The Investment Evidence Ledger
Direct Answer
An investment evidence ledger separates verified facts, rules, formulas, estimates, derived calculations, and interpretations. Each important claim links to a source, date, and review state. Without one, a thesis is hard to audit. With one, a changed source identifies every claim and page that depends on it.
What is an investment evidence ledger?
An investment evidence ledger is a structured record of every material claim in an investment thesis. Each entry records what the claim says, what type of claim it is, where the claim came from, when it was retrieved, and what conditions would make it stale. The ledger functions as an audit trail: if a source changes or a claim is later found to be incorrect, the ledger immediately reveals which other entries depend on it and which conclusions in the thesis are therefore affected.
The need for an evidence ledger arises from a structural problem in investment research. A typical thesis contains dozens of claims of different kinds. Some are observable facts that can be verified from primary sources. Some are regulatory rules that change on legislative or administrative schedules. Some are mathematical formulas whose outputs are only as valid as their inputs. Some are estimates or projections that carry explicit uncertainty. Some are analytical interpretations that represent the researcher's judgment rather than any external authority. These six types of evidence behave differently, age differently, and carry different risks when they prove wrong.
When all six types are mixed together in a single document without labeling, the thesis appears more certain than it is. An estimate dressed up as a fact, or a derived value whose input sources are not recorded, can survive long after the underlying evidence has been superseded. The evidence ledger forces separation: each claim is tagged with its type, anchored to a source, and given a staleness trigger so that it expires when it should rather than when someone remembers to check it.
The ledger is especially useful for theses that will be reviewed or revised over time, for research that will be shared with others, and for any situation where regulatory or tax rules affect the investment decision. It is also valuable for AI-assisted research workflows, where a language model may generate plausible-sounding claims that mix fact, estimate, and interpretation without distinguishing between them. A well-maintained ledger gives the researcher a mechanism for verifying AI-generated content entry by entry rather than treating the output as a uniform block of either reliable or unreliable text.
The six evidence types
Every entry in an investment evidence ledger belongs to exactly one of six types. The type determines how the entry ages, what a staleness trigger looks like, and what degree of confidence the claim carries.
Fact
A fact is a directly observable or verifiable datum: a closing price on a specific date, a statutory tax rate published by a regulatory body, an index level as of market close, a company's reported revenue for a specific fiscal quarter. Facts can be confirmed by checking the primary source, and the same check should produce the same result for any careful reader. Facts are the most durable evidence type when the underlying reality is stable, and the most quickly stale when it is not. A closing price from last quarter is a historical fact that will never change; a current interest rate is a fact that changes continuously.
When recording a fact in the ledger, the source must be the primary authority for that datum: the exchange record, the IRS publication, the company's SEC filing. A secondary source that quotes a fact from a primary source does not substitute for the primary source, because the secondary source may have introduced an error in transcription, may not update when the primary source changes, and does not resolve questions about jurisdiction or effective date.
Rule
A rule is a regulatory or legal requirement that constrains or enables an investment decision. Examples include the wash-sale disallowance window, the contribution limit for a specific account type in a given tax year, the short-selling circuit-breaker rule that applies after a 10% intraday price decline, or the margin requirement for a specific class of securities. Rules differ from facts in three ways: they are created by deliberate legislative or regulatory action, they have effective dates and sometimes sunset provisions, and they are jurisdiction-specific. A rule that applies in the United States may not apply in the European Union, and a rule that was current last year may have been amended or replaced.
A rule entry must record the jurisdiction and the effective date. It must also record the staleness trigger: for most rules, the trigger is any announced or enacted change in the relevant jurisdiction's laws or regulations in the area covered by the rule. The source for a rule should be the authoritative regulatory publication, not a secondary commentary. IRS publication numbers, SEC release numbers, and FINRA rule numbers are all resolvable to specific documents that represent the authoritative text.
Formula
A formula is a mathematical relationship used to derive a number from other inputs. Examples include the formula for compound annual growth rate, the dividend discount model, the Black-Scholes option pricing formula, or the net present value calculation used to build a discounted cash flow valuation. Formulas are stable in a way that inputs are not: the arithmetic relationship between variables does not change. What makes a formula entry stale is a change in the underlying inputs rather than a change in the formula itself.
Recording a formula entry requires stating the formula explicitly and recording the calculation lineage: the specific inputs used, where each input came from, and when each was retrieved. A formula entry that records only the output number without the lineage is unreviewable. If the output is later questioned, there is no record of whether the inputs were appropriate, which version of each variable was used, or which formula was applied when the investment was evaluated. The formula entry also provides a template for re-running the calculation: if an input is updated, the ledger tells the researcher exactly which formula to re-apply and which other derived values to update as a result.
