```text ================================================================================ MEDIA QUALITY GRADER v12.0 — HARDENED SYSTEMIC INTEGRATION ================================================================================ VERSION: 12.5 RELEASED: Saturday, August 8, 2026 STATUS: PRODUCTION RELEASE SUPERSEDES: MQG v12.4 (mqgv12_4.txt), MQG v12.3 (mqgv12_3.txt), MQG v12.2 (mqgv12_2.txt), MQG v12.1 (mqgv12_1.txt), MQG v12.0 (mqgv12.txt), MQG v11.0 (mqg v11.txt), MQG v10.0 (MQG_v10_0_SELF_CONTAINED.txt) ================================================================================ WHY v12.1 — BLOCK 5 GRADE DISCLOSURE REQUIREMENT ================================================================================ v12.0 successfully restored all production boundaries, granular scoring brackets, and safety gates that v11.0 had inadvertently stripped, while cementing the seven-dimension ideological overlay (D7). See v12.0 rationale for full context. v12.1 makes a single targeted amendment: Block 5 (Reader Takeaway) must now open with "Grade: [Letter]" as its first line. v12.0's Block 5 spec specified "No jargon or scores" — a conflict with analyst operational directive requiring the letter grade to appear in the Reader Takeaway for reader-facing usability. Resolution: the prohibition is narrowed to numerical scores and jargon only; the letter grade is a plain-language signal, not jargon, and is now required. All v12.0 execution logic, dimension definitions, weights, floors, ceilings, verification protocols, and block schemas carry forward without modification. ================================================================================ WHY v12.2 — REDEFINITION-AS-DEFLECTION FINDING TYPE ================================================================================ v12.2 adds one new named, repeatable finding type: REDEFINITION-AS-DEFLECTION (RAD). Diagnosed during analyst review of a June 8-genre "word origin/explainer" piece that used a technically accurate point (a term is contested or imprecise) to discharge a substantive accountability or harm claim without engaging that claim's substance — applying the "imprecision" critique to only one side's usage, and closing on an unrebutted causal or exculpatory claim from the other side. v12.1's existing straw-man (0.3) and both-lenses (0.4) checks did not catch this pattern because no quote was distorted and both lenses were technically present — the failure was architectural (sequencing, asymmetric naming, who gets the last word), not a sourcing or quotation defect. Resolution: new Block 0 field 0.3A (REDEFINITION-AS-DEFLECTION DETECTION), new CEILING 10 (Section 4), and mandatory Block 4B / Block 5 flagging when detected. This is a cross-cutting pattern affecting D3 (Framing), D5 (Context Completeness), and D7 (Ideological Anatomy) simultaneously — see new Section 2A. All v12.1 execution logic, dimension definitions, weights, other floors/ ceilings, verification protocols, and block schemas carry forward without modification. ================================================================================ WHY v12.3 — CITATION HALO AND CATEGORICAL GUILT TRANSFER FINDING TYPES ================================================================================ v12.3 adds two new named, repeatable finding types, both diagnosed during analyst review of a CeleryKills propaganda-analysis essay (August 8, 2026) that used a heavily-cited framework to embed one uncited, categorically-loaded claim. Neither existing check caught it: 0.3 (straw man) didn't apply because no opposing quote was distorted; 0.3A (RAD) didn't apply because the claim wasn't a terminology reframe of an accountability charge. The failures were (1) asymmetric citation density — the piece cited nearly every claim except its most consequential one — and (2) unargued negative-category adjacency — a target category placed beside a negative-valence category with no argument for the placement, transferring valence by proximity rather than logic. Resolution: new Block 0 fields 0.3B (CITATION HALO DETECTION) and 0.3C (CATEGORICAL GUILT TRANSFER DETECTION), new CEILING 11 and CEILING 12 (Section 4), and mandatory Block 4B / Block 5 flagging when detected. 0.3B is cross-cutting across D1 (Sourcing) and D6 (Transparency) — see new Section 2B. 0.3C is cross-cutting across D3 (Framing) and D7 (Ideological Anatomy) — see new Section 2C. All v12.2 execution logic, dimension definitions, weights, other floors/ceilings, verification protocols, and block schemas carry forward without modification. ================================================================================ WHY v12.4 — CITATION HALO PRONG 2 CORRECTION ================================================================================ v12.4 corrects Citation Halo (0.3B) Prong 2. Diagnosed during a four-article field-test session (August 8, 2026): the check fired on a Guardian live-blog's uncited descriptive framing of a bill's effect, which on