · modo: <mode>
 Prompts: <n_tasks> · tentativas por prompt e competidor: <k> · por competidor: <n> (= prompts × tentativas) · total: <total>
-Teto de custo: US$ <ceiling> · approved_by: <valor ou "não definido">
-Escala: <smoke (<=20 por competidor) | acima de smoke — exige approved_by>
+Teto de custo: US$ <ceiling> · teto por tentativa: US$ <attempt_cap_usd>
+Escala: <smoke (<=20 por competidor) | acima de smoke — exige aprovação do Lucas (wb approve)>
+Operador: <WB_OPERATOR> · semana no ledger: US$ <available> disponíveis de US$ 300 (wb budget status)
 Hash da configuração: <hash>
 
 | competidor | harness | provedor | US$/milhão (in / cached / out) |
@@ -197,7 +206,10 @@ makes resume refuse with a drift error, and the round has to start over.
 | Symptom | Cause and fix |
 |---|---|
 | `config error in ...: key_env: ... not set` | The API key is missing from `.env`. Show the line; do not run. |
-| `config error in ...: approved_by: ...` | Above smoke scale without approval. Ask Carlos; only he fills `approved_by`. |
+| `paid launch refused: WB_OPERATOR is not set` | Set `WB_OPERATOR=<name>` in the environment (who is launching), then retry. |
+| `<id> awaiting approval` | Above smoke scale, launched by a non-approver. Lucas runs `wb approve <id>`; then `wb run ... --request <id>`. |
+| `the week cannot cover this round ... short by US$ X` | The weekly ledger has no room. `wb budget status`; wait for Monday or reduce the plan. Never edit the ledger. |
+| `paid launch refused: ... milestone M5` / `M7` | Monarch or Claude Code is not verified yet. Say so; drop that competitor or wait for the milestone. |
 | Knowledge-base drift error | The seeds Monarch imported no longer match. `uv run wb monarch setup --product <product>` again, then say the config hash moved. |
 | Monarch unreachable / 502 | Railway is locked or asleep: `railway-ops.sh status`, then `unlock`. |
 | Monarch reaches nothing / front door times out | The `FRONT_DOOR_URL` tunnel is down. Restart ngrok; if its host changed, rerun `wb monarch setup`. |

.claude/skills/monarch-benchmark/references/tasks.md

@@ -75,3 +75,27 @@ rather than improvising a draw.
 
 Never draw a set by copying files by hand: a set without a recorded seed cannot
 be redrawn, and the round loses its source line.
+
+## A set listed by id is a slate
+
+When the members are chosen by a rule outside the bench (the first is
+`tasks/achievable-50-ids.txt`, the 50-task gauntlet of the unblock plan of
+8 Sep 2026), the set is frozen by the bench from that list, not by hand:
+
+```bash
+uv run wb corpus slate --ids tasks/<name>-ids.txt --out tasks/<name> --because "<why>"
+```
+
+The list holds one task id per line; its `#` comment lines become the
+manifest's selection rule, so write the rule and its source there. The
+command copies each corpus file unchanged into the folder and writes
+`tasks/<name>-manifest.yaml` beside it: when, the rule, why the set exists,
+the suite revision, the difficulty cut points it reused from
+`tasks/tiers-manifest.yaml`, the count per domain, and one row per task with
+its domain, hash, difficulty score and tier label. It refuses, naming every
+offender, when an id is not in the corpus, has no approval rule, does not
+match its own hash, or already sits in a frozen set (`tier-*`, `random-10`,
+or any folder with a manifest beside it); nothing is written before every id
+passes. A folder that already holds a set needs `--refreeze`, which keeps the
+ids the manifest records and refreshes the copies and hashes after an
+approval-rule change or a corpus re-import. Free and offline.

