This is how every Aegean brief is built.

Content type discipline.

Every sentence in every Aegean brief is classifiable as exactly one of three types. The types carry different evidentiary standards and serve different purposes for the reader. We do not blend them into a confident-sounding paragraph that hides the seam between evidence and inference. The reader can always tell which is which.

Source tier discipline.

Every cited source carries a tier. Tier 1 primary and official sources carry the highest weight and can stand alone; a high-impact claim needs two independent credible sources or one Tier 1 primary source. Tier and reliability are two things: every source also carries a reliability grade from A to F, refined claim by claim. Company materials are claim sources, not proof. State media is a claim source only and never counts toward verification. The source list is the curated registry the firm maintains: 626 entries across the four tiers. A source joins it only when a named Principal approves it, one source at a time; material from a source that is not on the list is labelled UNVERIFIED and never carries a fact in a finished deliverable.

Confidence calibration.

Every assessment carries an explicit confidence level and a one-line rationale. We surface what we don't know as plainly as what we do. Confidence is a function of source quality, corroboration depth, timeliness, and the plausibility of competing interpretations.

Risk velocity classification.

Every risk pathway and scenario driver in an Aegean brief is tagged with a velocity class. The same event can have IMMEDIATE consequences on one axis and STRUCTURAL consequences on another; the velocity tag makes the different planning horizons explicit so they reach the right decision owner with the right clock.

The anti-fabrication perimeter.

The most critical rule set in the operating standard. Aegean briefs do not invent, assume, estimate, or imply categories of information unless explicitly supported by a cited source. When the firm does not know, the brief says so, explains why, and adds the item to a Collection Plan. The gap is itself decision-relevant intelligence.

The seventeen-module collection model.

The cluster structure makes the architecture readable at a glance. Foundation governs source policy and analytical discipline at all times. Entity baseline establishes who the subject is. Domain exposure maps how that entity touches the world. Risk integration ties the exposures to outcomes. Consolidation and deliverables package the work for the audience. Commercial covers pipeline development. Each module within the standard documents inputs, outputs, scope boundary, sector adaptations, and hard constraints. Sector adaptation is built in: a module like Assets & Operations adapts metric vocabulary across extractive, manufacturing, technology, logistics, and financial subjects without changing its question architecture.

Heritage and influences.

The Aegean Standard is an adaptation, not an invention. The framework draws on structured analytic technique developed and published openly by the US intelligence community across more than two decades, then refined for the cadence and stakes of commercial decision support. The two traditions answer different questions. The IC literature was written to defuse cognitive bias inside national-security analysis on multi-year horizons; commercial briefs run on weeks, must reach a P&L-accountable audience, and must survive scrutiny from a board of directors. The methodology on this page is the bridge.

The forward-claim standard.

A forward claim is a dated, probability-weighted statement about an outcome that public sources will settle. A named Principal, one of the firm's two founders, locks it, and a locked claim is not edited: outcomes and corrections are new records filed beside it. Every brief on the register carries the forward claims it makes; the Scorecard keeps the running record. This chapter is the law that governs what qualifies, how a claim is worded, how it resolves, how a mistake in it is corrected, and when it is void.

What qualifies

A statement in a brief becomes a locked claim only when it passes all five tests. First, the brief states a probability for the outcome, or attaches a confidence band to the judgment that carries it. Second, the outcome is checkable from public sources: a filing, a published data series, a dated official record, or reporting that meets the source list in Chapter 02. Third, the resolution window is stated in the brief or is unambiguous from it. Fourth, the claim has a written resolution rule, approved before it locks. Fifth, a named Principal approves the lock. A statement that fails any test stays on file as a proposed claim, is never shown publicly, and is never scored.

