---
title: "Altss Source Reliability & Confidence… | Altss Frameworks"
description: "The Altss methodology for grading sources and assigning confidence labels to private-markets claims: source authority classes, reliability and credibility…"
canonical: "https://altss.com/knowledge-center/frameworks/source-reliability-and-confidence-methodology"
---

Methodology · Version 1.2 · Last reviewed 2026-10-01

# Altss Source Reliability & Confidence Methodology

The Altss methodology for grading sources and assigning confidence labels to private-markets claims: source authority classes, reliability and credibility grades adapted from the NATO Admiralty system, independence, use of validation status, conflict handling, label meaning and limits, calibration and decay.

Publisher: Altss LLCContent modified 2026-10-01

LIB-03

Image: Abstract illustration: a flowing wave of connected points on a dark blue field.

Altss editorial illustration. AI-generated and decorative; it does not show data.

## 1. Purpose

This methodology sets out how Altss judges the sources behind a private-markets claim and how it expresses the result as a confidence label. It separates questions that are often merged into one score: what standing this kind of source has for this kind of claim, how dependable this particular source has been, and how well this particular item of information is supported. It then defines how those judgements, together with the claim's validation status, combine into a label, what the label means, and what it does not mean.

It builds on the three evidence dimensions of the [Evidence & Provenance Standard](https://altss.com/knowledge-center/frameworks/evidence-and-provenance-standard): evidence origin, derivation status and validation status. It uses them as inputs and does not redefine them. Authority class and reliability grade, set here, are properties of the source. They are not evidence-origin values and not validation statuses.

The two-axis grading of sources and information is adapted from the NATO source-evaluation system, often called the Admiralty Code. The authority classes, the confidence labels, the decision rules and the decay classes described here are Altss-defined. They are Altss methodology, not an industry standard, and are labelled as such.

## 2. Scope

**In scope**

- Grading of sources that support factual claims about private-markets entities, relationships and events.

- Counting of independent originating sources, and the use of a claim's validation status, as inputs to confidence.

- The preference order used when a review resolves a conflict.

- Assignment, meaning and limits of confidence labels.

- Calibration of the labels and their decay over time.

**Out of scope**

- What evidence must be recorded for a claim, and the definitions of evidence origin, derivation status, validation status and independence. Those are set by the [Evidence & Provenance Standard](https://altss.com/knowledge-center/frameworks/evidence-and-provenance-standard), which this methodology takes as its input.

- Selection of sources for general statements in research and reference writing, which uses the T1/T2/T3 tiers of the [Research Citation Standard](https://altss.com/knowledge-center/frameworks/research-citation-standard). Those tiers rank sources for definitions and market facts; the authority classes here rank sources for claims about specific entities. The two are not interchangeable.

- Identity decisions. Confidence never merges two entities; see the [Entity Resolution Methodology](https://altss.com/knowledge-center/frameworks/entity-resolution-methodology).

- Assessments of an entity's quality, prospects or suitability as a counterparty.

## 3. Definitions

- **Source authority.** The standing of a class of source to state a given type of claim, decided by what the source is and its relation to the fact. Recorded as an authority class (5.2), a property of the source for a claim type. See source authority.

- **Source reliability.** A grade of how dependable a specific source has been for a given type of claim, based on its record (5.3). A property of the source: it is not an evidence-origin value and not a validation status. See [source reliability](https://altss.com/glossary/source-reliability).

- **Information credibility.** A grade of how well a specific item of information is supported (5.4): whether its claim is corroborated, consistent with what else is known, and plausible.

- **Evidence origin, derivation status, validation status.** The three evidence dimensions of the Evidence & Provenance Standard (sections 3.2, 6.1, 6.6 and 6.7). Evidence origin is recorded on each evidence item, derivation status on each value, and validation status, with its validation date, on each claim. This methodology reads them; it does not assign or redefine them.

- **Originating source.** The first source in an evidence chain with direct knowledge of the fact (Evidence & Provenance Standard, section 3.1).

- **Independent evidence items.** Evidence items that pass the independence test of the Evidence & Provenance Standard (6.8), repeated in 5.5.

