Insights Charaka Notes Analyse a Deal — £9
In production at a venture fund — running its autonomous analytical, research, and operational flows. This dashboard is that pod, anonymised.
UTC--:--:--
IST · Bengaluru--:--:--
PT · San Francisco--:--:--
Now firing Listening…
0 fired today · 0 upcoming
00:00 06:00 12:00 18:00 HUMAN · MD
Firing now (±15min) Fired today Upcoming
Data Engineering Analytical Council Finance & Capital Engineering Product GTM Founder's Associate Consulting
46Agents in production
8Divisions
43Scheduled firings / day
90,022Knowledge-graph entities
13,993Companies tracked
65,898Relationships modelled
298Scorecards graded (self)
63%Weighted backtest accuracy
53.7%Decline rate
79Learning entries

Backtest figures shown here are the strictest subset — 298 blind-backtest verdicts where the system was given no outcome data at scoring time. The broader rolling calibration corpus is larger and includes earlier non-blind verdicts; methodologies differ by subset.

The system grades itself

Last 15 of 298 backtest verdicts on private companies. Each verdict was generated from publicly available information only and scored against the real outcome — none of these are deals the Fund is currently evaluating. Average accuracy: 63%. Wins in green, misses in red — no cherry-picking.

CompanySectorVerdictOutcomeScore
Scapia Fintech Lean Aligned Raised up 0.70
Paper Boat D2C Lean Decline Raised flat 0.60
NoPaperForms Edtech Lean Aligned Raised up 0.70
Censys Cybersecurity Lean Aligned Raised up 0.70
15Five HR Tech Lean Aligned Stalled 0.15
Unbox Robotics Robotics Lean Aligned Raised up 0.70
Strapi SaaS Lean Aligned Raised up 0.70
Portal Space Systems Space Tech Lean Aligned Raised up 0.70
NeoGrowth Fintech Lean Decline Raised flat 0.60
Bankjoy Fintech Lean Decline Raised up 0.15
Illumio Cybersecurity Lean Aligned Raised up 0.70
Blue Frog Gaming Gaming Lean Decline Stalled 0.70
Caption Health Healthcare Lean Decline Raised up 0.15
Arkeus DefenceTech Lean Aligned Raised up 0.70
HyperPlay Gaming Lean Decline Raised up 0.15

What the system has learned from breaking

Recent entries from the failure log. Each one is a real bug, missed verdict, or process regression — captured so the next iteration doesn't repeat it.

  • 2026-05-03 Cross-agent learning open

    Weekly backtest fabrication near miss

    Scheduled tasks that synthesize output must precondition-gate on source data existence before any synthesis step. Without a gate, a task with an incorrect source path will fabricate data rather than fail loudly.

  • 2026-05-03 Entity quality open

    Entity Quality Regression: Null company_name and Default Mental Health Sector Contamination

    4 entities had null company_name field: freedom-forever, candid_health, yoco, meltplan. Add non-null company_name validation gate to scrape-to-KG write path

  • 2026-05-03 Filter calibration open

    NR-9.1 Amendment: Add Chapter 11 and Related Distress Stages to Pool Integrity Exclusion List

    The NR-9 pool integrity filter's stage exclusion list does not include Chapter 11 or other bankruptcy/distress stages. Freedom Forever passed the filter despite being in Chapter 11 bankruptcy.

  • 2026-04-11 Cross-agent learning open

    Website deployment cascade failure

    ManthanBot website launch session (11 April 2026) produced a cascade of 5 distinct failures across CTAs, progress bar, and results page rendering.

  • 2026-04-08 Cross-agent learning open

    Death diagnosis graphcore

    Four calibration rules from Graphcore blind vs outcome comparison: CR-1 revenue can decline (not just plateau) — check for ecosystem-dependent companies; CR-2 platform shift signals (like GPT-3 for Graphcore) should carry -0.10 confidence penalty; CR-3 exceptional team prevents…

  • 2026-04-05 Cross-agent learning open

    Aira sprint framework effective

    Aira sprint framework (request → research → findings → impact → next steps → closure) is effective when runtime exists. [sprint] (SEBI disclosure) produced a clear Option B decision and was properly closed with Tvashtar follow-up.

Showing 6 of 22 publicly-shareable entries · 79 total in the learning graph

Today's compute spend $0.00 0 firings completed
Cost per firing (avg) $0.00 Across the full agent mix
Projected daily $0.00 All 43 firings, end of day UTC
Cost per scorecard $0.38 Vs. analyst hour at $120

Estimated, based on a Sonnet-class run mix. Actual spend varies with prompt caching, model routing (Haiku for sweeps, Sonnet for analysis, Opus for synthesis), and agent verbosity. Real telemetry replaces these once the system moves fully to metered server infrastructure.

What you're looking at

This is not a mock-up. The clock above is your local browser's clock, mapped onto a 24-hour face. Each dot is a scheduled task running inside an actual venture fund. As you read this paragraph, an agent somewhere on that ring has just fired, or is about to.

The fund is two humans (partner + analyst) and forty-six agents. Eight divisions — research, deal flow, analytical deliberation, financial modelling, engineering, product, GTM, and consulting — each operate as small panels of specialised agents that disagree, deliberate, and produce structured outputs. Threads of work pass between them. The output lands on the partner's desk; the next loop starts.

The numbers above are live counts from our knowledge graph and backtest pipeline. The accuracy figure is the system grading its own analytical history — 298 deals scored against actual outcomes under strict blind methodology (the system sees no outcome data at scoring time). It improves week by week. The learning entries are the failures the system has chosen to remember.

The 50-second compression below is the same day, rendered as art. The browser version above is the same day, rendered as time.

24 hours, compressed. Same fund. Same day. Open full Day-Loop page →