The band values were fixed on 8 September 2026, when the standard was first applied to the register. They are a scoring convention for the bands as Chapter 03 defines them, not a claim about how often each band has held; the private calibration record on the Scorecard shows that separately, and the firm's calibration priors are measurements kept apart from this fixed convention. A change to them applies only to claims locked after the change. The mapping applies only to the confidence a brief attaches to an assessment or a key judgment. The confidence a brief attaches to how well a scenario was built describes the quality of the build, not the likelihood of its outcome, and is never turned into a probability.

Wording rules

1. A claim is entered in the brief's own words, lightly trimmed. Trimming removes lead-ins and citation markers and never changes the object, the direction, the quantity, or the window.
2. A claim names one outcome. A sentence carrying two expectations is entered as two claims or not at all.
3. A claim carries the brief's publication date as its made-on date and the date it entered the ledger as its locked-on date, so a retrospective entry is never presented as contemporaneous registration. A revised brief keeps the original date unless the claim first appeared in the revision.
4. Every public claim carries a record class and a claim family. BRIER-native claims are locked no later than publication and alone enter the headline score. Legacy prospective entries and legacy retrospective entries remain visible but score separately. Related claims share a family so raw claim count is never represented as independent sample size.
5. The locked fields, the statement, the probability, the made-on date and the window, are never edited. An outcome is added as a new, approved entry and a correction as a new record filed beside the claim, so the locked fields stay exactly as they were. The one exception is a revocation: a named Principal may revoke a claim, always with a stated reason, and the firm ledger records who revoked it, when, why and exactly what changed, before and after, with the earlier values kept in the claim's history.

Resolution rules

1. Every claim carries a written resolution rule, approved by a named Principal before the claim locks. The rule names the public source; the reading or event that settles the claim; how long after the window the source may publish and still count; and that the claim stays open if the source is silent.
2. A claim resolves only against a dated public source. A named Principal resolves it, recording with the claim a link to the source, one line saying what the source shows, and its date. Collectors and the AI may suggest an outcome; they never resolve a claim.
3. Resolution on inference is forbidden. If the only route to an outcome is reasoning from adjacent facts, the claim stays open.
4. A claim whose window has passed without a public source is open: it is shown as open, with its window and no outcome, and adds nothing to the score. The move to open in the record is approved by a named Principal.
5. Where the outcome is contested between sources of equal standing, the claim stays open until the record settles.
6. A claim resolves before the last day of its window only when its source already settles it, for example when the meeting it asks about has taken place.

Corrections and the void condition

A forecast that turned out wrong is not a mistake. It is scored, and it is never voided because it missed. A mistake in what was recorded is corrected by a correction record, a new record filed beside the claim that says what was wrong, what changed and why, approved by a named Principal and listed in Chapter 10. Under the founders' ruling of 23 September 2026 there are two cases. An incidental error, such as a date misprinted inside otherwise correct locked text, where the forecast and its resolution are not in dispute, gets a correction record alone: nothing in the claim is edited, and the claim stays resolved and keeps counting in the score. A claim that is wrong in its substance, in its probability, its premise or its outcome, is void. A void claim carries a note saying why and a correction record, is excluded from the score, and is counted separately so the reader can see how often it happened; a claim voided after it resolved leaves the score, which is recomputed without it.

A claim that cannot be settled as worded, because the measure it names is not published, the object it names has ceased to exist or changed beyond recognition, or the wording admits both outcomes, stays open: it is shown with its window and no outcome, and adds nothing to the score.

The scoring rule.

The headline public score is the Brier score over resolved BRIER-native probability claims only.