- **Corroboration.** Support for a claim from at least two independent evidence items. A claim with corroboration and no unresolved contradiction has the validation status CORROBORATED. See [triangulation](https://altss.com/glossary/source-triangulation).

- **Confidence label.** The Altss-defined ordinal statement of the strength of the evidence base for one claim at one assessment date (5.7). It is not a validation status: validation records what checking a claim has passed; confidence records how strong its evidence base is, and uses validation status as one input. See [Source Confidence (Altss)](https://altss.com/glossary/source-confidence) and, for the general data term, [confidence score](https://altss.com/glossary/confidence-score).

- **Calibration.** The degree to which claims carrying a given label are later confirmed at rates consistent with the label's stated meaning. See confidence calibration.

- **Review horizon.** The period after a claim's last observation within which a time-varying claim can still be read as current without a recheck (Altss-defined).

- **Decay.** The reduction in support for reading a time-varying claim as current, as time passes since it was last observed. See [Evidence Decay](https://altss.com/glossary/evidence-decay), an Altss-defined concept.

## 4. Inputs and source types

The inputs are the provenance records required by the Evidence & Provenance Standard: for each claim, its evidence items with the evidence origin of each, each source's description on the axes of proximity, form and interest, the evidence chain to the originating source, the value's derivation status (OBSERVED, DERIVED or ESTIMATED), the claim's validation status with its validation date, and its other dates.

Source types are those defined in that standard: primary and secondary sources, copies and compilations, regulatory filings, public records, publications, statements on request, and self-reported and third-party sources. This methodology does not add source types or evidence-origin values. It grades sources, and records each grade on the source for a claim type.

## 5. Method

**5.1 Three judgements, kept apart.** Three judgements are recorded separately:

- **Authority class** of the source for this claim type (5.2), assigned by rule and recorded on the source.

- **Reliability grade** of the source for this claim type (5.3), assigned from its record and recorded on the source.

- **Credibility grade** of the item of information (5.4), assigned from the claim's validation status and its consistency with other evidence.

A single combined "source score" is not recorded in place of these three, because a highly reliable source can carry a poorly supported item (a reputable newspaper repeating an unconfirmed rumour), and a source of unknown reliability can carry an item that independent sources confirm. None of the three is written into the evidence dimensions of the Evidence & Provenance Standard. An evidence item's origin, a value's derivation status and a claim's validation status are recorded there, in their own fields, and no grade set here changes them.

**5.2 Authority classes (Altss-defined).** Authority is assessed relative to the type of claim, not in general.

| Class | Description | Authoritative for |
| --- | --- | --- |

| RECORD_KEEPER | Regulators, statutory registries and courts, for the filings and records they keep; public bodies, for the official records and publications they issue (minutes, decisions, registers, statutory reports). | What the record is required to state, as the record defines it, at the date it specifies; a public body's own decisions and acts. |

| SUBJECT | The subject's own statements: annual reports, published financial statements, announcements, its website, statements on request. | The subject's own decisions and acts (an appointment it made, a commitment it approved, a fund it launched). Not for claims about its own quality or performance relative to others. |

| DIRECT_ACCESS | A party other than the subject with its own access to the fact: a counterparty, a journalist with their own sourcing, independent research. | Facts the party observed or took part in. |

| COPY_OR_COMPILATION | Copies, compilations and summaries without independent access: directories, aggregators, syndicated copies, machine-generated summaries. | Nothing on its own. Useful to locate an originating source. |

| UNATTRIBUTED | Statements that cannot be traced to an identifiable publisher. | Nothing. A research lead only. |

Two qualifications apply. First, most filed content is reported by the filer under a legal duty; the registry or regulator that publishes it is authoritative that the filing says what it says, and the truth of the content rests on the filer. Second, a SUBJECT source is interested: it is the originating source for its own acts, but for statements that flatter it (performance, rankings, "largest", "leading"), its authority is that of a self-description.

**Authority class is not evidence origin.** Evidence origin (Evidence & Provenance Standard 6.6) classifies the artefact relative to the claim's subject. Authority class states what standing the source has for the type of claim. The two often coincide, but they are recorded separately because they diverge. A manager's release naming an investor is OSINT_SOURCED for the investor's commitment and comes from a DIRECT_ACCESS source, because the manager is party to the commitment; an aggregator's profile of the same investor is also OSINT_SOURCED and comes from a COPY_OR_COMPILATION source. A registry record is REGULATORY_PUBLIC_RECORD for every field, while free-text fields its filers write carry SUBJECT authority (section 10).

