Publish 15 clearly labelled synthetic intelligence examples

README.md

@@ -6,31 +6,51 @@ tags:
 - company-intelligence
 - historical-analysis
 - mcp
-pretty_name: Vivameda Synthetic Intelligence Samples
+size_categories:
+- n<1K
 ---
-# Vivameda Synthetic Intelligence Samples
+# Vivameda synthetic intelligence samples
 
-**Synthetic demonstrations only. No real company data, panel extracts, numerical forecasts or research results.**
+15 fictional question-and-answer examples for developers exploring company intelligence and historical reasoning in agent workflows. Version 2, 10 October 2026.
 
-Three fictional examples show the intended answer style for strategy reasoning, diligence and reference forecasting. They are newly authored illustrations, not outputs of the production engine, training records, an evaluation benchmark or evidence of predictive accuracy.
+**Every row is synthetic and manually authored for this showcase. These are illustrations of reasoning boundaries, not outputs from the production engine, a benchmark, a training recommendation, or evidence of model performance. No proprietary panel or provider records were used to create these examples.**
 
-| Use case | Demonstrates |
-| --- | --- |
-| Strategy reasoning | Separate observed expansion from commercial conclusions |
-| Diligence | Preserve missing evidence rather than infer closure |
-| Reference forecasting | Separate historical reference outcomes from individual predictions |
+## Sample sets
+
+| Use case | Examples | Purpose |
+|---|---:|---|
+| Strategy reasoning | 5 | Separate structural observations, hypotheses and business conclusions |
+| Company comparison | 4 | Examine comparability, coverage and historical peer selection |
+| Reference forecasting | 4 | Explain reference classes, horizons and limits of transfer |
+| Diligence | 2 | Separate missing observations and corporate-event evidence |
+
+All sets are in `samples.jsonl`; filter on `use_case`. The original three examples remain unchanged.
+
+## Fields
+
+`sample_id` is a stable fictional identifier. `synthetic` is always true. `company` is explicitly fictional. `question` and `answer` illustrate an analysis interaction. `reference_sample_size` and `outcome_rate` are always null: this release supplies no numerical forecast or reference calculation. `limitations` states the evidence boundary.
+
+## Example
+
+Question: Can one named analog establish a base rate?
+
+Illustrative answer: No. One analog is a case illustration, not a population estimate. Separate analog narratives from aggregate reference rates and disclose how comparators were selected.
 
 ## About Vivameda
-Vivameda is building historical intelligence for agents, combining private longitudinal company evidence, numerical analysis and evidence-grounded reasoning through MCP.
 
-Canonical panel scope: 4M+ companies, 48M observed company-year records, 1950–2020, 100+ countries. The real panel is RETROSPECTIVE-RECONSTRUCTION; observed professional profiles are not payroll employees. It is not included in this repository.
+Vivameda's proprietary historical panel covers 4M+ companies and 48M observed company-year records across 1950–2020 and 100+ countries. The panel was reconstructed retrospectively. Observed professional profiles are not payroll employees; missing is not zero. These public examples contain none of that panel.
+
+Historical reference analysis can support strategy questions. Forecast validation, acquisition inference, commercial performance and current-company transfer require their own evidence and tests. This dataset grants no access to production tools, company data or model weights.
 
-## Files and fields
-samples.jsonl contains three fictional question-and-answer examples. Every record has synthetic=true. Null reference_sample_size and outcome_rate mean no calculated result is supplied; they do not mean zero.
+## Intended use
 
-## Limitations
-These samples illustrate wording and evidence boundaries only. They contain no proprietary data, scoring logic, feature recipes, weights or verified company findings. They cannot measure model accuracy or certify client availability. Independent evaluation and client release remain separate work.
+Inspect sample responses, discuss integration requirements and prototype how an agent presents evidence limitations. These examples do not specify a stable production MCP API and should not be treated as ground truth for investment decisions.
 
