SOVP Benchmark

Archived document. Completed 29 October 2026, version 1.1. Author: Thorsten Litzki, Litzki Systems LLC. Two measurement runs: 17/18 June 2026 (method v2.0) and 26 August to 4 September 2026 (method v3.0, cut-off 5 September 2026). No further runs are planned; these figures will not be updated. The benchmark is no longer sold; the former sector reports are discontinued.

Maturity levels by SOVP score (SRS): L0 below 30, L1 30 to 44, L2 45 to 59, L3 60 to 74, L4 from 75. The levels rank the score; they are not a protocol verdict. SRS 75 is not a SOVP threshold: CERTIFIED is decided by conformance (published SOVP artefacts plus an operational floor), not by the score. No domain in the September snapshot carries the verdict CERTIFIED.

01 // Measurement run, cut-off 5 Sep 2026 · method v3.0

Per domain the latest complete scan up to the cut-off. Most of it was measured in late August: 999 domains on 27 Aug, 433 on 26 Aug, 92 between 2 and 4 Sep, and one single domain as early as 18 Jun 2026 under method v2.0 (it is the only one with a decimal place). “September” names the cut-off, not the measurement period. 999 of the 1,525 domains come from the international Tranco 1000 list, the rest from ten German sector lists. The comparison with June is therefore mainly a change of population, not a statement that the market got worse.

• KPI-01 // V3.0
  1,525Domains
• KPI-02 // V3.0
  43.57Avg. SRS
• KPI-03 // V3.0
  44Median SRS
• KPI-04 // V3.0
  11.03Avg. AI readiness

02 // Maturity levels

Each dot is one scored measurement, sorted into its maturity level. For September that is one domain per dot (1,492 domains with a score); for June it is one measurement per dot (559 measurements on 540 domains, 19 of them measured twice). Switching shows the June run on the same scale.

September 2026

• L0198 · 13.3 %
• L1566 · 37.9 %
• L2621 · 41.6 %
• L3105 · 7.0 %
• L42 · 0.1 %

June 2026

• L011 · 2.0 %
• L1231 · 41.3 %
• L2309 · 55.3 %
• L38 · 1.4 %
• L40 · 0.0 %

Shares for September relative to the 1,492 domains with a score, for June relative to the 559 measurements. The 33 domains without a score (INCOMPLETE) are not counted in any level. Both domains in L4 come from the Tranco 1000 list; neither is CERTIFIED.

03 // Sectors (10 domains or more)

For each sector the average in the cut-off snapshot, the number of domains with a score and the avg. AI readiness. The line at 75 ranks the score; it is not a protocol threshold.

June and September not directly comparable

The June sector values are in the next section: between the runs, domains moved to other lists (for example seven automotive groups that count under DAX 40 in September), so the gap is not a trend.

Why the sectors add up to 1,522 domains

Three financial institutions whose latest scan ran under a sample list count in the totals but in no sector, so the sectors add up to 1,522 rather than 1,525 domains.

• Ø SRS
• SRS 75 line (ranking, not a protocol threshold)

• B2B e-commerce

  23 Domains

  49.52Ø SRS

  0SRS 75100

  Avg. AI readiness

  14.70

• DACH publishers & media

  39 Domains

  48.36Ø SRS

  0SRS 75100

  Avg. AI readiness

  11.77

• Financial sector (BaFin)

  44 Domains

  45.09Ø SRS

  0SRS 75100

  Avg. AI readiness

  10.64

• Critical infrastructure (KRITIS)

  48 Domains

  45.01Ø SRS

  0SRS 75100

  Avg. AI readiness

  9.62

• E-commerce Germany

  89 Domains

  44.30Ø SRS

  0SRS 75100

  Avg. AI readiness

  12.29

• Digital economy (BVDW)

  100 Domains

  44.00Ø SRS

  0SRS 75100

  Avg. AI readiness

  12.52

• Tranco 1000 (global)

