Translation-layer outcomes. Nine metrics, before vs current vs target.
Snapshot captured 2026-05-21T15:22:28.523Z. Audit endpoint version v3.6.0. 8 of 8 binary AI-readiness signals passing.
Begin values are zero for every metric because https://www.top10lists.us had no GEO infrastructure before the translation layer was built. Current values are computed from the canonical audit endpoint at geolocus.ai/api/audit for quantitative metrics, from a build-time multi-hit TTFB probe for network response, from JSON-LD parsing on the bot-served HTML for structured-data coverage, and from a placeholder pending the bot citation lookup wiring. Targets are aligned to the published audit thresholds.
| Code | Begin | Current | Target | Definition |
|---|---|---|---|---|
| TTFB | 0 ms | 34 ms | ≤ 200 ms | Time to First Byte. Median of 5 sequential probes from the build host to https://www.top10lists.us. Lower is faster. |
| TTLB | 0 ms | 33 ms | ≤ 500 ms | Time to Last Byte (median, seconds * 1000). Drawn from the canonical audit endpoint at geolocus.ai/api/audit (rtc.components.ttlb_p50_s). Lower is faster. |
| SGR | 0 | 0.5396 | ≥ 0.05 | Source Grounding Rate. Ratio of cited claims to total claims across the home page and one canonical neighborhood page, scored by sonnet-claim-extraction-v1 per the published audit methodology. Higher is better. |
| RR | 0 | 1.000 | ≥ 0.4 | Retrieval Ratio (signal-to-noise). Useful content chars divided by total response chars after presentation stripping. 1.0 = every byte is meaningful content. Higher is better. |
| RTC | 0 tok-s / 1k useful chars | 0.0585 tok-s / 1k useful chars | ≤ 1.000 tok-s / 1k useful chars | Retrieval Token Cost. Token-seconds spent per 1000 useful chars delivered, computed by the audit endpoint as response_tokens * ttlb_p50_s / useful_chars / 4. Lower is cheaper to ingest. |
| LMR | 0 days | 0.64 days | ≤ 14.0 days | Sitemap Lastmod Recency (median, days). Median age of <lastmod> timestamps across the sitemap tree. Lower means more recently maintained. |
| RPS | 0 URLs/sec | 49,600 URLs/sec | ≥ 50,000 URLs/sec | Retrievable Pages per Second. URLs in the sitemap tree divided by parallel-fetch wall-clock seconds at the audit endpoint's concurrency. Higher is better. |
| SCHEMA | 0 types | 17 types | ≥ 5 types | Distinct JSON-LD @type values present in the bot-served HTML across the home page and one canonical neighborhood page. Higher coverage means more structured entities for AI grounding. |
| CITE | 0 bots | 3 bots | ≥ 3 bots | Distinct AI bots with citation evidence (placeholder while bot_crawl_logs distinct-AI-bot-with-citation lookup is wired in a follow-up). Replace value will come from production logs. |
| AC | 0 | — | ≥ 0.8 | AC (Anti-Cloak): asymmetric containment ratio. Fraction of human-side tokens that appear in the AI-route payload. Uses asymmetric containment rather than symmetric overlap because the AI route is intentionally a superset (it adds machine-readable schema, citations, and entity grounding that humans do not need). A score below 0.80 means substantive human content is missing from the AI payload, which AI systems can interpret as deliberate omission. Pass threshold: 0.80 or higher. Currently "No data" — populator is in development. |
All eight binary signals from the canonical /api/audit response. Each is a single boolean: present and AI-permitted, or not.
robots_ai_bots_allowed: truellms_txt_present: truellms_full_txt_present: truesitemap_fresh: truejsonld_structured_data: trueprerendered_html: truemcp_server_live: trueai_content_feed: trueBegin column is fixed at zero per directive: prior to the GEOlocus translation layer the property had no llms.txt, no AI feed, no MCP endpoint, no JSON-LD entities beyond the SPA defaults, no sitemap lastmods, no measurable SGR, no edge bot routing, and no RPS measurement worth quoting. Saying "Begin = previous value of N" would imply infrastructure that did not exist.
The build-time snapshot generator at scripts/fetch-proof-snapshot.mjs calls https://geolocus.ai/api/audit?url=https://www.top10lists.us, takes RR / RTC / RPS / SGR / LMR straight from the response (no recomputation), multiplies rtc.components.ttlb_p50_s by 1000 for TTLB, runs 5 sequential TTFB probes against the home page and takes the median, and counts distinct JSON-LD @type values in the bot-served HTML of the home page plus the canonical neighborhood probe at https://www.top10lists.us/new-york/new-york/gramercy/top10realestateagents. The resulting JSON is written to /.well-known/proof-snapshot.json and consumed by this page.
The CITE metric is currently a placeholder; the production source will be a SELECT COUNT(DISTINCT bot_name) over bot_crawl_logs filtered to AI bots that have an associated citation event. The wiring lives in the follow-up.