Try the living X-Matrix with an agent
This is Hoshivest, a fictional turnaround fund with three industrial holdings. Its matrix carries the fund’s vision and objectives, and each holding is one of its projects. Each holding keeps its own matrix, with its own objectives, projects and indicators, and its first long-term objective takes up the mission the fund gave it. The fund reads in them, every month, the figures it requires: adjusted EBITDA, net cash, services still provided by the seller. Among the projects, an open question: can an agent read the reporting and prepare the review? Click a holding’s line, then “Open the matrix of…”, to see its own matrix, or ask the assistant in the corner.
Vision
Hoshivest takes over industrial businesses with €50m to €500m in revenue that their parent group wants to sell, and leads them in three stages: stabilize (exit the seller’s systems, hold the cash), turn around (margin, deliveries, customers), prepare the sale. By 2029, every holding sold is standalone, profitable and carries its own growth plan, and every sale has returned at least three times the capital invested.
Hoshivest 2026-2029
21 elements · 27 correlationsScroll or pinch to zoom, drag to move. Click an element to light its deployment chain and open its sheet. Esc closes.
The indicators, month by month
The same strategy, read through its numbers: each indicator under the annual objective it measures, its last readings, where it is heading and whether the target is met.
- 8 indicators
- 8 off target
- 1 with a reading missing
| Indicator | Target | Aug 26 | Sep 26 | Oct 26 | Trend | Owner |
|---|---|---|---|---|---|---|
| Every holding out of its seller’s systems in less than 12 months | ||||||
| Portfolio adjusted EBITDA, per monthTarget not met · result · also measures: Positive adjusted EBITDA in every holding before its 18th month ; One sale a year from 2027 ; Two build-ups per platform by 2028 | 2500 k€ | 1170 | 1970 | 2070 | improving | Sophie Lemaire |
| Hoshifonte net cash, month-endTarget not met · result · also measures: Positive adjusted EBITDA in every holding before its 18th month | 5000 k€ | 4600 | 4500 | 4450 | worsening | Sophie Lemaire |
| Services still provided by sellers (TSA)Target not met · performance · also measures: Positive adjusted EBITDA in every holding before its 18th month | 0 services | 22 | 18 | 15 | improving | Sophie Lemaire |
| Positive adjusted EBITDA in every holding before its 18th month | ||||||
| Savings from pooled purchasing, 2026 to dateTarget not met · performance | 1200 k€ | 60 | 140 | 230 | improving | Claire Vasseur |
| Hoshitherm margin on jobs signedTarget not met · result · also measures: Two build-ups per platform by 2028 | 19 % | 16.4 | 17.1 | missing | improving | Sophie Lemaire |
| One sale a year from 2027 | ||||||
| VDD findings still open at HoshipackTarget not met · performance | 0 findings | 26 | 17 | 11 | improving | Sophie Lemaire |
| Two build-ups per platform by 2028 | ||||||
| No indicator measures this objective. | ||||||
| Tied to no annual objective | ||||||
| Holdings with positive adjusted EBITDATarget not met · result | 3 of 3 | 2 | 2 | 2 | stable | Sophie Lemaire |
| Reporting delay, slowest holdingTarget not met · performance | 5 working days | 7 | 6 | 6 | stable | Sophie Lemaire |
One target per indicator, for the year. "Missing": the reading of the last closed month is due, or of the last closed quarter for a quarterly indicator (shown in the month that closes it). * provisional, its period is still running.
The bets of the year
Each initiative is a bet that this work will move those indicators. This reads the bet from what the matrix holds: how far the work has gone, and whether its indicators follow. It shows where to look; the review decides.
| Initiative | Progress | Indicators it should move | Reading |
|---|---|---|---|
| Hoshitherm: turnaround and first build-up (taken over in October 2024)Claire Vasseur | 70 % |
| A linked indicator is on target or improving |
| Hoshifonte: stabilization (taken over in March 2026)Claire Vasseur | 45 % |
| Too early to read |
| Hoshipack: preparing the sale (held since June 2022)Thibault Renard | 40 % |
| Too early to read |
| Pooled purchasing: energy, transport, insuranceClaire Vasseur | 30 % |
| Too early to read |
| AI trials: an agent reads the reporting and prepares the portfolio reviewSophie Lemaire | not stated |
| Too early to read |
| Two new platforms taken over in 2027Thibault Renard | not stated | none linked | Cannot be read: no indicator is linked |
A bet is read once the initiative is at least halfway or marked completed. An indicator on target does not prove the initiative caused it, and one that is not is a reason to ask why, not a verdict on the team.
Three things to ask the assistant
- 1.Which holding needs a decision from the investment committee, and why?
- 2.What is missing from last month’s reporting?
- 3.Prepare the monthly portfolio review: what is due and what needs a decision.
What is happening under the hood
- The agent reads through MCP. Answers rest on tool calls to the Hoshin Kanri MCP server: get_matrix_data, get_matrix_completeness, diagnose_theme_deployment. The same 44 tools your own agent gets.
- The page lends it four tools. highlight_element, focus_quadrant, get_canvas_state and reset_view are declared by this page (WebMCP). When the agent names an element, it can light it here. Read-only: the demo cannot change the matrix.
- Nothing is staged. Same matrix, same MCP server, same read-only demo token, open to an MCP client that accepts a bearer token. Your own matrix gets the same treatment, with your rights.
MCP URL: https://hoshin.app/mcp — works with claude.ai, Claude Desktop and Cursor.