  999 Domains

  43.31Ø SRS

  0SRS 75100

  Avg. AI readiness

  11.28

• DAX 40

  40 Domains

  43.25Ø SRS

  0SRS 75100

  Avg. AI readiness

  10.65

• Health & pharma

  47 Domains

  42.26Ø SRS

  0SRS 75100

  Avg. AI readiness

  6.98

• Federal authorities

  50 Domains, 17 with a score

  41.12Ø SRS (n = 17)

  0SRS 75100

  Avg. AI readiness

  7.16 (n = 50)

• Automotive & industry

  43 Domains

  39.35Ø SRS

  0SRS 75100

  Avg. AI readiness

  8.23

Excluded are an internal test entry on our own domain (score 100) and the sample list of six financial institutions; neither was a market survey. Three domains of the sample list also appear in a market list and are counted. The basis is the frozen snapshot of 29 Oct 2026 (cut-off 5 Sep 2026 23:59 UTC, latest full scan per domain). Avg. SRS, median and percentiles over the domains with a score (percentiles by the nearest-rank method); avg. AI readiness over all domains, because the partial value exists even where no overall score could be formed. For the federal authorities the scan did not get far enough on 33 of 50 domains (verdict INCOMPLETE); the sector value rests on the remaining 17 and is not comparable with the other sectors. Each list is a sample I compiled myself, not a complete survey of a sector: the values describe the listed addresses in this run, not the industry.

04 // June 2026 · method v2.0 · historical

First measurement run across ten German sector lists. 559 measurements on 540 domains (19 domains were measured twice; the original publication said 559 domains).

• KPI-05 // V2.0
  559Measurements
• KPI-06 // V2.0
  45.31Avg. SRS
• KPI-07 // V2.0
  0SRS ≥ 75
• KPI-08 // V2.0
  11.45Avg. AI readiness

Sectors in June

05 // Signal groups (means with sample size n)

In September the content groups cover only about 60 % of the domains, the infrastructure groups about 100 %. That is why every value carries its n. All September values come from the frozen snapshot of 29 Oct 2026; the internal test entry and the sample list are excluded.

• June
• September

• Meta & declarationJune 80.85 · n 406Sep 77.26 · n 915
• Code & renderingJune 88.18 · n 402Sep 85.06 · n 908
• Headings & structureJune 68.39 · n 402Sep 65.05 · n 912
• Narrative qualityJune 65.10 · n 404Sep 57.33 · n 910
• Integration & APIsJune 65.91 · n 531Sep 78.31 · n 1,522
• DNS & domainJune 64.47 · n 531Sep 62.49 · n 1,523
• Security & TLSJune 62.62 · n 507Sep 56.38 · n 1,446
• ALPN & protocolsJune 61.39 · n 531Sep 62.90 · n 1,518
• Images & mediaJune 55.35 · n 404Sep 57.65 · n 912
• Links & navigationJune 47.77 · n 404Sep 53.94 · n 921
• AI signalsJune 24.27 · n 530Sep 25.00 · n 1,521
• Semantic HTMLJune 18.62 · n 400Sep 17.94 · n 911
• llms.txt qualityJune 7.68 · n 531Sep 9.45 · n 1,521
• Agentic readinessJune 0.34 · n 531Sep 0.49 · n 1,521

06 // Method change v2.0 → v3.0

07 // Deviations and errors, named openly

Checking every number against the database (25 Oct 2026) revealed these points. They are listed in full, including where they concern my own earlier publications.

Figures of the measurement runs

1. F1

   June: avg. SRS originally published as 45.1, the database gives 45.31. The original calculation cannot be reconstructed; both values are given.

2. F2

   June: “559 domains” were 559 measurements on 540 domains.

3. F3

   September: what was published was 1,525 domains, avg. SRS 43.48 and avg. AI readiness 10.97; corrected to 1,522 on 26 Oct 2026. The frozen snapshot of 29 Oct 2026 confirms 1,525: three financial institutions of the sample list also appear in a market list and are counted; the correction had removed them twice. Against the original publication, avg. SRS (43.48 → 43.57), avg. AI readiness (10.97 → 11.03) and NOT_CONFORMANT (16 → 19) change; median, percentiles and the two domains at 75 or above are unchanged. The intermediate recalculation of 26 Oct 2026 (1,522 / 43.57 / 11.13) was never published on any page.

