VENTURE · INVESTMENT DECISION INTELLIGENCE
ThesisDelta
A falling stock price does not tell you whether the business is broken.
That is the decision problem ThesisDelta is built to solve. It separates price damage from thesis damage, asks what actually changed in the business, tests whether the apparent opportunity survives a serious bear case, and forces the research to end in a bounded decision.
Instead of rewarding more information, the system rewards better evidence, explicit assumptions, disconfirming analysis, portfolio awareness, and a clear answer to one question: Why is today’s price attractive rather than simply lower than yesterday’s?
Evidence state: working research system applied in real-capital operator use with observable portfolio outcomes; attribution, benchmark-relative performance, and probability calibration remain separate evidence gates.
UNDER THE HOOD
Current-source research · forensic vetoes · archetype-aware underwriting · explicit assumption sheets · hostile scenario analysis · portfolio overlap · adversarial review · falsifiers · monitoring triggers · outcome learning.
The architecture behaves more like a small institutional research process than a stock screener: evidence is ranked, disagreement is preserved, and missing information can cap confidence rather than being silently filled.
THE DECISION PROBLEM
Markets produce information faster than judgment can metabolize it.
A falling price can signal temporary dislocation, permanent impairment, or noise. Conventional screens identify the movement. ThesisDelta is designed to interrogate the difference.
01 · PRICE
What moved?
Magnitude, timing, volume, catalyst, valuation reset, and market context establish what the tape is actually saying.
02 · THESIS
What changed?
Filings, guidance, cash generation, balance-sheet survivability, competitive position, accounting quality, and catalysts test whether the business case moved with the price.
03 · DELTA
Does the gap survive?
The candidate advances only if the apparent discrepancy survives vetoes, adversarial review, explicit assumptions, portfolio overlap, and an intelligent bear case.
Why is today’s price attractive rather than simply less expensive than yesterday?
FROM DAMAGE TO DECISION
A research sequence designed to resist narrative momentum.
The system puts disconfirming evidence early, makes assumptions visible, and closes the loop before research becomes a substitute for deciding.
01
Lock the frame
Classify the mandate and market regime, then freeze a deduplicated, liquid universe before enthusiasm expands the search.
02
Veto before scoring
Surface accounting opacity, financing stress, fraud risk, destructive dilution, and uninvestable liquidity before a high score can disguise them.
03
Underwrite the survivors
Test business quality, cash economics, valuation, management, peers, revisions, catalysts, and the specific condition under which the market is wrong.
04
Stress the assumptions
Expose base, bull, hostile, liquidity, earnings, valuation-compression, and thesis-break conditions. Simulation compares assumptions; it does not prove outcomes.
05
Map the portfolio
Identify concentration, factor duplication, correlation spikes, and whether several attractive names are secretly the same risk.
06
Close the decision
Assign one bounded state, record dissent and falsifiers, specify the next observable trigger, and stop when more research would not change the tier.
EVIDENCE BEFORE CONVICTION
Every claim carries a source, a timestamp, and a consequence.
Source hierarchy
- Primary: filings, earnings releases, official guidance, proxies, and investor materials.
- Market: exchange, price, volume, options, and positioning data with stated timestamps.
- Reporting: reputable financial news and transcripts used with attribution.
- Consensus: estimates and analyst views treated as signals, not authority.
- Sentiment: social and promotional material used only to understand positioning.
Missing data is a result
Unknown fields are labeled unavailable, estimated with method, or not material. Key omissions cap confidence instead of being silently filled.
Evidence is separated into blocking evidence, important but non-blocking evidence, and future monitoring triggers. A fact that cannot be known today does not become an excuse to defer forever.
Every serious candidate carries evidence confidence, statistical confidence, and execution confidence because high upside and high confidence are not the same proposition.
ARCHETYPE-AWARE UNDERWRITING
Different companies should not win by the same mathematics.
ThesisDelta changes the questions and scoring emphasis by opportunity type instead of forcing every security through one generic model.
HIGH-CONFIDENCE DIP
Reversibility over drama
Rewards durable demand, survivability, intact guidance, cash quality, and a selloff whose cause is temporary rather than structural.
AGGRESSIVE GROWTH
Scale with cash discipline
Tests growth durability, unit economics, margin trajectory, dilution, competitive advantage, valuation, and the path from narrative to cash flow.
STABLE COMPOUNDER
Durability at a sane price
Emphasizes return on capital, pricing power, recurring cash generation, management discipline, reinvestment runway, and valuation resilience.
SPECULATIVE ASYMMETRY
Possibility without category error
Separates enormous upside from confidence, foregrounding financing, binary events, liquidity, dilution, failure probability, and strict confidence caps.
THE STOP RULE
Every candidate ends somewhere.
The system does not permit “interesting—wait for more evidence” as a permanent refuge. Once the evidence supports a tier, research stops unless another investigation could change it.
01
Research-cleared
Evidence is sufficient for action consideration; portfolio fit, risk tolerance, sizing, and execution remain outside the tool.
02
Hold / review
Existing exposure or an active queue name should be monitored, not escalated.
03
Underwrite now
Enough company-specific evidence exists for deeper work today, and that work could change the tier.
04
Track only
A useful signal or watch item, but not an active candidate today.
05
Do not consider today
The candidate is removed from the current decision set with the reason recorded.
06
Blocked
A genuinely necessary input is missing or stale, so classification would be unreliable.
ADVERSARIAL BY DESIGN
A research system that argues with itself.
Functional review desks reach explicit judgments—approve, object, veto, or insufficient evidence—so disagreement survives into the final record instead of disappearing inside a polished answer.