For each eligible resolved claim, take the square of the difference between the probability the firm locked and the outcome, one when the claim came true and zero when it did not, and average across all eligible resolved claims. Lower is better. A record that assigns one half to everything scores 0.25; a perfect record scores zero. Proposed, locked and open claims add nothing until they resolve. Void claims are excluded and counted separately, a claim voided after it resolved included. Legacy prospective and legacy retrospective claims remain public on their own tracks and are never blended into the headline score. Quantity forecasts, which carry a range instead of a probability, are scored by whether the outcome fell inside the range and how far it landed from the forecast, and are never mixed into the headline. The score is published to four decimals with its count and the date of the last eligible resolution, and is recomputed whenever an eligible claim resolves, a resolved claim is voided, or a revocation changes a scored claim. Each desk with resolutions carries a Brier score of its own. The reliability diagram plots the same BRIER-native claims in ten bins by probability, unsmoothed; below ten resolved claims it says it is not yet a calibration record. One generated sentence on the Scorecard gives the count of settled claims and says plainly that the score is early while it rests on few of them. Claim families are disclosed, so the count is never read as that many independent tests. Every figure is computed from the record; none is typed by hand. The firm's private benchmark and its public legacy tracks are shown separately and are never combined with the headline, and the private benchmark is never cited as evidence of calibration.

Every generated ledger version is copied to a content-addressed snapshot and published with a SHA-256 fingerprint, claim-level locked-field fingerprints, selection counts and dependency families in Record integrity.

The correction record.

Every correction to published or gated work is recorded here with its date, what was wrong, what changed and why. A correction found before publication is recorded all the same, because the discipline that caught it is part of the record. A correction to a claim on the record is a correction record: a new record filed beside the claim, approved by a named Principal before it is filed, which leaves the claim's locked fields as they were (Chapter 08). The correction records filed in the record follow the entries below, each with the claim it corrects.

Standards for objects.

The Intelligence Portal registers six kinds of object: a report, a Rapid Intel Brief, a claim, a series, a dataset and a desk. A Rapid Intel Brief is a compact assessment for a decision that cannot wait for a full report, and it names its evidence, its confidence, its decision window and any forward claim at publication. Each has exactly one page and exactly one stable identifier, and every page is built to the rules in this chapter. The rules exist so that consistency across the Portal is a published law rather than a design habit, and so that a reader who has learned to read one object can read all of them.

The header band

Every object page opens with the same band, in this order: the object's type; its identifier; the desk it belongs to, or "all desks" where it is filed to the firm as a whole; the date it was created; the date it was last updated; its source and the source's tier, where a source applies; and the standard it was built to, which is this page. The band carries the source and tier on a series, on a dataset, and on a claim once a public source has settled it. Where a slot does not apply it is omitted, never left blank. The dates are the dates the source file records, not the dates a page was generated.

Connected objects

Every object page ends with a block that lists the objects connected to it, by type. A report or a Rapid Intel Brief lists its claims, its desk, and any series a claim resolves against. A claim lists the report or Rapid Intel Brief it came from, its desk, its series where one is named, and the ledger package. A series lists its dataset, its desk, the claims that reference it, and the Dashboard tab it appears on. A dataset lists its series and the desks they belong to. A desk lists its reports, its Rapid Intel Briefs, its claims by status, its series and its datasets. A relation is stored once, in the registry, and rendered in both directions, so that no link can exist on one page and not on the other.

The rules

1. An empty list states what it is waiting for. A desk with no locked claim says so and names the next publication; a series no claim resolves against says so; a package not yet published says how many days of history exist against the seven it needs. No list on the Portal is ever a blank.
2. A chart is followed by its data. Every generated chart is an inline image with a title and a description that state the series, the last value and the direction of change, and it is followed by a table of the last thirty observations, so that no information on the Portal exists only inside a graphic.
3. Every figure on a page is read from a file the collectors or the record tooling wrote. No number on the Portal is invented, illustrative or seeded. A panel whose feed is down says so and shows its last good time.
4. Nothing retired is searchable. A page that has been retired behind a redirect keeps its file on disk and is excluded from the search index, the sitemap and the registry, so that nothing the firm has withdrawn can be found by search.
5. Generated regions are marked and idempotent. Every block a generator writes sits between named markers; hand-written copy sits outside them and is never overwritten; running any generator twice produces the same page.
6. The ledger is edited in one place. A claim page presents the ledger; it never carries a second copy of it and never computes a score of its own.