Estimate
An estimate is a projected or uncertain quantity derived from analysis, judgment, modeling, or third-party forecasts. Revenue growth projections, analyst consensus price targets, probability-weighted scenario outcomes, and macroeconomic forecasts are all estimates. Estimates carry explicit uncertainty that facts do not, and they should be labeled as such so that a reader does not mistake a judgment for a verified figure.
An estimate entry should record the method used to produce it, the range of uncertainty if one was calculated, and the source if the estimate comes from a third party such as an analyst or consensus service. A staleness trigger for an estimate is typically an event that provides new information relevant to the quantity being estimated: an earnings release, a guidance update, a macroeconomic report, or a competitor announcement. Because estimates are inherently uncertain, the threshold for updating them is lower than for facts. If the new information is material, the estimate should be revised; the ledger provides the hook that connects the new information to the specific estimate entries that depend on it.
Derived value
A derived value is a number computed from other evidence entries, typically by applying a formula to facts and estimates. A free cash flow yield is derived from operating cash flow, capital expenditure, share price, and shares outstanding. An enterprise value is derived from market capitalization, debt, and cash. A normalized earnings figure is derived from reported earnings adjusted for a set of line items whose treatment has been analyzed and justified. Derived values are important to label separately because they inherit staleness from every entry in their lineage: if any input changes, the derived value changes, and anything that depends on the derived value changes as a result.
The lineage field for a derived value lists every other ledger entry that feeds into the calculation. This makes the ledger a directed graph: each derived value node points to the entries that produced it, and those entries may themselves be derived from others. When a source changes, tracing the graph forward identifies every derived value that is now stale, and then every conclusion or recommendation that depended on those values. Without this graph, a researcher discovers stale derived values only by coincidence or by re-running the entire analysis from scratch.
Interpretation
An interpretation is an analytical judgment: a conclusion drawn from examining evidence that is not itself directly verifiable from external sources. Examples include a view that a company's management team has a strong track record of capital allocation, an assessment that a competitive moat is narrow and weakening, or a judgment that the market is underpricing a regulatory risk. Interpretations are the researcher's own analysis, and they are the least durable type of evidence because they depend on the researcher's understanding of the competitive landscape, the regulatory environment, and the company's strategy, all of which change continuously.
Recording interpretations in the ledger serves a different function than recording facts and rules. The purpose is not to anchor the interpretation to an external source but to make explicit which conclusions rest on judgment rather than on verified information. An investment thesis that presents judgments as facts is harder to revise when circumstances change, because the researcher may not recognize which parts of the thesis were based on evidence and which were based on reasoning. The ledger makes this visible: when revisiting the thesis, the researcher can immediately distinguish which entries require re-verification from a primary source and which require re-assessment of the underlying reasoning.
Ledger field reference
A complete evidence ledger entry includes the following fields. Not every field applies to every entry type, but recording all applicable fields makes the entry self-contained and reviewable without access to the original research context.
- Claim: The specific assertion the entry documents. Write it as a complete declarative sentence. "The wash-sale disallowance window is 30 calendar days before and 30 calendar days after the sale date" is a complete claim. "Wash-sale rule" is not.
- Type: One of the six types described above: fact, rule, formula, estimate, derived value, or interpretation. Only one type per entry; if a claim requires two types, split it into two entries.
- Owner page: The canonical URL of the page or document within your research system that publishes this claim. Used to identify which pages need updating when the claim changes.
- Source: The URL or full citation of the primary authority for the claim. For facts, this is the primary data source. For rules, this is the authoritative regulatory publication with publication number and section reference. For formulas, this is the textbook or regulatory document that defines the formula. For estimates and interpretations, this is either the third-party source or a note indicating the claim is the researcher's own analysis.
- Source class: A tier indicating the quality and authority of the source. Tier 1 sources are primary regulatory documents, official statistical releases, and company filings. Tier 2 sources are institutional research, central bank publications, and exchange operator documents. Tier 3 sources are reputable secondary commentary with traceable methodology. Tier 4 sources are informal, unverified, or AI-generated without primary support. Claims in a thesis should be supported by Tier 1 or Tier 2 sources wherever the claim is material to the investment decision.
- Retrieval date: The date on which the source was accessed and the claim was verified. Many sources change over time; the retrieval date establishes which version of the source supported the claim at the time the thesis was written.
- Effective date: For rules and regulatory facts, the date on which the rule or fact became current. A rule that was enacted in one year and amended in another has a different effective date for each version.
- Jurisdiction: The legal or regulatory jurisdiction to which the claim applies. Most investment rules in the United States are jurisdiction-specific; a rule that applies to U.S. tax treatment does not apply to investments held in accounts governed by another country's tax law.
