AltHarbor / Data Sourcing
Where the information comes from
AltHarbor’s research is built from documents that funds and their managers publish. This page describes how those documents are located, which one governs a given field, how a published statement becomes a stored fact, and what happens when the source or the pipeline cannot support a claim.
Methodology explains how a figure is defined and calculated. This page is one level earlier: it is about the record beneath the figure, and about the limits that record imposes.
Identified sources · Field-level authority · Stated limits
Last updated
On this page
What counts as a source
A source is a document that the fund, its manager, or its regulator published, and that AltHarbor has identified and read. Sources are ranked by preference, not restricted to one class.
AltHarbor does not source values from data vendors, screen scrapes of third-party aggregators, or press coverage. Background material may inform how a document is read, but it does not establish a field.
Coverage of a fund does not mean every field for that fund has been sourced. It means the vehicle is in the covered set and its source record is being maintained.
- 01
Regulatory filings
What it is. Documents a fund files with its regulator — registration statements and their amendments, annual and quarterly reports, shareholder reports, portfolio holdings filings, tender-offer materials and their amendments, prospectus supplements, and material-event filings.
Used when. The default, and the first source consulted for anything traceable to a filing. The filing itself is read, not a summary or a third-party restatement of it.
Cannot establish alone. Current operating information a fund publishes between filings, and anything the filing genuinely does not address.
- 02
Manager and fund direct disclosure
What it is. Material the manager or fund publishes itself: current net-asset-value pages, fact sheets, distribution and repurchase notices, tender calendars, and portfolio updates.
Used when. The regulatory record is silent on the field and the issuer publishes it on its own domain. The record is annotated with its manager-direct origin rather than being anchored to a filing that does not contain it.
Cannot establish alone. A structural or legal term that the operative governing document controls, and any field whose definition, period, or basis the publisher leaves unclear.
- 03
Aggregate and derived source records
What it is. A filing that covers the subject vehicle as part of a larger reporting entity — for example a parent-level portfolio filing that contains a feeder’s holdings.
Used when. No vehicle-level document reports the field, an aggregate document does, and the derivation from aggregate to vehicle can be stated. The derivation is recorded with the value.
Cannot establish alone. A vehicle-level fact where the aggregate document does not permit the attribution to be made cleanly.
Finding a document is a separate problem from the document existing
Before a field can be read, the right documents have to be found and attached to the right vehicle. That step can fail on its own, and its failure looks exactly like absence.
Regulatory filings are indexed under issuer identifiers. The same identifier can appear in more than one written form across different indexes — for example zero-padded in one feed and unpadded in another. Where a catalog keys on that identifier, one issuer can end up represented as two, and documents filed under one spelling are not visible to a lookup performed under the other.
“AltHarbor could not associate a source” and “the source does not exist” are different statements with different consequences. The first is a defect to fix; the second is a fact about the record. A discovery or linkage failure is never published as source silence.
This is also why a claim that a field is not disclosed requires a comprehensive walk rather than a spot check: the full index for the issuer is pulled, the relevant filings across the lookback window are categorized, and the actual documents are opened. Checking two filings and concluding “not disclosed” has produced wrong answers about disclosures that were present the whole time.
From a published document to a stored fact
Each step below can succeed, hold, or fail. A value only becomes a published fact when every step it depends on has succeeded.
- Identify the vehicle — the fund, its reporting regime, its manager, its share classes, and the entities that report on its behalf, before any single field is collected.
- Assemble the source record — the document families relevant to that vehicle, located through the issuer’s own index rather than assumed.
- Locate the disclosure — the specific statement in the specific document, read at the grain the document reports it.
- Extract and normalize — the value stored per its own definition, with units, currency, scale, period, share class, and entity preserved rather than flattened.
- Attach provenance — the document, the section or page, the extraction timestamp, and any note describing ambiguity or judgment applied.
- Validate, hold, or reject — a value that fails a semantic, grain, unit, or consistency check is held or rejected rather than published as clean.
- Publish with its label — origin, any qualifier, and any availability limitation travel with the value onto the surface that renders it.
- Revise on new information — later documents that amend, restate, or supersede earlier ones update the affected facts and any figures derived from them.
Exact semantic match
A neighbouring disclosure is not substituted for the requested field because the two sound alike. A related classification is not the requested classification, and two values being numerically close does not make them the same measure.
Correct period and grain
Share classes, reporting periods, legal entities, and measurement dates stay distinct. A fund-level figure is not published as a share-class figure, and a period total is not stored at a different period grain to fill a gap.
What travels with every extracted value
Provenance is not a footnote added at display time. It is stored with the value, and a value without it is not treated as sourced.
