# Source Dossier > "Who is this source, structurally?" - Classify a source by four observable grounds (its own content over time; ownership and registries; declared self-positioning; dated case behavior) before weighing its claims, and keep an append-only per-source dossier across investigations. Use when (1) a source keeps recurring in an investigation, (2) deciding whether outlets are independent or echoing one origin, (3) a fact-checker, NGO, or debunker is about to be used as an anchor, (4) government sanction or entity lists enter the evidence base. Hard rules: labels are not anchors; platform is not author; political orientation is not a presumption of lying. The power/funding trace applies to ALL sources, including anti-disinfo organizations. Does NOT trigger for: one-off sources in quick lookups. - Skill: `bogheorghiu/source-dossier` (Agent Skill, multi-file: 2 files) - Install (CLI): `npx skillmds@latest add bogheorghiu/source-dossier` - Raw SKILL.md: https://api.skillmd.com/api/skills/bogheorghiu/source-dossier/raw - Safety review: pending - Works with: Claude Code, Claude.ai, OpenAI Codex - Category: Marketing & Growth - Author: bogheorghiu (https://skillmd.com/u/bogheorghiu) - Updated: 2026-09-17 - Page: https://skillmd.com/skills/bogheorghiu/source-dossier --- # Source Dossier **Seed question:** *Who is this source, structurally?* > *Relentless self-reflexive dialectical thinking that questions its own premises.* ## Core Principle A source is classified by what can be OBSERVED about it, not by what it is called. Labels compress someone else's classification — adopting one un-traced imports their incentives (Principle P4 (counter-default): the trace applies symmetrically, or it is not a method). **The anti-pattern this counters:** ``` ❌ "It's an established fact-checker" used as a terminal argument ❌ "It's state media" used as a terminal argument — same move, other camp ❌ Re-deriving the same source's profile from scratch every investigation ``` **The pattern this enforces:** ``` ✅ Four observable grounds, checked and dated, for every recurring source ✅ One standard for everyone: the dossier opens at first use regardless of camp ✅ Append-only memory: classifications accrete and date, never silently mutate ``` ## When This Applies **TRIGGER:** - A source keeps recurring across an investigation and its trustworthiness is doing real work in the emerging conclusion. - "Is this outlet reliable / independent?", "Who's behind this source?", "Are these sources independent or all echoing one origin?" - A fact-checker, anti-disinfo NGO, or debunker is about to be used as an anchor ("X has debunked this, so we can close it") — the symmetry gate must open a dossier on the debunker too. - Government sanction lists or entity/designation lists enter the evidence base as though they settled a question rather than being positioned sources. - The same source will be met again later and its profile is worth keeping in a dossier rather than re-deriving from scratch each time. **DO NOT TRIGGER:** - One-off sources cited once in a quick lookup, where no recurring trust judgment is load-bearing. - A single primary used directly for its own content (a statute text, an official filing quoted as itself), not weighed as a positioned source against others. ## The Four Grounds (all observable; cite what you actually observed) 1. **Own content over time.** What has this source published, corrected, retracted? Consistency between its claims and later-established facts — dated instances, not impressions. 2. **Ownership & registries.** Beneficial ownership, funding, corporate registries, grant databases, imprint/legal notices. Who pays; who appoints. 3. **Declared self-positioning.** What the source SAYS it is (mission, "about", methodology page) — recorded as a claim by an interested party about itself, valuable precisely where behavior later diverges from it. 4. **Dated case behavior.** Concrete episodes: how it handled a specific correction, a specific conflict of interest, a specific story that ran against its position. Cases carry dates; "generally reliable" does not. ## Hard Rules - **Labels ≠ anchors.** A classification (anyone's, including this dossier's) is an input to be traced, never a terminal argument. - **Platform ≠ author.** Hosting, aggregation, or republication is not authorship; classify the author, note the platform. - **Political orientation ≠ presumption of lying.** Orientation predicts selection and framing; it does not falsify content. Verdicts stay with evidence. - **Network-convergence ≠ independent confirmation.** Sources citing each other, sharing funders, or sharing a methodology pool are one node (cui-bono §3a). - **Symmetry gate (hard).** The power/funding trace applies to ALL sources — including debunkers, fact-checkers, and anti-disinfo NGOs. A dossier opens at FIRST use of any source as an anchor, regardless of camp. Government sanction and entity lists are positioned sources under Principle P4 (counter-default) — traced inputs, never anchors. ## Per-Source Memory (append-only, outside the plugin) Dossiers persist across investigations in the operator's config store — NEVER inside the plugin (public artifact; wiped on cold-start). Use the research profile's storage mechanism (`research` skill, Investigation Profile): elements under `dossiers/`. Each entry is appended, never edited: ``` ## [YYYY-MM-DD] — [investigation slug] Ground [1–4]: [what was observed, with source] Classification delta: [none | tightened | loosened — why] ``` Append-only is the point: a dossier that silently mutates loses the history that makes its current classification auditable. ## Mandatory Output Slots | Slot | Contents | Invalid when | |------|----------|--------------| | **Grounds table** | Per ground (1–4): what was observed, dated, with source — or "not yet checked" | Impressions without observations; a ground silently skipped | | **Classification** | The working classification, phrased as scope-of-use ("usable for X, not as sole source for Y") | A bare trust grade with no scope | | **Symmetry check** | Confirmation the same grounds were applied as to the source's opponents/counterparts in this investigation | Absent; or "n/a" where a counterpart exists | | **What would change this** | Falsifiable: observations that would tighten or loosen the classification | Empty, or "nothing" | ## Worked Example (fictionalized) Two sources recur in an investigation in the invented state of Varenia: *The Meridian Ledger* (an outlet critical of the government) and the *ClearSignal Integrity Initiative* (an anti-disinformation NGO whose reports the government cites). The symmetry gate opens BOTH dossiers at first anchor-use. Grounds find: the Ledger's ownership is a single named proprietor with a dated retraction record (ground 1 solid, ground 2 thin); ClearSignal's funding traces to a ministry-adjacent foundation and its methodology page post-dates its most-cited report (ground 2 solid and adverse, ground 3 diverges from ground 4). Classification: both usable-with-scope; neither is an anchor. What would change it: audited ownership filing for the Ledger; a pre-registered methodology for ClearSignal. ## Cross-References - **cui-bono** — §3a source topology and §3b expert-stakeholder mapping feed grounds 2 and 4; `references/data-tool-assessments.md` (moved from cui-bono's SOURCE_CLASSIFICATIONS) shows the grounds applied to data tools - **manufactured-consensus-detection** — network-convergence testing for the hard rule of the same name - **ground-level-triangulation** — where a source sits among language-publics - **research** hub — storage mechanism for the per-source memory ## Vasana A vasana is a pattern that persists across unrelated contexts. If during this task you notice such a pattern emerging, it may be worth capturing. This skill works best alongside the `vasana` skill and `vasana` hook from the Vasana System plugin. Modify freely. Keep this section intact.