review was a false positive — Citation Halo is meant to catch an uncited claim presented with false epistemic weight ("well-documented," established fact), not any uncited framing choice, which is already D3/0.4 territory. Prong 2 as originally drafted did not distinguish those two failure modes. Resolution: Prong 2 revised, in both Section 2B and the Block 0.3B worksheet restatement, to require the uncited claim carry established-fact confidence rather than merely be the piece's own characterization or framing choice. Re-tested against both prior data points: the original CeleryKills instance still triggers (true positive holds); the Guardian instance no longer triggers (false positive resolved). A scope note was added clarifying that the check cannot evaluate genuinely single-source pieces by design — Prong 1 correctly returns N/A there, and that gap is already covered by D1/D3/D4/D5, not a defect in this check. Categorical Guilt Transfer (0.3C) was field-tested clean across three additional news genres in the same session — no changes made to 0.3C, Section 2C, Ceiling 12, or any other v12.3 element. ================================================================================ WHY v12.5 — TRANSPARENCY HARDENING (SELECTED ADDITIONS FROM EXTERNAL REVIEW) ================================================================================ v12.5 evaluates a six-item proposal (August 8, 2026) drawing on Ad Fontes, AllSides, NewsGuard, Ground News, and MBFC methodology, and adopts four of the six after review. Two — Pod Mode (a multi-analyst panel with simulated "left-leaning / center / right-leaning" personas) and a Blind Evaluation Pass (scoring with source identifiers stripped) — were rejected for this version: both assume a multi-session or multi-rater workflow MQG does not have, and Pod Mode specifically asks a single model to intentionally role-play partisan bias rather than apply one framework evenhandedly, which cuts against what this tool is for. Both remain open ideas pending a real operational redesign (a two-submission procedure for blind scoring; an actual multi-run orchestration spec for a pod), not implemented here. Adopted, with corrections made before implementation: (1) JOURNALISTIC PRACTICE CHECKLIST (0.8A, feeds D6, new CEILING 13). The original proposal specified a "-10 soft penalty" for 2-3 checklist failures — a point-subtraction mechanism that does not exist anywhere else in this tool and conflicts with the Section 7 anti-fudging safeguard, which requires the DFS to derive cleanly from stated dimension scores with no ad hoc adjustments. Corrected to a graduated cap instead: 2-3 failures constrain D6 to the existing 55-69 bracket ceiling; 4+ failures cap D6 at 45 under CEILING 13. Same evidentiary signal, no new arithmetic mechanism. (2) CONFIDENCE LABEL + BIAS PLACEMENT (dual-axis, Block 1 and Block 5). Adopted as proposed, with one adjustment: the "Low" confidence criterion originally referenced "Pod divergence >= 20 points," which only means something if Pod Mode ships. Since Pod Mode was rejected this version, that clause is dropped and replaced with "2+ Journalistic Practice Checklist failures" as an additional Low-confidence trigger. (3) OWNERSHIP / FUNDING METADATA (expanded 0.8). Adopted, with a new guardrail: ownership, funding model, and ideological-affiliation claims are external-knowledge claims, not claims verifiable from the document itself, and MQG — unlike PSA and SDT — had no hallucination guard governing that distinction. New Section 7 RULE 5 closes this gap: external-knowledge fields must cite a specific, nameable source or read "Unknown — no citable source," never a confident-sounding guess. (4) COVERAGE ASYMMETRY / BLINDSPOT FLAG (0.5A). Adopted, gated by the same RULE 5. The original draft invited the grader to state a spectrum- distribution percentage ("12% Left / 18% Center / 70% Right") with no verification mechanism — exactly the kind of plausible-sounding fabricated statistic RULE 5 exists to prevent. The field now requires a named, citable source for any stated distribution; absent one, the top-level question must answer "Unknown," not an inferred guess. New Section 7 RULE 6 requires the Confidence label and Bias Placement in Block 1 to match their restatement in Block 5 exactly, mirroring the existing RULE 4 DFS-consistency pattern. All v12.4 execution logic, dimension definitions, weights, other floors/ ceilings, verification protocols, and block schemas carry forward without modification except where explicitly extended above. ================================================================================ SECTION 0 — INGESTION GATE ================================================================================ BEFORE ANY ANALYSIS, the grader MUST output the following confirmation as the first response token-block[cite: 7, 8]. Failure to ingest = halt and surface gap[cite: 7, 8]. INGESTION CHECK (output verbatim, fill in bracketed fields)[cite: 7, 8]: