llmXive automates scientific discovery end-to-end. A registry of 28 specialist agents — brainstormer, flesh-out, specifier, clarifier, planner, tasker, implementer, reviewer, paper-writer, figure-generator, statistician, proofreader, LaTeX builder, citation validator, and others — drives each project through two complete Spec Kit pipelines: one for the research itself and one for the paper that reports it.
+llmXive automates scientific discovery end-to-end. A registry of specialist agents — brainstormer, flesh-out, specifier, clarifier, planner, tasker, implementer, paper-writer, figure-generator, statistician, proofreader, LaTeX builder, citation validator, and others, plus two panels of focused review lenses — drives each project through two complete Spec Kit pipelines: one for the research itself and one for the paper that reports it.
Research review and paper review run an identify → revise → re-review convergence loop driven by the lane's panel — 8 research reviewers (idea quality, creativity, implementation correctness, completeness, code quality, data quality, filesystem hygiene, plus a generic reviewer) and 12 paper specialists (writing, logic, claims, over-reach, safety, evidence, statistics, code, data, formatting, figures, jargon). The gate is unanimous panel acceptance within a 3-round cap; otherwise the project is kicked back to a prior stage carrying full provenance. Human and simulated-personality reviews are advisory inputs via stage-aware triage; self-review is rejected by the schema. (Spec 015 supersedes the prior point-based gate.)
Beyond the review gates, every factual claim in a generated artifact is detected, registered, and resolved against a real source; an unverifiable claim is marked [UNRESOLVED-CLAIM: …] and hard-blocks advancement (spec 016) — execution receipts are harness-signed so an agent can't forge a pass. When a claim can't be verified as written, an authoritative-fill step searches real sources (OEIS, Wikipedia, Wikidata, papers) and substitutes a value only if it is actually present in a fetched source, never model memory (spec 017). Verification picks a per-claim mode — exact count, approximate constant, safe symbolic computation, or source-fact (spec 018). For prose sources, a value is accepted only when the source semantically asserts that this subject has this value, not a coincidental digit match (spec 019).
Each pipeline step is associated with an appropriate open model. Long, complex tasks (planning, paper writing, deep review) are routed to Qwen3.5 122B; faster, less complex tasks (clarifying questions, atomization, quick judgments) are routed to Gemma 3 27B. All inference runs on Dartmouth’s Discovery Cluster, with a fallback to open-weight Hugging Face models run locally via transformers as needed.
diff --git a/web/js/app.js b/web/js/app.js index d3f655798..e74a5b04b 100644 --- a/web/js/app.js +++ b/web/js/app.js @@ -917,6 +917,23 @@ activate(tab.dataset.tab, tab); } if (document.fonts && document.fonts.ready) document.fonts.ready.then(init); else init(); + + // #262: in-page anchors (the brand `` and the About + // page's "Browse projects" / "Find something to review" buttons, which + // point at `#inProgress`) only rewrote the hash — there was no + // `hashchange` listener, so the active tab never switched. Route every + // hash that names a real tab through the same `activate()` path used on + // load, then scroll it into view. Unknown/empty hashes are a no-op so + // unrelated in-page anchors (e.g. `#how-to-contribute`) keep working. + window.addEventListener("hashchange", () => { + const name = (location.hash || "").slice(1); + if (!name) return; + const tab = tabs.find(t => t.dataset.tab === name); + if (!tab) return; // not a tab anchor — leave it alone + if (!tab.classList.contains("active")) activate(name, tab); + const panel = document.querySelector('.panel[data-panel="' + name + '"]'); + (panel || tab).scrollIntoView({ behavior: "smooth", block: "start" }); + }); // Recompute on font load (widths change once the web font swaps in). if (document.fonts && document.fonts.ready) document.fonts.ready.then(positionUnderline); diff --git a/web/js/auth.js b/web/js/auth.js index 16dd315dd..72c1a7897 100644 --- a/web/js/auth.js +++ b/web/js/auth.js @@ -70,14 +70,13 @@ } const state = _randomState(); sessionStorage.setItem(KEY_STATE, state); - const authorize = _authorizeUrl(); - // FR-011 (account-switching): route through github.com/logout first so the - // user is presented with GitHub's account/consent screen and a *different* - // account can be chosen. This works regardless of whether the OAuth proxy's - // /revoke route is deployed (it's the no-Worker-change fallback path); when - // /revoke IS deployed, signOut() also revokes the grant for good measure. - const logoutUrl = "https://github.com/logout?return_to=" + encodeURIComponent(authorize); - location.href = logoutUrl; + // Navigate STRAIGHT to GitHub's OAuth authorize URL so the user reaches the + // consent screen directly. We must NOT pre-route through github.com/logout: + // GitHub renders a sign-out confirmation page there and does not reliably + // honor return_to to a cross-site authorize URL, so the user would land on a + // sign-out page and never reach consent (issue #217). Account-switching lives + // behind the explicit signOut() action, which revokes the grant via /revoke. + location.href = _authorizeUrl(); } // Non-blocking, dismissible notice (used when grant-revocation fails). diff --git a/web/js/dialog.js b/web/js/dialog.js index deaf12551..da7055313 100644 --- a/web/js/dialog.js +++ b/web/js/dialog.js @@ -68,15 +68,16 @@ + '" + esc + ""; } + // Convert GFM pipe tables to