Studio: design system, reports, live views, Genesis as an autonomous scientist, identity file (features 007-020) - #2

Merged

LucasWaki merged 41 commits into

mainfrom

007-benchmark-foundations

Sep 10, 2026

Commits on Sep 8, 2026

feat(007): foundation increment: weekly budget ledger, durable evidence, regrade revisions, runtime manifests, Studio with product graphs, offline Enterprise adapter; unblock plan of 8 Sep

LucasWaki
and
claude
committed
e033c90
feat(plans): evaluation track field (agentic-request | create-run) on plan files, hashed when not the default

LucasWaki
and
claude
committed
cf745f1
feat(studio): hosted mode: basic auth, public host allowlist, data folder on a volume, host flag; Dockerfile and .railwayignore for Railway

LucasWaki
and
claude
committed
921f52a
fix(studio): log the cause of a failed GET on the server side

LucasWaki
and
claude
committed
0034329
feat(007): paid dispatch through the weekly ledger: one reservation per API request, attempt cap, round admission, Monarch attempts reserve their ceiling

LucasWaki
and
claude
committed
c83c5f0
test(studio): build name without git; MONARCH_BUILD fallback

LucasWaki
and
claude
committed
3829be8
docs(plan): 8 Sep evening findings: the 67% run identity, the fork-only Railway branch, seeds pin the front-door host, hosted Studio

LucasWaki
and
claude
committed
ed2ec45
docs(research): PG-Waki v1 provenance: the 67% run record and the reconstructed catalog it descends from

LucasWaki
and
claude
committed
6a16bae
feat(007): approval flow per decision D5, capability checks in place of the blanket refusal, doctor through the ledger, wb budget reconcile

LucasWaki
and
claude
committed
b159432
docs(007): the D5 rule in the constitution, CLAUDE.md and the benchmark skill; config README on the attempt cap, approvals, the ledger and wb budget reconcile

LucasWaki
and
claude
committed
5316845
chore: ignore agent worktrees

LucasWaki
and
claude
committed
ff6a229
feat(corpus): wb corpus slate freezes a task set listed by id with its manifest; the achievable-50 id list (unblock plan M2, T2.1 and T2.2)

LucasWaki
and
claude
committed
a21b23f
config: the two gauntlet plans, achievable-50-request (track agentic-request) and achievable-50-workflow (track create-run) (unblock plan M2, T2.4)

LucasWaki
and
claude
committed
38b09e6
research: pre-registrations EXP-2026-001 (gauntlet, one-off request track) and EXP-2026-002 (workflow track), ledger lines, and the two round docs (unblock plan M2, T2.5 and T2.6)

LucasWaki
and
claude
committed
96d8fd3
Merge branch 'worktree-agent-aef8182935801fb3d' into 007-benchmark-foundations
LucasWaki
committed
4841c96
tasks: achievable-50 frozen on corpus revision 1.0.6 (50 prompts, overlap with two tier ids recorded); re-freeze pending the evalrepair.10 import

LucasWaki
and
claude
committed
41f6ae4
config: control-only gauntlet plans (bare Opus 5 beside the answer key) on both tracks, runnable before the Monarch instances exist

LucasWaki
and
claude
committed
2a26250
feat(007): adopt AutomationBench 1.0.6+evalrepair.10 as the bench's world; vendor script with a version check and a provenance record

LucasWaki
and
claude
committed
c69e2f8
feat(007): every task records its world revision; suite ids carry it; wb run refuses a set imported under another world; reports never pool suites

LucasWaki
and
claude
committed
a2cd731
corpus: the seven AutomationBench domains imported under revision evalrepair10 (1.0.6+evalrepair.10), approval rules derived, manifest written

LucasWaki
and
claude
committed
08bf6fe
docs(007): dependency adoption record for 1.0.6+evalrepair.10

LucasWaki
and
claude
committed
58eb3c5
feat(M6): lab seeds: `wb monarch knowledge` writes the stock seeds with descriptions from the PG-Waki catalog through an explicit table

LucasWaki
and
claude
committed
b18eca2
research(pg-waki v1): the catalog-to-seed mapping report of the first lab seed set

LucasWaki
and
claude
committed
c757764
Merge origin/main (Carlos, 8 Sep: seeds v5.3, front-door access log, authoring backstops) into 007-benchmark-foundations

LucasWaki
and
claude
committed
6845376
specs: track features 007 to 011 (they were wrongly ignored with the stray files on 8 Sep)

LucasWaki
and
claude
committed
5943f02
Merge M1 (AutomationBench 1.0.6+evalrepair.10 adopted; corpus-evalrepair10; world revision on every task; reports never pool suites) into 007-benchmark-foundations

LucasWaki
and
claude
committed
afabbb5
Merge M6 (lab seeds: wb monarch knowledge, the PG-Waki catalog-to-seed table, mapping report) into 007-benchmark-foundations

LucasWaki
and
claude
committed
c977677
tasks: achievable-50 re-frozen on corpus revision 1.0.6+evalrepair.10 (same 50 ids)

LucasWaki
and
claude
committed
cdefc9b
docs(plan): state at the end of 8 Sep: paid dispatch, slate, approval flow, evalrepair.10 done; what is ready to trigger and what is not

LucasWaki
and
claude
committed
3fb41e6
test(slate): the frozen-set check reads the corpus folders from the manifest (re-freeze on another revision)

LucasWaki
and
claude
committed
9f4fbb4
docs(plan): the fork tree is on origin (ailabs/unattended-authoring); hosted front door; the three steps left for the Monarch arms

LucasWaki
and
claude
committed
565e640
feat(studio): hosted front door and Langfuse key derivation
LucasWaki
committed
65c7a6a
chore: ignore the generated seeds for the hosted front door

LucasWaki
and
claude
committed
d3d91ff
feat(M5): wb monarch verify; a fresh passing verification of the named instance admits Monarch competitors for two hours

LucasWaki
and
claude
committed
a847ecb
config: knowledge-base hashes of the seeds imported for the hosted front door (47 apps, 8 Sep night)

LucasWaki
and
claude
committed
c49485b
fix(studio): a declared build (MONARCH_BUILD, optional MONARCH_BUILD_COMMIT) freezes its manifest on a hosted bench

LucasWaki
and
claude
committed
7ba9a56
docs(plan): Monarch stock arm verified end to end from the web app; pilot monarch-pilot-001; how Monarch rounds launch

LucasWaki
and
claude
committed
a691ec7

Commits on Sep 9, 2026

Commits on Sep 10, 2026