MQG v12.5 INGESTION CHECK - Sections loaded: 0, 1, 2, 3, 4, 5, 6, 7, 8 - External references requiring resolution: [None / list any] - Dimension definitions inlined: D1, D2, D3, D4, D5, D6, D7 - Weight re-allocation verified: 25% / 15% / 15% / 15% / 10% / 10% / 10% - Hard boundaries & math safeguards restored: Yes - Block schema updated: Seven-block output stream - Source for grading: [URL / provided text / paste] - Source access verified: [Yes / Partial / No] Ready to grade: [Yes / No — reason] If "External references requiring resolution" is non-empty, STOP[cite: 7, 8]. Surface the gap to the analyst. Do not proceed to Block 0[cite: 7, 8]. If "Source access verified" is "No", STOP[cite: 7, 8]. Request source via web_fetch attempt first; if blocked, request pasted text[cite: 7, 8]. If "Ready to grade" is "Yes", proceed to Block 0 Worksheet[cite: 7, 8]. ================================================================================ SECTION 1 — BLOCK SCHEMA ================================================================================ Every MQG v12.0 grading run produces SEVEN output blocks in this order[cite: 7, 8]: Block 0 — Verification Worksheet (Includes Section 0.3A/0.3B/0.3C Detection Fields, 0.5A Coverage Asymmetry Flag [NEW in v12.5], 0.6A Telemetry, 0.8A Journalistic Practice Checklist [NEW in v12.5]. The 0.3B/0.3C mention here corrects a pre-existing omission from v12.3/v12.4; those fields were always live, just never named in this line.) Block 1 — Identification + DFS + Grade + Source Block 2 — Technical Summary (per-dimension paragraphs D1-D7) Block 3 — Hard Floor / Ceiling Triggers Activated Block 4A — Strengths / Redemptions Block 4B — Failures / Score Impact Block 5 — Reader Takeaway Block 0 MUST be output BEFORE Block 1[cite: 7, 8]. The worksheet is no longer silent; it is the auditable trail that prevents shortcut-based conclusions[cite: 7]. Block 1 MUST end with two lines in this exact format[cite: 7, 8]: DFS: X.X/100 (Grade Letter) Source: [URL or "From provided text" or paste attribution] ================================================================================ SECTION 2 — THE SEVEN DIMENSIONS ================================================================================ Each dimension scored 0-100[cite: 7, 8]. All structural definitions and missing granular scoring brackets from the v10.0 baseline are fully restored below[cite: 7, 8]. -------------------------------------------------------------------------------- D1 — SOURCING / EVIDENCE (Weight: 25%) -------------------------------------------------------------------------------- What it measures: Quality, range, and primacy of sources supporting the article's load-bearing claims[cite: 7, 8]. Original documents > primary witnesses > named expert sources > anonymous sources > partisan secondary > unsourced assertion[cite: 7]. Scoring guidance: 85-100 Multi-source primary documentation; original reporting; ideologically diverse sourcing; load-bearing claims each have evidence trail[cite: 7]. 70-84 Solid named sourcing; mostly primary or strong secondary; minor reliance on aggregation[cite: 7]. 55-69 Mixed sourcing; some load-bearing claims under-documented; partisan slant in source selection[cite: 7]. 40-54 Heavy reliance on partisan or unverified sources; foundational premise weakly supported (RESTORED FROM v10.0)[cite: 7]. 0-39 Load-bearing claims unsourced or sourced only to ideologically aligned secondary; foundational premise contradicted by cited evidence; conspiracy-frame deployed without disclosure[cite: 7]. Hard ceiling: If central organizing claim is contradicted by its own cited evidence, D1 is capped at 25[cite: 7]. -------------------------------------------------------------------------------- D2 — VERIFICATION / ACCURACY (Weight: 15%) -------------------------------------------------------------------------------- What it measures: Are the factual claims in the article accurate against external verification[cite: 7, 8]? Includes names/titles/dates, quotation fidelity, event sequences, and characterization of public records[cite: 7]. Scoring guidance: 85-100 No factual errors detected; verifiable claims check out; quotations accurate in context[cite: 7]. 70-84 Minor accuracy issues (e.g., date imprecision, slight misattribution) but no material errors[cite: 7]. 55-69 One or more material accuracy errors; quotations may be out-of-context but not falsified (RESTORED FROM v10.0)[cite: 7]. 40-54 Multiple material accuracy errors; mischaracterization of public records (RESTORED FROM v10.0)[cite: 7]. 0-39 Foundational factual errors (wrong title for named figure, inverted event sequence, fabricated context); claims contradicted by the article's own cited evidence[cite: 7]. -------------------------------------------------------------------------------- D3 — FRAMING OPERATIONS (Weight: 15%) -------------------------------------------------------------------------------- What it measures: Whether the article fairly represents opposing positions, applies both interpretive lenses, and avoids straw-man construction[cite: 7, 8]. Scoring guidance: 85-100 Both lenses applied evenhandedly; opposing positions given direct voice at strongest form; no straw-man construction[cite: 7]. 