feat(front): position the site as the live rules engine for AI agents - #242

Intent

Rework the site so every page says the approved category: The live rules engine for AI agents, with the h1 "Take the rules out of your prompts." and the bold line "Your agents act on what's true now." Source: the gtm-tagline report, sections 2, 3, 4.2, 4.4, 4.6 and 9, with the challenger's factual fixes applied.

What changed

Home (front/app/page.tsx)

• Hero: category eyebrow, h1, bold line plus definition, "the model proposes; the rules decide", and the scope line (single node, tens of concurrent agent sessions per knowledge graph, draft feat(protocol): share evaluations across standing queries that differ only in bound constants, and cache their plans #232 linked, subscriptions and claims ship with the next release).
• New Replaces / Keeps block and a five-sentence Where it sits block (adapter with per-key revisions, claim-then-run outside the tool's call path, cancel on retraction, freshness as a rule, one node with a WAL).
• How it works: the annotated hero code. The loop (subscribe(), claim(), .any()) is labelled upcoming Python SDK. The declarations are labelled runs on today's Python SDK, using today's negation form ~t.tool.in_(KillSwitch.tool).
• The four capability statements from §4.4, "How it recovers", the stack line moved below the fold, "What you delete" (with the time-driven-cron boundary), the comparison table with a commit-gate row, and proof points. The benchmark is labelled "a recursive-query result, not an agent-latency figure". The get-started section is now titled "Pilot it on one flow".

Shared pieces: site title, meta and OG description; the nav adds "What it replaces" and "Compare"; footer links and tagline; all six CTA banners; compare index metadata and subtitle; compare detail descriptions.

Content: the category sentence in the three compare pages, the docs index, the LangGraph and voice-article intros and the LangChain sample. The benchmark post gets a framing note. The vector comparison cell now reads "6.83ms in the recursive benchmark". All benchmark numbers are unchanged (6.83 ms, 11.3 s, 1,652x).

Challenger fixes applied

• workflow-id dedupe is no longer listed as replaced. Temporal is kept, and its workflow id becomes the claim key.
• A claim is not a TTL lock. What it replaces is the "already handled" rows and claimed_by columns; TTL locks follow with leases in a later release.
• Only crons that re-check data go. A check that waits on time keeps a one-line clock writer.
• Stale-source wording names the health lease, the guard at commit and the watchdog.
• kg.why is not shown as working today.
• "the moment" is now "when". "in review" is now "in progress (draft)". "(one per tenant)" is removed from the scope line.

Not in this PR, because it is outside the site: the category string in Cargo.toml, docs/README.md, the server's --help banner and the repository README.

Verification

• cd front && npm ci && npm run build: passes, and all pages are exported.
• Every internal link touched resolves in dist/: quickstart, websocket-api, ingestion, persistence, deployment, python-sdk, langgraph, langchain, verified-completions, the benchmark and voice posts, compare, commercial, docs, and the anchors #replaces, #where, #how and #start. feat(protocol): share evaluations across standing queries that differ only in bound constants, and cache their plans #232 returns 200.
• The declarations block on the page was extracted verbatim and run against an engine built from main's source with today's Python SDK installed from source. Results: on time []; late ETA [['ORD-4821','SHP-9']]; kill switch []; switch lifted [['ORD-4821','SHP-9']].
• At 375 px, scrollWidth == clientWidth (375) on the home page, so nothing scrolls the page sideways. Code scrolls inside its own block. Screenshots were taken at 375 px and 1280 px.