fix(site): remove weak homepage proof cards

[drewstone]

What changed\n\n- Removed the homepage hero proof-card row with cold-start/model-count/WebBench/open-source count stats.\n- Removed matching proof claims from agent discovery metadata.\n- Replaced benchmark/stat-oriented public copy with evaluation and deployment language.\n\n## Validation\n\n- JSON parse check for .well-known manifests\n- pnpm build\n- pnpm check:links\n- Playwright desktop/mobile homepage screenshots

[tangletools]

✅ Auto-approved drewstone PR — bd8a0b4a

This PR was opened by the trusted drewstone account.
The full PR reviewer audit still runs separately and will publish findings if it detects issues.

_{tangletools · auto-approval · reason: drewstone_author · 2026-07-01T01:52:36Z}

[tangletools]

🟡 Value Audit — sound-with-nits

Verdict	sound-with-nits
Concerns	2 (2 weak-concern)
Heuristic	0.0s
Duplication	0.0s
Interrogation	180.2s (2 bridge agents)
Total	180.2s

💰 Value — sound-with-nits

Removes weak homepage proof cards and benchmark framing from discovery metadata, aligning public copy with the product loop; one stale OG image still shows the removed stats.

• What it does: Removes the four-stat homepage hero proof-card row (0.6s cold start, 671 models, 90% WebBench, 30+ packages), deletes the matching proof_points array from public/.well-known/tangle-agent.json, and swaps "benchmarks" wording for "evaluation/deployment" language across src/pages/index.astro, src/layouts/BaseLayout.astro, and the .well-known manifests.
• Goals it achieves: Replaces vanity benchmark stats with the concrete product narrative (runtime → router → evidence → eval → deployment), so the homepage and agent-readable discovery metadata make a coherent claim about what Tangle builds rather than leading with numbers that may not age well or prove product value.
• Assessment: Good change. The proof cards were weak social proof disconnected from the rest of the page; removing them tightens the hero. The copy edits are consistent across the four changed files and match the existing loop/product sections. JSON files parse correctly, and no new component or duplicated capability was introduced.
• Better / existing approach: none for the homepage/discovery edits themselves. Searched for existing proof-point components or manifest generators and found none; the change correctly edits the source files directly. One completeness gap: public/images/og-self-improving-agents.svg (and the PNG it produces, which BaseLayout.astro:18 uses as the default OG image) still renders the removed stats ('0.6s runtime', '671 models'
• Model: opencode/kimi-for-coding/k2p7
• Bridge attempts: 1

🎯 Usefulness — sound-with-nits

Cleanly removes the weak/unverified proof-card stats and matching metadata claims from the homepage and agent discovery manifests, with no dangling references or dead surface left behind.

• Integration: The removed proofPoints const was only consumed in src/pages/index.astro's hero (now removed), and the .home-proof* CSS rules were removed alongside it, including the responsive overrides at the 720px breakpoint. The proof_points field dropped from public/.well-known/tangle-agent.json was a custom field in a non-standard manifest; a repo-wide grep for proof_points/proofPoints across .jso
• Fit with existing patterns: The change follows the codebase's grain: data, markup, CSS, JSON-LD, and the .well-known manifests are all edited in lockstep, and the 'benchmarks' wording is stripped consistently from the hero copy (index.astro:188), WebPage JSON-LD (index.astro:140), Organization JSON-LD (index.astro:119), BaseLayout default description (BaseLayout.astro:17), and the default keywords list (BaseLayout.astro de
• Real-world viability: Removing a rendered section and its dedicated CSS (including the mobile grid override) leaves the homepage structurally intact; the hero still has its kicker, title, copy, and CTAs. Both edited JSON files remain valid (descriptions shortened, arrays reformatted, one field dropped). No error paths or edge inputs are affected; this is pure static content.
• Model: opencode/zai-coding-plan/glm-5.2
• Bridge attempts: 1

🎯 Usefulness Audit

🟡 Proof stats still live on the homepage's OG image and the overview page [problem-fit] ``

The PR's stated goal is removing weak proof stats from the homepage surface, but the same numbers it deleted still render elsewhere that are reachable from the homepage. (1) The social-share card is og-self-improving-agents.png (referenced at src/pages/index.astro:173 and src/layouts/BaseLayout.astro:18); its source public/images/og-self-improving-agents.svg:30 still bakes in '671 models' as a proof figure, so sharing the homepage still advertises the stat the PR is retiring. (2) src/pages/o

💰 Value Audit

🟡 OG image still shows removed proof-point stats [maintenance] ``

public/images/og-self-improving-agents.svg contains the exact stats this PR removes ('0.6s runtime', '671 models', '90% WebBench'), and the rendered PNG is the default social image in src/layouts/BaseLayout.astro:18. This leaves a public-facing artifact contradicting the new positioning. Fix: edit the SVG to match the new loop/deployment framing and regenerate the PNG.