**5.3 Reliability grade.** Adapted from the NATO source-evaluation scale:

| Grade | Meaning in this methodology |
| --- | --- |

| A | Completely reliable: a record of accuracy for this claim type with no material errors found. |

| B | Usually reliable: minor errors found, corrected when identified. |

| C | Fairly reliable: errors found at a rate that warrants corroboration for material claims. |

| D | Not usually reliable: frequent errors or uncorrected errors. |

| E | Unreliable: shown to publish false or fabricated information. |

| F | Reliability cannot be judged: no adequate record. |

A reliability grade is a property of the source for a claim type and is recorded on the source. It is not an evidence-origin value and not a validation status, and it is not inherited from either: an evidence item with origin REGULATORY_PUBLIC_RECORD can come from a source graded C for a free-text field, and a CORROBORATED claim can rest on sources graded C. A source's default grade follows its authority class for the claim type, and is changed only with a recorded reason: errors found on recheck, corrections history, or a demonstrated change in the source's practices. Reputation, size or brand are not reasons. A source is graded per claim type: a registry can be A for registration status and C for a free-text business description its filers supply.

**5.4 Credibility grade.** Adapted from the same system, and read from the claim's validation status and the evidence:

| Grade | Meaning in this methodology |
| --- | --- |

| 1 | Confirmed: the claim has at least two independent originating sources (validation status CORROBORATED, or RESEARCH_VALIDATED with that count recorded) and is not contradicted. |

| 2 | Probably true: one originating source; consistent with other evidence held. |

| 3 | Possibly true: one originating source; no other evidence bears on it. |

| 4 | Doubtful: the claim's validation status is CONFLICTING, or the value is inconsistent with other evidence without a real conflict under the Evidence & Provenance Standard 7.1 (for example a figure implausible for the subject's size). |

| 5 | Improbable: the value is contradicted by evidence preferred in a review under section 7. |

| 6 | Cannot be judged. |

A credibility grade does not replace the validation status, and setting one never changes it.

**5.5 Independence test.** The test is defined in the Evidence & Provenance Standard (6.8) and applied here unchanged. Two evidence items are independent only if all of these hold:

- neither states the claim on the basis of the other;

- their chains do not reach the same originating source (one press release, one filing, one interview, one dataset);

- they are not copies, translations or syndications of one text;

- neither relies on a third source that the other also relies on for this claim.

Signals that items are not independent include identical wording or quotes, the same figures with the same rounding, publication within a short window after a single announcement, and attribution to the same release. Where independence cannot be established, the items are treated as one originating source. Different evidence-origin values do not establish independence.

**5.6 Count originating sources, not documents.** Corroboration is the number of independent originating sources, not the number of documents. Two articles quoting one press release are one source. Ten aggregator profiles copying one directory are one source. A subject's filing and the same subject's press release are one originating source for facts only the subject supplies; corroboration of such a fact needs a different party (a counterparty, a regulator acting on its own findings, the other side of the transaction). A claim has the validation status CORROBORATED only on two or more independent originating sources.

**5.7 Confidence labels (Altss-defined).** Each claim receives one label at each assessment. No label reuses a word from the evidence dimensions, and a label is always shown separately from the validation status it relied on.

| Label | Criteria |
| --- | --- |

| HIGH | All of: derivation status OBSERVED or DERIVED; credibility 1 or 2; no unresolved contrary evidence; for a claim read as current, last observed within its review horizon; and either (a) a pinpointed evidence item from a source with authority for this claim type (5.2), graded A or B, with the claim RESEARCH_VALIDATED against it, or (b) two or more independent originating sources, each graded A to C. |

| MODERATE | Any of: a claim that meets HIGH (a) except that it has not been RESEARCH_VALIDATED; a claim that meets HIGH except that it is read as current beyond its review horizon; one originating source graded B or C that is not authoritative for this claim type, with credibility 2 or 3; an ESTIMATED value whose method and inputs are recorded. |

| LOW | Any of: evidence only from COPY_OR_COMPILATION sources; a single originating source graded D; credibility 4; an ESTIMATED value with incomplete inputs or a wide range. |

| SUSPENDED | The claim's validation status is CONFLICTING. No strength is stated until a review resolves the conflict; the competing values remain visible with their evidence. |

| UNASSESSED | No assessment has been recorded. This is not the same as LOW. |

Claims supported only by UNATTRIBUTED sources, or with credibility 5 or reliability E, are not published as facts.