.env.example

@@ -0,0 +1,10 @@
+# Server-side credentials for AI Labs. Copy to .env; never commit actual values.
+# Supported bounded API control:
+GEMINI_API_KEY=
+# Native runners also require a verified isolated runtime; keys alone do not enable them.
+ANTHROPIC_API_KEY=
+OPENAI_API_KEY=
+
+# Model catalog; optional account includes your custom models.
+FIREWORKS_API_KEY=
+FIREWORKS_ACCOUNT_ID=

.gitignore

@@ -12,3 +12,86 @@ graphify-out/
 
 # Generated Monarch knowledge-base seeds (wb monarch setup); regenerate, do not version
 monarch-benchmark/workflowbench/out/monarch-seeds/
+
+# Local Studio jobs contain private task evidence and are not source.
+monarch-benchmark/workflowbench/out/studio/
+
+# Files from other workspaces that landed in this checkout (Lucas, 2026-09-08): kept out of the lab repo, not deleted
+/WorldRoadBridgeProvenance.cs
+/_drainage_audit.py
+/_drainage_seams.py
+/authenticated-principal-snapshot-report.md
+/bridge_catalog_deep_freeze.ps1
+/bridge_catalog_fix.patch
+/bridge_catalog_fix.ps1
+/bridge_catalog_more_tests.ps1
+/bridge_catalog_tests_fix.ps1
+/bridge_catalog_typed_failure.ps1
+/command-receipt-query-contract-review.md
+/complete_f3a_archive_plumbing.py
+/contract-at-99cb205.md
+/edit_g0_capture_test.py
+/edit_g0_replay.py
+/edit_g0_replay_tests.py
+/explore_floor_slices.py
+/explore_obj.py
+/explore_obj_components.py
+/finish_f3a_witness_regressions.py
+/implement_f3a_archive_witness.py
+/inspect_cycle_runs.py
+/inspect_floor_cycles.py
+/inspect_longitudinal_edges.py
+/measure_tunnel_sections.py
+/mission-crown07-review-report.md
+/mission-crown08-review-report.md
+/mission-persistence-probe-20260908/
+/mission-underflow01-review-report.md
+/mission-underflow02-review-report.md
+/mission-underflow03-review-report.md
+/plot_floor_slices.py
+/provenance_catalog_tests_mapping.ps1
+/provenance_catalog_types.ps1
+/provenance_endpoint.ps1
+/provenance_hard_terminals.ps1
+/provenance_junction.ps1
+/provenance_prepare_adoption.ps1
+/provenance_smoothing.ps1
+/provenance_world_route.ps1
+/provenance_world_route2.ps1
+/sample_curved_x.py
+/seal_bridge_catalog.ps1
+/section_probe.py
+/streets-path-attestation-current-packet.md
+/strengthen_f3a_witness_tests.py
+/underpressure-party-architecture-review.md
+/visual-combat-integration-report.md
+/wire_f3a_witness_tests.py
+/artifacts/exact-lease-approval-20260908/
+
+# Agent worktrees (Claude Code)
+.claude/worktrees/
+monarch-benchmark/workflowbench/out/monarch-seeds-lab/
+monarch-benchmark/workflowbench/out/monarch-seeds-studio/
+
+# Private local preview snapshot
+/artifacts/studio-refactor/studio/
+/artifacts/studio-refactor/research/
+/artifacts/studio-refactor/*.log
+.tmp/research-matrix/
+artifacts/studio-refactor/implement_*.py
+artifacts/studio-refactor/fix_review.py
+artifacts/studio-refactor/tests.xml
+artifacts/studio-refactor/taskset-check/
+artifacts/studio-refactor/*-check.json
+artifacts/studio-refactor/rebuild-*.py
+artifacts/studio-refactor/wire-*.py
+artifacts/studio-refactor/launcher-*.py
+artifacts/studio-refactor/settings-copy.py
+artifacts/studio-refactor/tasksets-sample.py
+
+# Local snapshots and scratch, never committed
+artifacts/enterprise-local/
+.tmp/
+.impeccable/
+monarch-benchmark/workflowbench/.testagent/
+artifacts/genesis-check/

