Information arrives late.
An event can happen before it is disclosed. A backtest that uses the event date can give a model knowledge the market never had.
Loading EquityLens…
INDEPENDENT EQUITY RESEARCH
Research the signal.
Verify the evidence.
Markets are connected. Information has a history.
EquityLens investigates predictive equity signals from public information—with point-in-time integrity built in.
Research in development. Paper only. Evidence first.
RESEARCH WORKSPACEIllustrative interface · synthetic examples
A change is an observation. A signal is a hypothesis.
Compare a filing with a valid prior-year counterpart, using only documents visible at the decision time.
Knowledge cutoff enforced- 01Periodic disclosure
- 02Prior-year counterpart
- 03Entity continuity
- 04Pairwise text feature
PUBLIC EVIDENCE → DATED FEATURES → CONTROLLED EVALUATIONCONCEPT DEMO / NO LIVE DATA
Built around a simple standard
Public informationHistorical contextReproducible evidence
01 /THE PROBLEM
A convincing backtest
can tell the wrong story.
The hard part isn’t finding another pattern. It’s knowing whether the pattern survives an honest reconstruction of the past.
The past gets rewritten.
Restated data and today’s surviving companies can make historical results look better than they were. Missing and failed companies matter.
A result needs a trail.
A chart without its data vintage, assumptions, and experiment history is difficult to trust—and even harder to reproduce.
02 /THE PLATFORM
From public information
to testable hypotheses.
A research system for asking what was knowable, how it connects, and whether it adds predictive information.
Reconstruct what was known.
Not what we know now.
Availability timestamps separate the event from its disclosure. Dated security identity follows the instrument behind a symbol. Missing evidence stays missing.
- Knowledge-time feature cutoffs
- Dated identity and eligibility checks
- Explicit exclusions and quarantine
THE KNOWLEDGE BOUNDARYIllustrative
Make research reproducible.
Deterministic features, content-addressed snapshots, and hash-verified artifacts connect a result to the exact inputs that produced it.
VERSIONED INPUTS / TRACEABLE OUTPUTSChallenge every hypothesis.
Chronological evaluation, purging, embargo, negative controls, cost stress, and trial accounting make “no signal” an acceptable result.
CONTROLLED TESTS / EXPLICIT VERDICTS03 /RESEARCH METHODOLOGY
Every conclusion
has a chain of evidence.
Keep the historical boundary intact from acquisition to evaluation. Each stage makes its assumptions explicit and its outputs inspectable.
- 01
Acquire
Archive public evidence
- 02
Reconstruct
Respect knowledge clocks
- 03
Resolve
Identify the security
- 04
Generate
Build causal features
- 05
Guard
Check for leakage
- 06
Experiment
Freeze the hypothesis
- 07
Evaluate
Challenge the result
An auditable result. Including a clear “no signal.”
TIME IS A CONSTRAINT
Training precedes testing. Purging removes labels that cross a test boundary; embargo separates neighboring windows.
SKEPTICISM IS A FEATURE
Shuffled-label and planted-leak controls challenge the machinery itself. Baselines, uncertainty, and costs challenge the candidate.
04 /RESEARCH AREAS
Look beyond the chart.
Price and technical features serve as baselines. The broader research asks how disclosures, corporate context, and economic relationships contribute information.
01
Disclosure drift
Compare periodic SEC filings with their prior-year counterparts. Investigate whether changes in corporate language precede a market response.
Evaluated · not promoted
Evidence & limitations
Implemented deterministic text comparison; the evaluated disclosure hypothesis did not meet the promotion standard.
02
Customer → supplier
Ask whether information travels along disclosed economic relationships, from a customer’s news to its suppliers.
Design · data constrained
Evidence & limitations
A proposed directed relationship graph. Historical customer identity and edge coverage remain constraints.
03
Product-market peers
Explore connections between companies that compete in similar product markets, beyond broad sector classifications.
Experimental infrastructure
Evidence & limitations
Text-derived graph construction and eligibility tools exist. Predictive usefulness has not been established.
04
Insider context
Distinguish a transaction from its context: who traded, what they traded, and how it compares with prior disclosed activity.
Feasibility · limited coverage
Evidence & limitations
Form 3/4/5 parsing and purchase features exist. Prior aggregate hypotheses were rejected; contextual studies remain constrained.
05
Corporate events
Turn structured disclosures into dated, auditable events. Study the information content of an event without borrowing from its outcome.
Infrastructure · prior tests rejected
Evidence & limitations
Filing event features and event infrastructure are implemented. No event family has produced a promoted model.
06
Historical identity
Follow the security behind the ticker through renames, reused symbols, and changing eligibility windows.
Implemented foundation
Evidence & limitations
Dated identity and ticker-lease infrastructure is implemented. Complete historical membership and delisted price coverage remain unresolved.
Research areas describe hypotheses and engineering work, not demonstrated trading signals. No model has met the programme’s promotion standard.
05 /ENGINEERING
The infrastructure
is part of the science.
Reliable research needs more than a model. EquityLens connects acquisition, historical reconstruction, and evaluation through explicit contracts.
Purposeful data storage
Columnar research data and transactional ledgers, linked by content hashes and provenance.
Historical security identity
Issuer, security, listing, and dated ticker leases represent identity changes explicitly.
Testing and scientific controls
Offline fixtures, causal feature checks, chronological splits, and negative controls challenge correctness.
Governed experimentation
Registered hypotheses, append-only trial records, promotion gates, and immutable evidence preserve the research history.
06 /CURRENT STATUS
Ambitious by design.
Measured in practice.
EquityLens is an independent research project under active infrastructure development. This website presents the work; it does not offer access to a trading service.
Research foundations
Causal features, historical identity, evaluation controls, experiment governance, and paper-simulation machinery.
Evidence and data readiness
Governed acquisition and admission, historical reconstruction, and stronger coverage of identities and corporate actions.
Broader historical coverage
Historical membership, delisted prices, and relationship evidence remain barriers. Completeness is not assumed.
Future predictive research
Further evaluation depends on admissible data and explicit scientific authorization. Closed hypotheses are not silently retested.
THE STANDARD DOESN’T CHANGE
No promoted model. No claimed predictive performance. Paper-only scope.
07 /ABOUT EQUITYLENS
Better questions.
Evidence you can revisit.
EquityLens exists to make quantitative equity research more trustworthy: reconstruct the information available at the time, preserve the evidence, and evaluate the hypothesis under rules set before the outcome is known.
The long-term vision is a research platform where public corporate information and economic relationships can be studied together—without sacrificing historical integrity for a more flattering result.
A well-supported rejection is progress, too.
LET’S COMPARE NOTES
Interested in the research?
For research conversations and business inquiries, contact the founder.