**5.8 Ceiling rules.** An ESTIMATED value is never HIGH, whatever its validation status: validation never converts an ESTIMATED value into an OBSERVED one (Evidence & Provenance Standard 6.7). A DERIVED value is never higher than its weakest input. A claim about a natural person's current role or contact details is never read as HIGH beyond its review horizon, whatever its original support.

**5.9 Review changes confidence only through evidence.** A review raises a label in three ways only: by adding evidence (a new independent originating source, a pinpointed primary source), by correcting an error in the provenance record, or by checking the claim against its evidence and confirming it, which sets the validation status RESEARCH_VALIDATED with its date (Evidence & Provenance Standard 6.7). A reviewer's opinion without a check against evidence does not raise a label. Where a reviewer departs from these rules, the departure is recorded as a Research Override with its reason and evidence, and is visible as such in the provenance record. An override changes the confidence label only. It never changes an evidence origin, a derivation status or a validation status.

**5.10 Confidence is per claim.** A record does not have one confidence, and it does not have one validation status. Where a summary across claims is needed, it is shown as the count of claims at each label, not as an average or a single record-level label.

**5.11 Commercial independence.** No commercial relationship, payment or request from the subject changes a source's grade, a claim's validation status or a claim's label. A subject's submission is a SUBJECT source with evidence origin DISCLOSED, graded like any other.

## 6. Classification rules

The label is assigned by this procedure, in order. The first step that applies decides.

- If the claim has no evidence item, it is not published as a fact (Evidence & Provenance Standard 5.2).

- If the claim's validation status is CONFLICTING, the label is SUSPENDED.

- If the only support is UNATTRIBUTED, or credibility is 5, or reliability is E, the claim is not published as a fact.

- If the derivation status is ESTIMATED, the label is MODERATE when method and inputs are complete and the range is stated, otherwise LOW.

- If the HIGH criteria in 5.7 are met, the label is HIGH, subject to the ceiling rules in 5.8.

- If the MODERATE criteria are met, the label is MODERATE.

- Otherwise the label is LOW.

Each assessment records the label, the assessment date, the step that decided it, the evidence items considered, and the validation status and validation date it relied on. [Evidence weight](https://altss.com/glossary/evidence-weighting) in this methodology is expressed by these rules, not by a numeric sum of source scores.

## 7. Conflict handling

**7.1 First, test whether the conflict is real.** The pre-checks of the Evidence & Provenance Standard (7.1: subject, definition, time, unit and scale, independence, compatibility) are applied first. A difference that a pre-check explains is not a conflict: it becomes two correctly qualified claims, and neither is CONFLICTING. See source conflict.

**7.2 A review applies the preference order.** While a real conflict is unresolved, the claim's validation status is CONFLICTING and its label is SUSPENDED. A review compares the competing values in this order, and the first rule that separates them decides which is preferred:

- **Authority for the claim type.** The value from the source whose authority class covers this claim type is preferred (a registry for registration status, a pension's own minutes for its commitment decision).

- **Proximity to the originating source.** A primary source is preferred to a secondary report of it.

- **Independent corroboration.** The value with more independent originating sources is preferred.

- **Specificity.** A pinpointed, dated statement is preferred to a vague or undated one.

- **Reliability grade.** The value from the more reliable source is preferred.

**7.3 If no rule separates them, the claim stays CONFLICTING.** Both values are shown with their evidence and the label stays SUSPENDED. The conflict is not resolved by averaging, by majority vote among non-independent sources, or by choosing the more recent publication when both state the same as-of date.