- Calculation lineage: For derived values: the list of other ledger entries that serve as inputs to this entry's calculation, along with the formula applied. For formulas: the formula written out explicitly with variable names defined.
- Conflicting evidence: Any source that contradicts or qualifies the claim. Recording conflicting evidence does not invalidate the claim, but it documents that the researcher was aware of the conflict and made a deliberate choice about which source to rely on and why.
- Staleness trigger: The specific event or date that should prompt re-verification of this entry. For regulatory rules, this is typically any enacted or proposed change to the relevant regulation. For market facts, this may be any trading day when the position is being evaluated. For estimates, this is typically each material company announcement or earnings release. A well-defined staleness trigger is what converts the ledger from a static record into a live audit system.
- Dependent pages and tools: The list of other pages, tools, or documents in the research system that use this claim. When the claim changes, this list identifies every artifact that needs to be reviewed or updated. This field is the outbound equivalent of calculation lineage: lineage records what feeds into an entry, dependent pages record what the entry feeds into.
Why an evidence ledger matters
Auditable investment thesis
A thesis that cannot be audited is a thesis that cannot be improved. Without a ledger, revising a thesis after a material development requires a researcher to re-read the entire document and identify by intuition which claims might be affected. With a ledger, a regulatory change fires the staleness triggers for all rule entries in the relevant jurisdiction, the researcher re-verifies those entries, and the dependent pages list tells them where to make updates. The audit is mechanical rather than judgmental, which means it is faster, more complete, and less dependent on the researcher's memory of which parts of the thesis were sensitive to which inputs.
Cascading impact of changed sources
Investment theses typically contain chains of inference: an estimate of revenue growth drives a projection of earnings per share, which drives a valuation multiple, which drives a price target, which drives a position size recommendation. If the revenue growth estimate changes because a competitor announced better-than-expected results, the change cascades through every downstream derived value in the chain. A ledger with calculation lineage makes this cascade visible before the researcher discovers it by accident. Every derived value whose inputs have changed is immediately identifiable, and the researcher can update them in dependency order rather than in the order they appear in the document.
AI and LLM content integrity
Language models are widely used in investment research workflows to draft analysis, summarize filings, and suggest interpretations. A structural challenge with AI-generated content is that facts, estimates, and interpretations are produced in the same prose style and at the same apparent confidence level, even when the model is uncertain or working from training data that predates the current regulatory environment. An evidence ledger is a tool for verifying AI-generated claims entry by entry: each claim is extracted, typed, and checked against a primary source before being incorporated into the thesis. Claims that cannot be verified are either flagged as estimates with explicit uncertainty or removed. This workflow does not eliminate the usefulness of AI assistance; it channels it through a verification step that catches the most common failure mode of AI-generated research, which is the confident statement of an unverified or outdated fact.
Worked example: the wash-sale rule
The following is a sample ledger entry for a specific regulatory rule. This example illustrates how the field set works in practice and what level of specificity is appropriate for a rule entry.
| Field | Value |
|---|---|
| Claim | A loss on the sale of a security is disallowed for U.S. federal income tax purposes if the same or substantially identical security is purchased within 30 calendar days before or 30 calendar days after the sale date. |
| Type | Rule |
| Owner page | https://www.getswoopr.com/learn/taxes-and-rules/ |
| Source | IRS Publication 550: Investment Income and Expenses, section "Wash Sales" |
| Source class | Tier 1 (primary regulatory document) |
| Retrieval date | 2026-09-08 |
| Effective date | IRC Section 1091, originally enacted in 1921, most recent relevant guidance as of tax year 2025 |
| Jurisdiction | United States federal income tax |
| Calculation lineage | N/A (this is a rule, not a derived value) |
| Conflicting evidence | Several broker interfaces describe the window as "30 days" without clarifying whether this means 30 calendar days or 30 trading days. IRS Publication 550 uses calendar days. This entry relies on the IRS publication as the authoritative source. |
| Staleness trigger | Any enacted or proposed change to IRC Section 1091, or any new IRS guidance material to wash-sale treatment of securities. Also flag for review if the investment context changes to include cryptocurrency, which as of the retrieval date is not subject to the wash-sale rule under U.S. law. |
| Dependent pages | Tax-loss harvesting strategy pages, account tax efficiency comparisons, any tool that calculates post-tax realized loss amounts |
This entry documents not only what the rule says but why the source was chosen over an alternative, what the boundary conditions are (calendar days versus trading days), and which adjacent areas are specifically excluded (cryptocurrency wash-sale treatment). A researcher reviewing this entry a year later knows exactly what to re-verify and which downstream tools to update if the rule changes.
Building a ledger for an existing thesis
If you have an existing investment thesis without a ledger, building one retroactively requires working through the thesis systematically rather than creating a comprehensive ledger on the first pass.