 ## Contact
-For a scoped sample or integration discussion: https://vivameda.com
-Commercial delivery through Vivameda Ltd, Cyprus.
+
+For company intelligence, data partnerships or historical context for agent workflows: https://vivameda.com
+
+## Version history
+
+v1: 3 synthetic examples, published commit 25cc52273b79f39d094e587083e8b21eb51562f4.
+v2: retains those examples and adds 12 across the four use cases. Publication pending.

samples.jsonl

@@ -1,3 +1,15 @@
 {"sample_id":"SYN-001","synthetic":true,"company":"Fictional Northbridge Instruments","use_case":"strategy reasoning","question":"Does recent expansion establish sustainable commercial growth?","answer":"The synthetic observations describe expansion, not commercial success. A production analysis would test coverage consistency, calculate a historical reference class and distinguish observed outcomes from inference. Financial performance remains unverified.","reference_sample_size":null,"outcome_rate":null,"limitations":"No real records or computed reference results. Profile observations are not payroll employee totals."}
 {"sample_id":"SYN-002","synthetic":true,"company":"Fictional Meridian Biosystems","use_case":"diligence","question":"Do missing observations establish company closure?","answer":"No. Missing observations are a coverage gap. A production analysis would retain the missing state, verify identity and seek dated corporate evidence before discussing closure.","reference_sample_size":null,"outcome_rate":null,"limitations":"Illustrative reasoning only. No company, corporate event or outcome has been verified."}
 {"sample_id":"SYN-003","synthetic":true,"company":"Fictional Elmfield Software","use_case":"reference forecasting","question":"What does a historical reference class tell us about this company?","answer":"A reference class describes resolved observations among defined historical comparators. It does not automatically establish an individual-company probability. A production response would state match criteria, sample size, horizon, missing follow-up and comparability limits.","reference_sample_size":null,"outcome_rate":null,"limitations":"No actual reference class or forecast supplied. Retrospective reconstruction does not establish point-in-time forecasting validity."}
+{"sample_id":"SYN-004","synthetic":true,"company":"Fictional Cedar Robotics","use_case":"strategy reasoning","question":"Should expansion into sales change our strategy assessment?","answer":"A shift toward commercial roles can motivate a distribution hypothesis, but does not establish revenue growth. Verify observation coverage and compare the earlier operating structure with historical peers before interpreting the shift.","reference_sample_size":null,"outcome_rate":null,"limitations":"Fictional demonstration authored for this showcase. No panel records, computed reference results or validated forecast. Observed profiles are not employees."}
+{"sample_id":"SYN-005","synthetic":true,"company":"Fictional Harbour Components","use_case":"strategy reasoning","question":"Does a broader capability mix demonstrate diversification?","answer":"Breadth can describe organizational structure; it does not establish diversified customers or revenue. Separate workforce breadth from business-segment evidence and request dated revenue or customer information.","reference_sample_size":null,"outcome_rate":null,"limitations":"Fictional demonstration authored for this showcase. No panel records, computed reference results or validated forecast. Observed profiles are not employees."}
+{"sample_id":"SYN-006","synthetic":true,"company":"Fictional Larch Therapeutics","use_case":"strategy reasoning","question":"Is technical hiring evidence of a successful pipeline?","answer":"Technical expansion can be consistent with investment in development. It does not establish clinical success. Use dated program evidence to test the hypothesis and keep research activity separate from commercial outcomes.","reference_sample_size":null,"outcome_rate":null,"limitations":"Fictional demonstration authored for this showcase. No panel records, computed reference results or validated forecast. Observed profiles are not employees."}
+{"sample_id":"SYN-007","synthetic":true,"company":"Fictional Summit Systems","use_case":"strategy reasoning","question":"Does slowing growth mean the strategy failed?","answer":"A slowdown may reflect coverage, maturation or a change in operating model. Check observation consistency and compare relevant historical trajectories before choosing an explanation.","reference_sample_size":null,"outcome_rate":null,"limitations":"Fictional demonstration authored for this showcase. No panel records, computed reference results or validated forecast. Observed profiles are not employees."}