**7.4 A preference is not a deletion.** When a review records a preference, the claim takes the preferred value and becomes RESEARCH_VALIDATED as of the review date (Evidence & Provenance Standard 7.3), and it is assessed again under section 6. The evidence items for the other value stay attached to the claim with the reason they were not preferred; that value has credibility 5 and is not shown as current. If later evidence reverses the preference, the history shows both decisions.

**7.5 Corrections by the subject.** A subject's correction of its own earlier statement is a superseding claim version, not a conflict. A subject's statement that contradicts a RECORD_KEEPER record the subject did not file is a conflict, handled under 7.2.

## 8. Confidence and limitations

**What a confidence label means.** It states how strong the evidence base for one claim was, under these rules, on the assessment date.

**What a confidence label does not mean.**

- It is not a probability that the claim is true. It becomes a statement about observed accuracy only through calibration (section 9), and only with the sample, population and date of that calibration.

- It is not a validation status. A claim can be RESEARCH_VALIDATED and MODERATE (a reviewed estimate), or CORROBORATED and LOW (two independent sources, both graded D). Both are shown, separately.

- It is not a statement that the claim is current. A HIGH claim as of 2023 is a well-supported statement about 2023.

- It does not describe the record. Other claims on the same record can carry different labels.

- It is not an assessment of the entity: not its quality, importance, creditworthiness or suitability.

- It is not legal verification, identity verification for regulatory purposes, or an audit.

- For contact details, it describes a checking process, not a guaranteed delivery rate. Delivery also depends on sending practices, recipient policy and the current mailbox state.

- It is not an endorsement by any source that is cited.

**Limitations.** Reliability grades depend on rechecks; a source with few rechecks is graded F, not A. Independence can be hidden: two sources may share an originating source that neither names, so a CORROBORATED status is only as good as its recorded independence basis. RECORD_KEEPER records contain errors, typically transcription or classification errors made by filers. The labels are ordinal; the distance between HIGH and MODERATE is not defined as a number.

## 9. Temporal rules

**9.1 Assessment date.** Every label carries the date on which it was assessed. A label without a date is UNASSESSED. An assessment also records the validation status and validation date it relied on; a later change of validation status (a completed review, a new contradicting item) is a reason to reassess.

**9.2 Decay applies to the "current" reading, not to the dated claim.** A claim states a value with an as-of date or an effective period. That statement does not decay: "X was chief investment officer as of 1 March 2024" remains as well supported in 2026 as when it was observed. What decays is support for reading the claim as current. See [data freshness](https://altss.com/glossary/data-freshness).

**9.3 Volatility classes (Altss-defined).** Each attribute belongs to a volatility class, and each class has a review horizon recorded with the assessment.

| Class | Typical attributes | Treatment |
| --- | --- | --- |

| STATIC | founding date, legal form at a date, vintage year, a dated past event | No decay. |

| SLOW | legal name, domicile, registered address, stated strategy, legal entity hierarchy | Long review horizon. |

| FAST | a person's current role, contact details, fundraising status, current allocation targets, current mandates | Short review horizon. |

When a claim read as current passes its review horizon, its label for the current reading is lowered one step (HIGH to MODERATE, MODERATE to LOW), it is flagged for recheck, and its last-observed date is shown. The dated claim keeps its original label. Passing the horizon does not change the validation status or its date: a RESEARCH_VALIDATED status still says what was confirmed on its validation date.

**9.4 Calibration.** Labels are calibrated by audit:

- Draw a stratified random sample of claims at each label, and within each label by claim type.

- Recheck each sampled claim against fresh evidence from sources authoritative for its type, by a reviewer who did not make the original assessment. Each recheck is recorded as a review under the Evidence & Provenance Standard (6.7).

- Record the confirmation rate per stratum with a confidence interval (a Wilson score interval is suitable for proportions).

- The labels are meaningful only if confirmation rates are ordered (HIGH above MODERATE above LOW) with intervals that separate them. Where they do not separate for a claim type, the rules for that type are revised and the revision is versioned.

- Any published calibration result states the population, the sample size per stratum, the period and the date. No calibration figure is implied by this methodology.