Start with the claims that do the most work: the central valuation assumption, the key regulatory constraint, the primary competitive differentiator. These are typically the claims whose incorrectness would change the investment decision. For each, ask which type it is, where it came from, and what would make it stale. Create one entry per claim, filling in as many fields as available sources support.
Next, work through derived values. Identify every number in the thesis that was calculated rather than observed directly. Trace each back to its inputs and record the lineage. If an input cannot be traced to a ledger entry, either create the entry or acknowledge that the derivation is unverifiable from the current documentation.
Then identify the interpretations. These are the judgments: the view on management quality, the assessment of competitive position, the thesis on why the market is mispricing the asset. Record each as an interpretation entry, noting the evidence that supports the judgment even if the judgment is not itself directly verifiable from primary sources.
Finally, set staleness triggers. Go through each entry and define the specific event that should prompt re-verification. Build a process around those triggers: a monitoring list of regulatory areas to watch, a calendar of earnings releases, a set of alerts on competitor announcements. The ledger is only as valuable as the process that keeps it current.
The ledger does not need to be comprehensive on the first pass. A ledger that covers the ten most consequential claims in a thesis is more valuable than a ledger that does not exist, even if it does not yet cover every claim. The goal is to reduce the probability that a changed source propagates undetected through the thesis, and partial coverage is substantially better than none.
FAQ
What is the difference between a fact and an estimate in an evidence ledger?
A fact is a directly observable or verifiable datum: a price on a given date, a statutory tax rate, an index level, a reported earnings figure. A fact can be confirmed by checking the primary source and should produce the same result for any careful reader. An estimate is a projected or uncertain quantity derived from analysis, judgment, or a model: a forecast of future earnings growth, an estimated probability that a regulatory change passes, or an analyst's projection of market size. Estimates carry uncertainty that facts do not. The distinction matters because a changed source can invalidate a fact automatically, while a changed assumption invalidates an estimate that may not be labeled as such anywhere in the thesis. Keeping them in separate ledger types forces you to count how many of your convictions rest on estimates rather than verified facts.
How do I decide which claims need ledger entries?
A claim needs a ledger entry when changing it would change a decision or conclusion in your thesis. Start by identifying the claims that do the most work: the revenue growth rate used in your valuation, the regulatory requirement that makes a business model legal, the formula used to compute a normalized margin. Claims that are decorative background do not need entries; claims that anchor a numerical output or a go/no-go conclusion do. A practical test: if someone proved the claim wrong, would you reconsider the investment? If yes, it belongs in the ledger.
What does "calculation lineage" mean in practice?
Calculation lineage is a record of which inputs and which formula produced a derived value. For example, if your thesis states that a company's free cash flow yield is 6.2%, calculation lineage would record: free cash flow = operating cash flow minus capital expenditure (figures from the latest annual report, retrieved on a specific date); market capitalization = share price on a specific date times shares outstanding; yield = free cash flow divided by market capitalization. If any input later changes, the lineage tells you immediately that the derived value is stale. Without lineage, a derived number floats free of its sources and can survive long after the inputs that produced it have been updated or retracted.
How often should ledger entries be reviewed?
The review cadence depends on the entry type and the staleness trigger. Regulatory rules change at defined legislative or administrative events and require review whenever those events occur or whenever a jurisdiction publishes proposed rule changes in the relevant area. Market facts such as prices or rates change continuously and should be refreshed whenever they are used in a live calculation. Estimates based on analyst models should be reviewed when the company reports earnings, issues guidance, or announces a material event. The staleness trigger field in each ledger entry defines this explicitly for that entry rather than applying a blanket schedule. For most investment theses, a structured review at every material company announcement and a comprehensive review at each quarterly filing is a reasonable minimum.
What is a staleness trigger and how do I define one?
A staleness trigger is the specific event or date that should prompt re-verification of a ledger entry. For a regulatory rule entry, the staleness trigger might be: any legislative change in the relevant jurisdiction, or the rule's scheduled sunset date. For a price-based fact, the trigger might be: any trading day when the position is being evaluated. For a company earnings estimate, the trigger might be: each quarterly earnings release. A well-defined staleness trigger converts passive monitoring into an active checklist: when the trigger fires, you know exactly which entries to re-verify and which downstream derived values may have changed as a result. Without a staleness trigger, entries age silently and the ledger gradually becomes a historical record rather than a live audit trail.
Educational use
This page is educational and informational. It does not tell a reader what to buy, sell, hold, or contribute, and it does not account for an individual's objectives, taxes, legal situation, benefits, debts, time horizon, or risk tolerance. Verify rules, limits, product terms, fees, and market data from current primary sources before acting.
References
Reviewed by the Swoopr Editorial Team in September 2026.