4. F4

   September: P25 originally 36; by the nearest-rank method the database also gives 36 (35 with interpolation). The method is now stated in the report.

5. F5

   The September snapshot was defined without a fixed cut-off and today is only reproducible with the cut-off of 5 Sep 2026; that cut-off therefore applies.

6. F6

   33 domains have the verdict INCOMPLETE and no score. They are listed as a separate group and not counted in any maturity level.

7. F16

   The verdict distribution of the September run follows the scan date, not the market: the change to the three-state verdict (CERTIFIED / NOT_CONFORMANT / INCOMPLETE) went live on 3 Sep 2026. The 1,473 domains carrying the old value NOT_READY were scanned before that day, the 19 NOT_CONFORMANT and 33 INCOMPLETE after it. How the 1,473 would be classified under the new logic has not been measured.

8. F17

   Several hundred partial values of the Tranco run of 27 Aug 2026 are not measurements: where a scoring group could not be collected at all, the scanner substituted a neutral value (50, or 25 for AI readiness). In the Tranco run this affected 470 domains for content quality and 624 for authority. Since 3 Sep 2026 such cases produce the verdict INCOMPLETE; rows written before that keep the neutral values. Avg. content quality and avg. authority are biased upwards as a result.

Earlier texts and pages

1. F7

   The former llms.txt described the benchmark with v2.0 numbers as a product and contradicted itself. It is replaced together with this page.

2. F8

   Former machine-readable texts promised report contents (per-domain values, sector ranking) that were never delivered; only sector aggregates and the buyer’s own domain were delivered.

3. F9

   The former glossary quoted v2.0 values as current industry values without marking them as historical.

4. F10

   The former Deep Dive page described 14 signal groups, the overview 4 scoring groups; the method change had not been carried over there.

5. F13

   The former Deep Dive page announced availability “from early August”, which was not met.

6. F15

   A CERTavia blog post of 10 Jun 2026 gave values (289 domains, avg. CES 16.6) that the database does not confirm (288 measurements on 271 domains, avg. CES 22.85).

7. F18

   The page labelled level L4 as “SOVP-ready” and showed a count “SOVP-ready” per sector. SRS 75 is not a protocol threshold; both domains at 75 or above carry NOT_READY in the database. The scoring groups also rested on an outdated population (1,489 instead of 1,525) and three signal groups on none that could be traced. All values were recalculated from the frozen snapshot on 29 Oct 2026.

Technology and evidence

1. F11

   The four additional clusters of the Deep Dive report were empty in the database and appeared as 0.0 in the delivered CSV.

2. F12

   The report generator did not filter by measurement run. The reports delivered on 10 Jul 2026 were correct; later ones would not have been.

3. F14

   The benchmark measurements are not anchored in the vault (field vault_hash empty in all 3,333 rows) and therefore cannot be proven cryptographically after the fact.

08 // What this report does not contain

No domain names, no individual values per domain (including no top values per sector), no sectors under 10 domains (sample list of 6 financial institutions), no internal test entry and no subgroups covering only a small share of the domains (TLS detail group with 49, anchor signals as a raw group with 92 domains).

Origin of the protocol

Disclosure: the protocol measured here (SOVP) was designed by the author. Until October 2026 paid sector reports were offered on the basis of this data; that offer has been discontinued.

Script and database

The script shows every calculation rule and the SQL queries; the database itself contains domain names and is not published. The texts in the data set are in German.

Data set and script: Data set (JSON, 21 KB, aggregates only) · Script (Python, SHA-256 6c68f0d6…ea81c)

Suggested citation: Litzki, T. (2026): SOVP Benchmark 2026 – Machine readability of 1,525 domains for AI agents. Version 1.1, Litzki Systems LLC, 29 October 2026. https://litzki-systems.com/sovp-benchmark/ – Published under CC BY 4.0.

What agentic readiness means for your website: agentic SEO →