.impeccable/review/confirm-graph.cjs

@@ -0,0 +1,2 @@
+const {chromium}=require('C:/Users/Lucas Wakigawa/.cache/codex-runtimes/codex-primary-runtime/dependencies/node/node_modules/playwright');
+(async()=>{const b=await chromium.launch({headless:true,channel:'msedge'}),p=await b.newPage({viewport:{width:1600,height:1080}});const errors=[];p.on('pageerror',e=>errors.push(e.message));await p.goto('http://127.0.0.1:8765');await p.getByRole('button',{name:'Monarch setups',exact:true}).click();const option=await p.locator('#blueprint-library option').filter({hasText:/Product graph enrichment/}).first().getAttribute('value');await p.locator('#blueprint-library').selectOption(option);await p.locator('[data-graph-node]').filter({hasText:'Enrich the graph'}).click();const original=await p.locator('#node-instructions').inputValue();await p.getByRole('button',{name:'Save draft',exact:true}).click();await p.locator('#blueprint-state').filter({hasText:/Draft saved/}).waitFor();await p.locator('#node-instructions').fill(original+'\nKeep citations with each field.');await p.getByRole('button',{name:'Save draft',exact:true}).click();await p.locator('#blueprint-state').filter({hasText:/Draft saved/}).waitFor();const data=await p.evaluate(async()=>await(await fetch('/api/blueprints')).json());if(!data.items.find(x=>x.id===option).graph.nodes.find(n=>n.type==='enrich').config.instructions.endsWith('Keep citations with each field.'))throw Error('Post-save edits lost');await p.locator('#node-instructions').fill(original);await p.getByRole('button',{name:'Save draft',exact:true}).click();await p.locator('#blueprint-state').filter({hasText:/Draft saved/}).waitFor();await p.locator('[data-graph-node]').filter({hasText:'Enrich the graph'}).focus();await p.keyboard.press('Enter');if(!await p.locator('#node-label').evaluate(el=>el===document.activeElement))throw Error('Keyboard focus lost');await p.screenshot({path:'.impeccable/review/graph-desktop.png',fullPage:true});await p.locator('#node-label').fill('Discard test');p.once('dialog',d=>d.accept());await p.getByRole('button',{name:'New architecture',exact:true}).click();if(await p.evaluate(()=>localStorage.getItem('ailabs-architecture-draft')))throw Error('Discarded edits retained');await p.locator('#blueprint-library').selectOption(option);await p.setViewportSize({width:390,height:844});await p.screenshot({path:'.impeccable/review/graph-mobile.png',fullPage:true});await p.getByRole('button',{name:'Back to runs'}).click();await p.getByRole('button',{name:'New run',exact:false}).click();await p.locator('#task-search').fill('Lisa Park');await p.locator('.difficulty-badge.easy').waitFor();await p.screenshot({path:'.impeccable/review/graph-difficulty.png',fullPage:true});console.log(JSON.stringify({postSaveEditsRetained:true,keyboardFocus:true,discardCleared:true,realDifficultyBadge:true,errors,overflow:await p.evaluate(()=>document.documentElement.scrollWidth>innerWidth)}));await b.close()})().catch(e=>{console.error(e);process.exit(1)});

.impeccable/review/verify-builder-2.cjs

@@ -0,0 +1,8 @@
+const {chromium}=require('C:/Users/Lucas Wakigawa/.cache/codex-runtimes/codex-primary-runtime/dependencies/node/node_modules/playwright');
+(async()=>{const b=await chromium.launch({headless:true,channel:'msedge'});const p=await b.newPage({viewport:{width:1600,height:1000}});const errors=[];p.on('pageerror',e=>errors.push(e.message));p.on('dialog',d=>d.accept());
+await p.goto('http://127.0.0.1:8765');await p.getByRole('tab',{name:'Activity'}).click();await p.waitForTimeout(800);await p.screenshot({path:'.impeccable/review/builder-activity-2.png',fullPage:false});
+await p.getByRole('button',{name:'Monarch setups',exact:true}).click();await p.locator('.bp-node').first().waitFor();await p.evaluate(()=>localStorage.removeItem('ailabs-architecture-draft'));
+const options=await p.locator('#blueprint-library option').allTextContents();const target=options.findIndex(o=>o.startsWith('Opus planner'));await p.locator('#blueprint-library').selectOption({index:target});await p.waitForTimeout(1000);
+await p.locator('[data-node="planner"]').click({position:{x:30,y:20}});await p.waitForTimeout(500);await p.screenshot({path:'.impeccable/review/builder-published.png',fullPage:true});
+await p.getByRole('button',{name:'Run latest version'}).click();await p.waitForTimeout(1500);await p.screenshot({path:'.impeccable/review/builder-launch-2.png',fullPage:false});
+console.log(JSON.stringify({errors,options}));await b.close()})().catch(e=>{console.error(e);process.exit(1)});