αx Benchmark
Composite, pillar and theme scores across the companies GreySpark has assessed, and where any one company sits among them. Internal to GreySpark.
Scores are compared only within one methodology version. v1.8 changes how most themes are scored, and Themes 4 and 5 now measure different things, so v1.8 results will form a separate series.
Composite score
Pillars
Each pillar on its own 1 to 5 scale. Risk Mitigation reads on the risk scale, Unmanaged to Optimised.
Themes
The highlighted company's score as a bar. The grey line spans the lowest to the highest company, and the black tick marks the average.
Market signals beside these themes
Open Market signalsPublic signals from Stream 3 that relate to a theme. They are shares of companies where a signal was found, not 1 to 5 scores, so they sit beside the benchmark and never enter it.
Companies in this view
Select a row to highlight it above.Notes on this mockup (v0.3)
- What changed in v0.3. Themes 4 and 5 carry the names from William’s 5 October question set, "AI in the Product" and "AI in Software Development". A Methodology filter keeps v1.7.1 and v1.8 results apart: v1.8 changes the scoring of 24 of the 36 metrics, and Themes 4 and 5 now measure different things. The market signals for AI hiring and skills now relate to O2 and O7, since O3 is no longer talent, and the engineering signal relates to O5 to O8. Signal codes keep Cara’s original numbering. The ±5 point note is gone, as it is not yet validated.
- Segments. All ten segments, with "All segments" as the base. Each shows how many companies it holds.
- Both assessment types together by default. Due-diligence companies are filled dots and self-assessments are rings, so the two stay visible in one pool. "Assessment type" narrows to one. Self-assessment scores are indicative.
- Period and comparison. "Completed" filters by the date the assessment finished, by quarter. "Compare with" adds that period's average as a grey line on the composite and pillars, with the difference in the headline figures.
- One average everywhere. The Companies and Deciles views, the tiles and the table all use the same companies, so they show the same average.
- Deciles, simplified. Ten equal blocks with each decile's score range, and the highlighted company's block filled. They need at least 20 companies in the selection; below that the Companies view shows them all.
- Market signals. The toggle adds Stream 3's public signals beside the themes they relate to, for the segments Stream 3 covers so far. They are never mixed into scores. The full picture is on the new Market signals tab.
- Nothing below the themes, and risk on its own scale, as in v0.1.
Market signals
Public signals about companies in GreySpark's segments, collected from company websites and Companies House by the Stream 3 proof of concept. No interviews and no scores: each figure is the share of companies where a signal was found.
What was found, by αx pillar
Each signal shows the αx metrics it relates to (from Cara's Stream 3 set), how it was collected, and the share of companies where it was found in each run.
In Cara's set, not collected yet
Cara's Stream 3 set has 21 public signals. The proof of concept collected seven of them as counts, plus four rubric measures and three context signals. These are the rest.
How Stream 3 could connect to the benchmark: to investigate
- Clean overlaps sit beside a theme. Senior AI leadership (V1), launch cadence and agent readiness (V6), hiring (O2 and O7), governance disclosure, control language and model providers (R1 to R4) each relate to one metric. The Benchmark toggle shows them as shares of companies, beside the theme.
- Rubric measures use their own scales. AI Product Centrality runs 0 to 3 and AI Engineering Signal runs 0 to 4, and the programme proposes only the lower bands. Putting them on αx's 1 to 5 scale would need a mapping that GreySpark agrees first.
- Assessor measures need a person. Product Replaceability Exposure and Competitive AI Pressure are judgements by design, so they appear here as gaps until an assessor scores them.
- The other direction. Companies with both a due-diligence or self-assessment score and Stream 3 signals would show which public signals track real scores. That needs companies in both sets, which there are not yet.
- Coverage. Three of ten segments so far, and most signals are found for a minority of companies. Low shares partly reflect what companies publish, not only what they do.