Fundamental quality
Business durability, moat, free cash flow, valuation, management, and capital allocation.
Quantitative discipline
Base rates, factor exposure, distributions, volatility, correlation, sensitivity, and overfitting risk.
Risk and tails
Left-tail exposure, crowded trades, liquidity traps, correlation spikes, and downside survivability.
Forensic accounting
Revenue quality, cash conversion, dilution, concentration, auditor signals, and balance-sheet opacity.
The final record preserves the intelligent bear case, council dissent, falsifiers, evidence freshness, and the next observable trigger.
CURRENT EVIDENCE
A working research system now operating against real capital.
The underlying design artifact is Investment War Room Protocol v4.0: a research decision-support operating system for disciplined judgment under uncertainty, not an automated trading system.
ThesisDelta is being applied in a live six-figure portfolio with observable outcomes. That materially strengthens the evidence for functional use: research judgments are being tested against consequential decisions rather than hypothetical examples. Historical portfolio outcomes remain operating evidence, however—not proof that ThesisDelta caused those returns, generated alpha, or produced calibrated probabilities.
WORKING SYSTEM
Six linked research modules
- Research engine for current evidence, forensic pre-screening, company dossiers, and falsifiers.
- Ranking engine that separates opportunity archetypes before comparing them.
- Simulation engine with explicit assumptions, hostile sensitivity, and downside analysis.
- Portfolio-construction engine for concentration, correlation, factor overlap, and risk budgeting.
- Monitoring engine for review cadence, regime change, and thesis invalidation.
- Performance-learning engine for outcome tracking, calibration, and prospective model improvement.
The protocol also requires source hierarchy, assumption sheets, adversarial review, dissent, audit logs, and explicit sell or review logic.
PERFORMANCE GATE
Real outcomes are necessary; prospective attribution is stricter.
Historical positions can show that the system is being used in a real-money decision environment, including both successful and unsuccessful outcomes. They do not by themselves prove that ThesisDelta selected the position, established the entry, or caused the result.
The stronger test is prospective: freeze the research state before the subsequent return is known, retain the entry reference, thesis, archetype, confidence, bear case, falsifiers, monitoring triggers, and decision tier, then evaluate the outcome at predefined horizons.
Current public gate: accumulate enough prospectively frozen observations to test attribution, benchmark-relative outcomes, drawdown behavior, and calibration before making performance, alpha, or calibrated-probability claims.
Claim boundary. ThesisDelta is a working research system in real-capital operator use with observable portfolio outcomes. System maturity, historical outcomes, prospective attribution, benchmark performance, and probability calibration remain distinct evidence states.
COMPARATIVE EFFECTIVENESS
The test is not whether ThesisDelta can find winners. It is whether it improves decisions versus simpler alternatives.
A sophisticated research process has not demonstrated investment advantage just because a selected security rises. ThesisDelta is designed to face a harder counterfactual: under the same date, same available information, and same observation horizon, did its classification produce a better risk-adjusted decision than a reasonable traditional baseline?
THE COMPARISON DESIGN
Same date → same information set → ThesisDelta decision and predetermined baseline → same holding horizon → comparative outcome.
The baseline is specified before the outcome is observed. Retrospective reconstruction is labeled separately and never treated as prospective evidence.
01 · PASSIVE
Relevant benchmark
Compare the same capital window with an appropriate broad-market or sector benchmark. Absolute gains do not establish advantage if a passive alternative produced the same or better result.
02 · SIMPLE RULE
Naive dip heuristic
Test ThesisDelta against a deliberately simpler rule, such as buying a qualifying drawdown when core earnings or guidance remain intact. The system should add information beyond an obvious screen.
03 · INTERNAL CONTROL
Accepted versus rejected
Preserve serious candidates that ThesisDelta rejects. If research-cleared names do not subsequently separate from thesis-damaged or rejected names, the central classification has not demonstrated predictive value.
What would count as evidence of incremental value?
RETURN
Benchmark-relative outcome
Cumulative and median excess return versus the predetermined comparison, evaluated only after the relevant observation horizon matures.
DOWNSIDE
Loss avoidance
Maximum drawdown, severe-loss frequency, and downside capture test whether the process avoids structurally broken theses rather than simply maximizing upside.
SEPARATION
Price damage versus thesis damage
The core ThesisDelta hypothesis is supported only if “price damage / thesis intact” classifications subsequently separate from “thesis damage” classifications under comparable conditions.
CALIBRATION
Confidence that means something
Realized outcomes should eventually correspond to stated confidence bands. Misses, thesis breaks, and model changes remain in the record rather than being edited away.
Current evidence state. Comparative effectiveness is the next gate, not a current performance claim. ThesisDelta has established real-capital use and observable outcomes; it has not yet established benchmark-relative alpha, superior risk-adjusted returns, or calibrated predictive advantage.
THE BOUNDARY
What ThesisDelta refuses to do.
No trading
No broker access, order routing, automated execution, options execution, or hidden action.
No personalized sizing
No share counts, allocation targets, margin instructions, or commands to buy, sell, trim, or rebalance.
No certainty theater
No invented data, silent estimates, guaranteed outcomes, simulated proof, or confidence unsupported by fresh evidence.
No professional substitution
No tax, legal, fiduciary, accounting, or individualized financial advice.
THE INVITATION
A better answer is not more certainty. It is a decision that can be reconstructed.
ThesisDelta is being developed as a calmer, harder-to-fool research system: simple enough to use, rigorous enough to audit, and explicit about where its authority ends.
Research-system design only; not individualized investment advice. Please do not send account values, positions, credentials, confidential records, or other sensitive financial information in an initial inquiry.