---

What this audit checks

It judges the change on its merits — not whether it was tasked out in an issue. Unticketed, fast-moving work is fine; the question is whether the change is good and whether a better or existing approach should be used instead.

Pass	What it asks
Heuristic	Vague title? Whitespace-only or cruft-bearing diff? (content signals only)
Duplication	Do added function/class names already exist elsewhere in the repo?
Value Audit	What does it do? What goal does it achieve? Is it good? Better architecture or already-exists?
Usefulness Audit	Does it integrate and fit? Will it hold up in real use and actually get used?

Findings are concerns, not blocks — the human reviewer decides what to do with them.

_{value-audit · 20260701T015745Z}

[tangletools]

✅ No Blockers — bd8a0b4a

Review health 100/100 · Reviewer score 86/100 · Confidence 85/100 · 4 findings (4 low)

	deepseek	kimi-code	aggregate
Readiness	86	92	86
Confidence	85	85	85
Correctness	86	92	86
Security	86	92	86
Testing	86	92	86
Architecture	86	92	86

Reviewer score is advisory once the run is complete and the verdict has no blockers.

Full multi-shot audit completed 5/5 planned shots over 6 changed files. Global verifier still owns final merge decision. | Full multi-shot audit completed 5/5 planned shots over 6 changed files. Global verifier still owns final merge decision.

🟡 LOW 'billing headers'→'auth headers' changes described tcloud purpose — public/.well-known/tangle-agent.json

tcloud package purpose changed from 'Router SDK and CLI for model routing, billing headers, and agent service calls' to '...auth headers...'. The blog post at src/content/blog/openai-compatible-routers-for-agents.mdx:19 still says 'billing headers'. If tcloud now handles general auth (x402 + API keys) rather than only billing, the change is correct but the blog post is now inconsistent.

🟡 LOW proof_points array removed without equivalent replacement in machine-readable manifest — public/.well-known/tangle-agent.json

The entire proof_points section (0.6s sandbox p50, 671 models, 90% WebBench-50, 30+ open source packages) was removed. These concrete metrics are still referenced across the website (overview.astro, og image SVG, blog post, .evolve variants) but no longer present in the machine-readable agent manifest. If downstream agent discovery tools relied on this field they will now see it missing. This is likely intentional — proof points stale quickly and may not belong in a machine schema — but the removal should be coordinated with any consumers of this manifest.

🟡 LOW 'usage attribution' removed from Router does but still in blog post — public/.well-known/tangle-products.json

Router 'does' field dropped 'usage attribution' from the capability list. The blog post at src/content/blog/openai-compatible-routers-for-agents.mdx still references 'usage attribution' in its lede (line 19), alt text (line 16), and takeaway (line 94). This creates an inconsistency between the machine-readable product catalog and the marketing content.

🟡 LOW Router capability description narrows documented feature set — public/llms.txt

Line 22 changed Router purpose from 'policy, fallback, and usage attribution' to 'policy, fallback, and health checks'. Health checks are correct (src/pages/index.astro:52 also uses 'health checks'), but usage attribution/billing separation is still a documented Router capability (src/content/blog/openai-compatible-routers-for-agents.mdx:94). This is not false, just a narrowing; verify the omission is intentional for llms.txt positioning.

---

_{tangletools · 2026-07-01T01:58:39Z · trace}

[tangletools]

✅ Approved — 4 non-blocking findings — bd8a0b4a

Full multi-shot audit completed 5/5 planned shots over 6 changed files. Global verifier still owns final merge decision. | Full multi-shot audit completed 5/5 planned shots over 6 changed files. Global verifier still owns final merge decision.

Full immutable report for this review: trace

Summary comment for this run: full summary

---

_{tangletools · 2026-07-01T01:58:39Z · immutable trace}