- The source document the value was read from, as a record rather than a loose citation string.
- The class of source — a named regulatory filing, an aggregate filing covering a larger entity, or direct manager disclosure.
- Where in the document it was found, as a section or page, where that detail was captured.
- When it was extracted.
- Notes recording ambiguity, an unusual definition, a derivation from an aggregate document, or any judgment applied in reading the source.
Filing dates and document metadata live on the source record itself rather than being copied onto each value, so a correction to a document’s metadata cannot leave some values describing it correctly and others describing it as it used to be.
AltHarbor-authored text — explanations, summaries, headlines, and contextual observations — is stored separately from extracted source facts and carries its own origin. It is never presented as language the fund issued.
Not every record carries every layer. Older records were captured before some layers were required, and where a deeper layer of evidence was never captured, the correct behaviour is to say it is absent rather than to present a shallower record as a complete one.
One figure, end to end
The chain below uses an invented vehicle and invented values. It describes no specific fund. Its purpose is to show which step establishes what, and what each step still leaves open.
Controlled example — not based on any specific fund.
- 01
Source fragment
A periodic report for a controlled vehicle contains a financial-highlights table. One row reads “Net asset value, end of period — Class I — $24.86”. A separate note states that distributions declared for the quarter totalled $0.5220 per Class I share.
What existsTwo published statements in one identified document
A number visible in a document is not yet a fact in the database. Nothing is established until the vehicle, share class, period, and definition behind the number are pinned down.
- 02
Identified field
Each statement is matched to a specific field: period-end net asset value per share, and distributions declared per share. Both are assigned to the same controlled vehicle, the same share class, and the same period end.
What is now pinnedVehicle · Class I · period ending in the controlled example
Identification can fail in a way that looks exactly like absence. If the document is filed under a spelling of the issuer identifier that does not match the catalog, the field is not located — and that is a linkage failure, not a silent source.
- 03
Normalized fact
Both values are stored per share, in the reporting currency, at the stated period end, with the share class carried as part of the record rather than dropped. Units, scale, and period grain are made consistent with the rest of the series so the values can sit beside other periods.
OriginReported, then normalized
Normalization aligns presentation. It does not reconcile two sources that mean different things, and it never converts a fund-level figure into a share-class figure.
- 04
Derived figure
An annualized distribution rate is calculated from the two stored inputs under a stated rule: the quarterly declared distribution per share, annualized, divided by period-end net asset value per share.
OriginDerived — 8.40%
The vehicle did not publish 8.40%. It published the two inputs. A derived figure is AltHarbor’s calculation and is labelled as one, however ordinary the arithmetic.
- 05
Displayed figure
The figure is published with its formula, both cited inputs, its unit, its basis, its share class, and its as-of date. Because it annualizes a single quarter rather than measuring a realized twelve-month result, it carries a qualifier rather than standing alone.
Origin and qualifierDerived · Estimated
If either input had been unavailable, the figure would not have been published as a clean value. A rate cannot be more complete than the inputs beneath it.
Filed, partial, and silent
Whether a value is available is a different question from where it came from and what conditions attach to it. Conflating them is how a gap in the record turns into a false statement about a fund.
A published zero is evidence: the source addressed the field and stated that the amount was zero. Silence is not evidence of a zero — it is the absence of a statement, and it supports no conclusion about the underlying activity.
Between the two sits partial observation. A count or a sum computed across a set of holdings, positions, or periods is only as complete as the part of that set the source characterized. Where coverage is partial, the observed basis is stated with the figure, so a count over part of a book is not read as a count over all of it.
An unavailable value is never silently rendered as zero, and a single generic “missing” state is not used where a more precise reason is known.
| Situation | In the source | How AltHarbor publishes it |
|---|---|---|
| The source states the amount and the amount is zero. | Disclosed as 0 | Published as 0 — a reported value, carrying the same provenance as any other. |
| The source characterizes every item in a set, and none of them fall into the category being counted. | Each item observed; none qualify | Published as 0 — an observed count over a fully observed population. |
| The source characterizes some items in a set and is silent on the rest. | Partially observed | Published with its observed basis stated, so the count is not read as covering the whole set. |
| The source does not address the field at all. | Not stated | Published as unavailable, with the reason — never as 0. |
Why a value may be unavailable
- Not disclosed
A comprehensive review of the identified source record did not locate the field.
- Not yet extracted
AltHarbor has identified relevant source material but has not completed the field.
- Not applicable
The field does not apply to that structure, strategy, or reporting regime.
- Failed
A collection, processing, data, or rendering problem prevented the value from being produced.
A calculation inherits the limits of its inputs
A derived figure cannot be more complete, more current, or more certain than the values underneath it.