## 10. Edge cases

- **Registries with filer-supplied text.** A registry entry's evidence origin is REGULATORY_PUBLIC_RECORD for every field retrieved from it. Its authority class is RECORD_KEEPER for registration facts; free-text fields the filer writes (a business description) carry the authority of a SUBJECT statement. The origin value does not change; the authority does.

- **Anonymous sources in reporting.** A news report citing unnamed people is graded on the publication's reliability for that claim type, and its credibility cannot be better than 3 without an independent evidence item (Evidence & Provenance Standard 6.8).

- **Circular reporting.** An article citing an aggregator that copied an earlier article is one originating source. If the chain loops back to a page derived from Altss's own published records, that evidence is excluded entirely.

- **Generated web content.** Pages produced by generative systems without attributable sourcing are COPY_OR_COMPILATION at best and usually UNATTRIBUTED.

- **The subject disputes an official record.** The record and the dispute are both kept; the claim is CONFLICTING until a review under 7.2 resolves it (7.5).

- **A reviewed claim with one source.** A RESEARCH_VALIDATED claim resting on one originating source has credibility 2 or 3, not 1. A review confirms the claim against its evidence; it does not add an independent source.

- **A reliable source in one domain, unknown in another.** A specialist publication graded B for fund-closing news is graded F for claims about personnel in a sector it does not cover, until it has a record there.

- **Corrections policies.** A source that publishes and corrects errors promptly can merit a higher grade than one that never issues corrections, because uncorrected errors are invisible.

- **Very old official records.** A dated record kept by a RECORD_KEEPER stays authoritative for the past event it records. It is not evidence of the present state.

## 11. Worked examples

**Example 1: an investor relationship confirmed by two parties.** A public pension plan's annual investment report lists a named fund among its private-equity commitments, with a commitment of $75 million. The fund manager's final-close announcement lists the plan among its investors.

|  | Plan's annual report | Manager's announcement |
| --- | --- | --- |

| Evidence origin, for the plan's claims | OFFICIAL_INSTITUTIONAL | OSINT_SOURCED (the manager is a counterparty, not the subject) |

| Authority class | RECORD_KEEPER, authoritative for the plan's own acts | DIRECT_ACCESS, party to the commitment |

| Reliability grade | A | B |

| States | the plan is an investor; commitment $75 million | the plan is an investor |

The claim "the plan is an investor in the fund" has two independent originating sources: different parties, neither statement derived from the other. Derivation status OBSERVED; validation status CORROBORATED; credibility 1. Label: **HIGH**. The claim "the commitment is $75 million" rests on one originating source, the plan. Its validation status is UNVERIFIED until a reviewer confirms it against the report's pinpoint; until then it meets HIGH (a) in every respect except the review, and is **MODERATE**. After the review it is RESEARCH_VALIDATED, dated, and **HIGH**. It is still not CORROBORATED. The board's earlier approval of the commitment, if recorded in public minutes, is a separate claim about a decision, dated to the meeting.

**Example 2: one release, many articles.** A manager announces a new hire. Four trade articles repeat the announcement with the same quote. Independence test (5.5): all derive from the release, so there is one originating source. The release is DISCLOSED, from a SUBJECT source (the appointment is the manager's own act), grade B; the articles are OSINT_SOURCED and, for this claim, COPY_OR_COMPILATION sources. Validation status UNVERIFIED; credibility 2. The dated claim "appointed effective [date]" is **MODERATE** until a reviewer checks it against the release and confirms it. It is then RESEARCH_VALIDATED, dated, and **HIGH**, because the subject is authoritative for its own appointments. Two years later the reading "currently holds the role" has passed its FAST-class review horizon: **MODERATE** for the current reading, flagged for recheck. The dated claim stays HIGH, and its validation status still describes the review on its date.

**Example 3: aggregator only.** A directory lists an organisation as a family office with a stated AUM. No originating source can be traced. Evidence origin OSINT_SOURCED; authority COPY_OR_COMPILATION; reliability F; validation status UNVERIFIED. Label: **LOW**. The listing is a lead for research and is not used as evidence of classification or AUM.

