Work
Individual ProjectInvestment Analytics · Filings · Information RetrievalIn development

Private Markets Screening & Filing Intelligence

The system narrows a company universe with deterministic financial rules, then links the resulting screen to the filings and passages an analyst should review.

Central findingThe current system links 280 companies with SEC fundamentals and retrieves relevant evidence from 15,965 filing chunks, while keeping scoring rules and source passages inspectable.

Evidence

The result in context

Companies screened
283
Companies with SEC fundamentals
280
Filing chunks
15,965
Precision@5
89.4%
Relevant evidence among the top five retrieved passages.

Question

Can financial screening and filing retrieval reduce repetitive work without hiding the evidence or replacing analyst judgement?

A transparent screening and filing-retrieval workflow combining company fundamentals, SEC evidence and local search.

Workflow

From a broad universe to reviewable evidence

A deterministic scorecard filters companies using financial fundamentals. SEC XBRL data supplies structured evidence; filing retrieval then surfaces relevant passages for analyst review.

The system is intentionally not a black-box recommendation engine. Scores, source filings and retrieved text remain inspectable.

Research workflow

From universe to reviewable evidence

Automation narrows and retrieves; an analyst still owns the decision.

  1. 01Universe

    288 attempted · 283 successfully screened.

  2. 02Financial screen

    Transparent rules narrow the set without hiding exclusions.

  3. 03SEC fundamentals

    280 companies connect to structured public fundamentals.

  4. 04Filings

    35 filings supply primary-source context.

  5. 05Evidence retrieval

    15,965 chunks make passages searchable and testable.

  6. 06Analyst review

    The workflow ends with judgement, not automatic recommendation.

Interactive evidence

Valuation and growth are reviewed together, not collapsed into one rank

256 public-company observations with complete EV / sales and year-on-year revenue-growth fields.

Read: The screening score directs attention; the scatter preserves sector, valuation and growth context before any company is promoted for review.

Data table · 256 verified rows
Representative preview rows; the full 256-row local dataset loads with the interactive chart.
TickerCompany NameSector ThemeEv To SalesRevenue Growth PctMarket CapInvestment Screening Score
AMPAmeriprise Financial, Inc.Alternative Asset Managers / Market Infrastructure1.8421925.49481239,526,850,56060.3
BENFranklin Resources, Inc.Alternative Asset Managers / Market Infrastructure2.258353.45246516,293,222,40060
MCOMoody'S CorporationBusiness Services11.0027078.88826279,273,402,36844.6
SBUXStarbucks CorporationConsumer Services3.7402512.786915115,616,866,30439
ARRYArray Technologies, Inc.Energy Transition1.89529140.219611,429,816,19254.3
HCAHCA Healthcare, Inc.Healthcare Services1.8399537.07760385,968,453,63237.7
ROKRockwell Automation, Inc.Industrials6.5253920.9414150,827,714,56046.1
AFRMAffirm Holdings, Inc.Payments / Financial Services27.74950434.86140624,316,985,34460
KIMKimco Realty CorporationReal Estate / REITs11.5719565.06142816,438,250,49663.2
NETCloudflare, Inc.Software / Saa S33.81687929.84566679,696,035,84044.8
SEATVivid Seats Inc.Travel / Leisure / Consumer Platforms0.773865-30.4653194,381,86434.5
UBERUber Technologies, Inc.Travel / Leisure / Consumer Platforms2.90995118.279594142,726,021,12047

Retrieval

Evaluation is attached to the research workflow

Across a deterministic 96-question gold set evaluated against cached sample filing chunks, the current benchmark reaches 100% Hit@5, 89.4% Precision@5 and 100% mean reciprocal rank.

Hit@5 asks whether at least one relevant passage appears in the first five results; Precision@5 asks how many of those five are relevant.

Interactive evidence

Retrieval evaluation is attached to the research workflow

Selected committed evaluation metrics across 96 questions.

Read: Hit rate is 100% at three and five results; Precision@5 is 89.44%, keeping the quality trade-off visible.

Data table · 6 verified rows
Complete dataset
MetricMetric TypeValue Pct
Hit Rate At 3Retrieval100
Hit Rate At 5Retrieval100
Precision At 3Retrieval98.148148
Precision At 5Retrieval89.444444
Citation CoverageGroundedness100
No Answer AccuracyGroundedness100

Interactive evidence

Risk flags remain separate from the opportunity score

Mean score across all 283 screened companies.

Read: Data-gap risk is lowest on average; drawdown, red-flag and volatility measures cluster around 50 and remain review prompts rather than verdicts.

Data table · 5 verified rows
Complete dataset
LabelValue
Credit Risk43.284452
Red Flags50.032155
Drawdown Risk50.176678
Volatility Risk50
Data-Gap Risk27.396631

Status

In development, with boundaries kept visible

The system currently supports structured screening, filing ingestion and local retrieval. Optional local generation remains downstream of source retrieval and human review.

Limitations

What this evidence does not establish

  • The retrieval metrics describe a bounded evaluation set and should not be generalised to every filing or analyst question.
  • The benchmark uses cached sample filing chunks and is not evidence of legal accuracy, exhaustive diligence or complete real-filing coverage.
  • A screen prioritises attention; it does not replace diligence, valuation judgement or primary-source review.

Source and reproducibility

Trace the evidence

Source code, evaluation outputs and supporting material are available in the repository.

View repository
  1. Valuation versus growthdata/processed/investment_scores.csvCommit / evidence ID: 164ad28bd806ccdc182bef3886d4a6e8a9ac4193
  2. Retrieval evaluationdata/processed/rag_eval_results.csvCommit / evidence ID: 164ad28bd806ccdc182bef3886d4a6e8a9ac4193