+{"sample_id":"SYN-008","synthetic":true,"company":"Fictional Alder and Birch Manufacturing","use_case":"company comparison","question":"Which company is stronger when observed profile totals differ?","answer":"Observed profile totals alone cannot rank business quality. Compare coverage, sector, age and operating structure; state which comparisons remain unsupported.","reference_sample_size":null,"outcome_rate":null,"limitations":"Fictional demonstration authored for this showcase. No panel records, computed reference results or validated forecast. Observed profiles are not employees."}
+{"sample_id":"SYN-009","synthetic":true,"company":"Fictional Shore and Valley Software","use_case":"company comparison","question":"Can two companies with similar growth have different structures?","answer":"Yes. Similar observed growth can coexist with different role distributions. Present comparable role summaries and keep a structural difference separate from a claim about future performance.","reference_sample_size":null,"outcome_rate":null,"limitations":"Fictional demonstration authored for this showcase. No panel records, computed reference results or validated forecast. Observed profiles are not employees."}
+{"sample_id":"SYN-010","synthetic":true,"company":"Fictional Ember and Frost Biotech","use_case":"company comparison","question":"How should missing years affect a peer comparison?","answer":"Missing follow-up reduces comparability and stays missing. Use comparable observed periods, disclose excluded periods and do not replace missing values with zeros.","reference_sample_size":null,"outcome_rate":null,"limitations":"Fictional demonstration authored for this showcase. No panel records, computed reference results or validated forecast. Observed profiles are not employees."}
+{"sample_id":"SYN-011","synthetic":true,"company":"Fictional Oak and Pine Logistics","use_case":"company comparison","question":"Can the present-day peer list be used for historical comparisons?","answer":"Present-day peers may introduce selection and survivorship bias. State how historical peers were defined and whether the comparison uses retrospective reconstruction.","reference_sample_size":null,"outcome_rate":null,"limitations":"Fictional demonstration authored for this showcase. No panel records, computed reference results or validated forecast. Observed profiles are not employees."}
+{"sample_id":"SYN-012","synthetic":true,"company":"Fictional Atlas Devices","use_case":"reference forecasting","question":"What should a three-year reference response contain?","answer":"It should describe the reference class, match criteria, resolved follow-up, horizon and uncertainty. Report code-computed counts and rates only when available; withhold a result when the evidence is insufficient.","reference_sample_size":null,"outcome_rate":null,"limitations":"Fictional demonstration authored for this showcase. No panel records, computed reference results or validated forecast. Observed profiles are not employees."}
+{"sample_id":"SYN-013","synthetic":true,"company":"Fictional Delta Analytics","use_case":"reference forecasting","question":"Can one named analog establish a base rate?","answer":"No. One analog is a case illustration, not a population estimate. Separate analog narratives from aggregate reference rates and disclose how comparators were selected.","reference_sample_size":null,"outcome_rate":null,"limitations":"Fictional demonstration authored for this showcase. No panel records, computed reference results or validated forecast. Observed profiles are not employees."}
+{"sample_id":"SYN-014","synthetic":true,"company":"Fictional Horizon Materials","use_case":"reference forecasting","question":"Can historical reference rates become a current risk score?","answer":"Historical reference rates do not establish transfer to a current company. Label the historical observation separately and require a transfer test before making a validated forecast claim.","reference_sample_size":null,"outcome_rate":null,"limitations":"Fictional demonstration authored for this showcase. No panel records, computed reference results or validated forecast. Observed profiles are not employees."}
+{"sample_id":"SYN-015","synthetic":true,"company":"Fictional Willow Services","use_case":"diligence","question":"Does a falling observed profile count establish acquisition?","answer":"No. Contraction alone does not establish acquisition or closure. Verify dated corporate events independently and distinguish the workforce observation from the event explanation.","reference_sample_size":null,"outcome_rate":null,"limitations":"Fictional demonstration authored for this showcase. No panel records, computed reference results or validated forecast. Observed profiles are not employees."}