**Example 4: an apparent conflict that is not one.** A fund's Form D reports, in Item 13, the total amount sold as of its filing (REGULATORY_PUBLIC_RECORD). The manager later announces a final close at a larger figure (DISCLOSED). Pre-checks: the attributes differ (amount sold as of a filing date vs total commitments at final close), and the as-of dates differ. Two claims, no conflict; neither is CONFLICTING, and each is labelled on its own evidence.

**Example 5: calibration (hypothetical numbers, not Altss measurements).** An audit rechecks 200 HIGH claims and confirms 188 (94.0%, 95% Wilson interval 89.8% to 96.5%), 200 MODERATE claims confirming 154 (77.0%, 70.7% to 82.3%), and 100 LOW claims confirming 41 (41.0%, 31.9% to 50.8%). The labels are ordered and their intervals separate. Within HIGH, a stratum of 40 personnel claims confirms 30 (75.0%, 59.8% to 85.8%), which does not separate from MODERATE. The rule for personnel claims is revised (for example, a shorter review horizon) and the revision is versioned.

**Example 6: a real conflict.** A companies-registry record, retrieved from the registry's system, gives an investment manager's date of incorporation as 14 March 2011 (REGULATORY_PUBLIC_RECORD, RECORD_KEEPER, grade A). The claim was RESEARCH_VALIDATED in an earlier review and labelled HIGH. The manager's annual report, captured later, says it was incorporated in 2012 (DISCLOSED, SUBJECT). The pre-checks find the same entity, the same attribute and definition, and incompatible values. The claim becomes CONFLICTING and its label SUSPENDED. A review applies 7.2: the registry is authoritative for incorporation, so its value is preferred at rule 1. The claim is RESEARCH_VALIDATED as of the review date and is assessed again: **HIGH**. The annual-report value stays attached, not preferred, with credibility 5. The same case is worked through in the Evidence & Provenance Standard, Example 7.

## 12. External references

- **NATO source-evaluation system (Admiralty Code).** NATO intelligence doctrine (AJP-2.1) grades source reliability on a letter scale, A (completely reliable) to E (unreliable) plus F (cannot be judged), and information credibility on a number scale, 1 (confirmed) to 5 (improbable) plus 6 (cannot be judged), recorded separately. Irwin and Mandel (2019) review the system, argue that the verbal scales hide subjective judgement, and recommend numeric probability estimates and periodic re-evaluation. This methodology adapts the two scales and their separation. It differs in five ways, all Altss-defined: it adds authority classes relative to claim type; it defines grades by a recorded error history for a claim type; it reads credibility from the validation status defined in the Evidence & Provenance Standard, so that credibility 1 requires independent originating sources; it combines the grades and the validation status into confidence labels by explicit rules; and, rather than numeric probabilities, it keeps ordinal labels and tests them by calibration audits and dated reassessment (section 9). The grades are recorded on the source and the item of information, apart from the evidence origin, derivation status and validation status of the Evidence & Provenance Standard.

- **US Intelligence Community Directive 203 (Analytic Standards).** ICD 203 requires analysis to describe the quality and credibility of its underlying sources, data and methods, to express uncertainty, and to distinguish assumptions from information. The confidence labels here serve the first of those purposes: they describe the evidence base for a claim, not a likelihood of an event.

## 13. Version and change log

- **Version 1.2 (2026-10-01).** Independent review. Example 4 no longer states the Form D filing deadline (not needed for the example; the deadline is carried by the Form D concept page) and cites the form's Item 13. Summary phrased as a methodology.

- **Version 1.1 (2026-10-01).** Aligned with the three evidence dimensions of the Evidence & Provenance Standard version 1.1 (Altss methodology). Corroboration and confidence now use validation status: credibility 1 means at least two independent originating sources (CORROBORATED); a single-source claim reaches HIGH only when RESEARCH_VALIDATED; the conflict label CONFLICTED is replaced by SUSPENDED, assigned while the claim's validation status is CONFLICTING; a review that resolves a conflict sets RESEARCH_VALIDATED. The independence test is the standard's test (6.8), applied unchanged. Authority classes are renamed so that no class name repeats an evidence-origin value, a derivation status or the independence test: OFFICIAL to RECORD_KEEPER, INDEPENDENT to DIRECT_ACCESS, DERIVATIVE to COPY_OR_COMPILATION. Reliability and authority are stated to be properties of the source, not evidence-origin values or validation statuses. "Origin" of a chain is now "originating source".