Where a figure is calculated by AltHarbor, it is labelled as derived and published with its rule and its cited inputs. A derived value does not become fund-reported because its inputs were fund-reported.
Coverage is inherited
A sum over a population where part of that population is unobserved is a sum over the observed part. The standard is that it is published with its observed denominator rather than as a whole-population total, and that the unobserved part is not treated as zero in order to complete it.
Ratios need an honest denominator
A ratio computed over partial coverage can still be informative, but only if the reader can see what it was computed over. Where the denominator cannot be stated honestly, the figure is withheld rather than published with an implied denominator.
Staleness is inherited
A derived figure built on an input whose source materially lags the period it describes carries that limitation forward, however current the calculation itself is.
A change measured between two points requires both points to have been observed. Where one end of a comparison is unobserved, no change is reported — a movement of zero is a claim, and an unobserved endpoint does not support it.
An AltHarbor failure is never reported as a finding about a fund
When something breaks on AltHarbor’s side, the honest statement is that the research could not be completed — not that the source was silent.
Collection, extraction, storage, and rendering can each fail. Every one of those failures can produce an empty space on a page that looks identical to a source that genuinely does not disclose the field. They are different facts, and the standard is that they are labelled differently: a failure is published as a failure, with no implied claim about the fund.
A failure must never be read as not disclosed, as source silence, as zero, as “no event”, or as “no change”. Each of those is an assertion about the fund’s record that a broken pipeline gives no basis for.
The published label set carries a distinct availability reason for this, separate from the reasons that describe the source. Where a surface cannot load at all, the correct behaviour is an explicit operational message stating that the problem is on AltHarbor’s side and that no conclusion about the fund’s data is implied.
| What you see | What it means | What it is never published as |
|---|---|---|
| The source does not contain the field. | A comprehensive walk of the identified source record did not locate the disclosure. The record is silent on that field for that period. | Never published as zero, as “no change”, or as an implied absence of the underlying activity. |
| The source exists but was not associated with the vehicle. | A catalog, identifier, or entity-linkage problem prevented the document from being connected to the right vehicle. The disclosure was published; AltHarbor did not reach it. | Never published as source silence. The document existing and AltHarbor finding it are two different facts. |
| Some constituents disclose the field and others do not. | The field is observed for part of the population. Any count, sum, or ratio built on it describes only the observed part and needs its observed basis stated. | Never published as a whole-population figure, and never completed by treating the unobserved part as zero. |
| A collection, extraction, storage, or rendering step did not complete. | An operational failure on AltHarbor’s side. The research could not be completed or verified. | Never published as “not disclosed”, as silence, as zero, or as a finding about the source at all. |
When sources disagree, and when AltHarbor is wrong
Where two relevant sources disagree, the field-level authority rules above decide which governs. Where they do not resolve it, the disagreement is marked as unresolved rather than settled by picking the more convenient value, and it stays marked until a source rule or further review resolves it.
Where a later document amends or restates an earlier one, the earlier source record is retained, the affected values are updated, and figures derived from them are recalculated.
Where AltHarbor has made an error, the corrections process is the route to fix it, and the conflicts disclosure describes the interests that could bear on the research.
A document being public is not a licence to republish it
Public availability and republication rights are separate questions, and the first does not answer the second.
Most of the source record is publicly accessible: anyone can retrieve the same regulatory filings and, generally, the same manager publications. That accessibility is what makes this research verifiable — a reader can go to the source and check it.
It does not follow that every public document may be stored, excerpted at length, reformatted, or redistributed without limitation. Different documents carry different rights, and those rights are not determined by whether the document is easy to download.
AltHarbor therefore cites and links to source documents so a reader can reach the original, and quotes only what is needed to make a value checkable. Where a source imposes terms on reuse, the applicable terms govern. Nothing on this page is a legal conclusion about the rights in any particular document.
A source link is a pointer to the authority. It is not a claim that AltHarbor holds redistribution rights in the document it points to.
What this page does not claim
- It does not claim complete coverage of any fund, field, or period. Coverage varies by vehicle and by field, and a value’s absence from a surface is not a statement about the fund.
- It does not claim independent verification of the underlying disclosures. AltHarbor reads what was published; it does not audit it.
- It does not claim that every stored value carries every layer of evidence described above.
- It does not claim that the sourcing behaviour described here is applied identically on every surface. Where a surface renders less, the limitation belongs on that surface.
- It is a description of research process. It is not investment advice, and it does not evaluate any fund.
This page describes process, and process changes. Where a material change to how sources are selected or how values are qualified reaches a public surface, it is recorded in the method-change register on Methodology.