- **Version 1.0 (2026-10-01).** First publication. Defines authority classes, reliability and credibility grades adapted from the NATO system, the independence test and origin counting, the label procedure with ceiling rules, conflict preference order, the meaning and limits of labels, volatility classes and decay, and calibration.

**Change control.** A change to the label criteria, the authority classes or the preference order is a major change and is versioned; claims assessed under an earlier version keep the version under which they were assessed until reassessed.

## Concepts used

- [Source Reliability](https://altss.com/glossary/source-reliability)

- Source Authority

- [Triangulation (Source Triangulation)](https://altss.com/glossary/source-triangulation)

- [Confidence Score](https://altss.com/glossary/confidence-score)

- Confidence Calibration

- [Evidence Weight](https://altss.com/glossary/evidence-weighting)

- Source Conflict

- [Source Confidence (Altss)](https://altss.com/glossary/source-confidence)

- [Evidence Decay](https://altss.com/glossary/evidence-decay)

- Research Override

## Sources

- [Improving information evaluation for intelligence production](https://doi.org/10.1080/02684527.2019.1569343). Daniel Irwin; David R. Mandel, Intelligence and National Security, Vol. 34(4), pp. 503-525. Status: Published (paywalled) (checked 2026-10-01). Vol. 34(4), pp. 503-525

- [Intelligence Community Directive 203: Analytic Standards](https://archive.dni.gov/files/documents/ICD/ICD-203.pdf). Office of the Director of National Intelligence, ODNI, Signed 2 January 2015; technical amendments incl. 21 January 2022. Status: In force (as amended) (checked 2026-10-01). Analytic tradecraft standards

- [Form D - Notice of Exempt Offering of Securities (form and instructions, paper version)](https://www.sec.gov/files/formd.pdf). U.S. Securities and Exchange Commission, SEC1972 (5/17); OMB No. 3235-0076, expires 2027-07-31. Status: in force (checked 2026-10-01). Item 13 (Offering and Sales Amounts)

## Related frameworks

- Standard · Version 1.2

### [Altss Evidence & Provenance Standard](https://altss.com/knowledge-center/frameworks/evidence-and-provenance-standard)

The Altss standard for recording evidence behind private-markets facts: claim-level provenance, evidence chains, three separate evidence dimensions (evidence origin, derivation status, validation status), distinct date types, null states such as unknown vs not disclosed, negative evidence, conflicts and claim citation.

- Methodology · Version 1.2

### [Altss Temporal Data & Freshness Methodology](https://altss.com/knowledge-center/frameworks/temporal-data-and-freshness-methodology)

The Altss methodology for time in private-markets data: valid time vs transaction time, point-in-time reconstruction, as-of reporting, freshness as an attribute of each claim rather than a page, decay, and why a content edit never implies that a fact was verified.

- Methodology · Version 1.2

### [Altss Entity Resolution Methodology](https://altss.com/knowledge-center/frameworks/entity-resolution-methodology)

The Altss methodology for deciding whether two records describe the same entity: identity, relation and role kept separate; organisations merged only on registry identifiers or a filing naming both as one; similarity creates relation candidates, never merges; no metrics summed across entity families.

- Standard · Version 1.2

### [Altss Private Markets Research Citation Standard](https://altss.com/knowledge-center/frameworks/research-citation-standard)

Altss's rules for citing sources in private-markets reference and research content: source tiers, no self-citation or competitor evidence for general claims, a fixed claim format with unit and as-of date, recency markers, and rules for AI-assisted drafting.

- Standard · Version 1.1.0

### [Altss Private Markets OSINT Verification Standard](https://altss.com/knowledge-center/frameworks/osint-framework-for-private-markets-intelligence)

The Altss standard for verifying publicly accessible private-markets information: Berkeley Protocol-aligned source, item and content verification, the three evidence dimensions with origin kept separate from collection method, source grading, ICD 203 language for judgments, and source-rights recording.

## Canonical URL

https://altss.com/knowledge-center/frameworks/source-reliability-and-confidence-methodology