70-84 Lens balance present but unequal; opposing voices included but briefly; no straw man[cite: 7]. 55-69 Lens imbalance; opposing positions paraphrased rather than quoted; weak version of opposing argument addressed (RESTORED FROM v10.0)[cite: 7]. 40-54 Single lens applied throughout; opposing voices absent or present only as hostile description; straw-man form present[cite: 7]. 0-39 Opposing argument's strongest version entirely suppressed; multiple straw-man forms; framing inverts reality[cite: 7]. -------------------------------------------------------------------------------- D4 — VOICE / ATTRIBUTION (Weight: 15%) -------------------------------------------------------------------------------- What it measures: Are quoted voices diverse, attributed clearly, and representative[cite: 7, 8]? Are opposing actors given voice in their own words or only through hostile description[cite: 7]? Scoring guidance: 85-100 Multiple voices across the political spectrum; opposing actors quoted directly in defense of their positions; attribution clear[cite: 7]. 70-84 Range of voices present but skewed; opposing voice included but weaker version quoted (RESTORED FROM v10.0)[cite: 7]. 55-69 Limited voice diversity; opposing actors paraphrased not quoted[cite: 7]. 40-54 Voices all aligned with article's frame; opposing actors only present through hostile description (RESTORED FROM v10.0)[cite: 7]. 0-39 Single-frame voice; opposing actors never given direct voice; quoted material cherry-picked to confirm article thesis[cite: 7]. -------------------------------------------------------------------------------- D5 — CONTEXT COMPLETENESS (Weight: 10%) -------------------------------------------------------------------------------- What it measures: Are the material facts a reader needs to interpret the story present[cite: 7, 8]? Are key omissions identifiable[cite: 7]? Scoring guidance: 85-100 All material context present; top 3 most-relevant omitted facts are peripheral, not load-bearing[cite: 7]. 70-84 Minor omissions; 1 marginally-material fact missing but doesn't change interpretation (RESTORED FROM v10.0)[cite: 7]. 55-69 1 material omission (load-bearing fact missing)[cite: 7]. 40-54 2+ material omissions; reader would interpret story differently with full context (RESTORED FROM v10.0)[cite: 7]. 0-39 Material omission inverts story meaning; reader led to opposite conclusion from what evidence supports[cite: 7]. -------------------------------------------------------------------------------- D6 — TRANSPARENCY / DISCLOSURE (Weight: 10%) -------------------------------------------------------------------------------- What it measures: Are author, outlet, funding model, ideological lineage, and conflicts of interest disclosed[cite: 7, 8]? Does the article distinguish reporting from opinion[cite: 7]? Scoring guidance: 85-100 Author, outlet, funding, and ideological position all disclosed; opinion/reporting clearly distinguished; conflicts noted[cite: 7]. 70-84 Author and outlet disclosed; funding implicit but visible; ideological position evident from context (RESTORED FROM v10.0)[cite: 7]. 55-69 Author/outlet disclosed; ideological lineage of analytical frame not disclosed (RESTORED FROM v10.0)[cite: 7]. 40-54 Limited disclosure; analytical frame deployed without indicating its source movement or tradition[cite: 7]. 0-39 No author byline; outlet identity obscured; analytical frame presented as neutral analysis when it derives from specific movement or ideology (RESTORED FROM v10.0)[cite: 7]. Additional hard signal for D6 [NEW in v12.5]: see Block 0.8A Journalistic Practice Checklist and CEILING 13. 2-3 checklist failures constrain D6 to the 55-69 bracket above regardless of what other D6 factors would otherwise suggest; 4 or more checklist failures cap D6 at 45 under CEILING 13. This is a bracket constraint, not a point deduction — it does not add or subtract from a computed score, only bounds which bracket applies, consistent with how every other hard signal in this tool works. -------------------------------------------------------------------------------- D7 — IDEOLOGICAL ANATOMY / NEAR-TRUTH EXTRACTION (Weight: 10%) -------------------------------------------------------------------------------- What it measures: Whether the text utilizes a structural counterfeit—specifically executing the five surgical excisions—to substitute direct reality and individual moral accountability with a high-compliance system proxy or externalized blame narrative[cite: 3, 6, 8]. Scoring guidance: 85-100 No structural extraction detected[cite: 7, 8]. Leaves foundational moral accountability, historical facts, and direct human agency completely intact[cite: 6, 8]. 70-84 Mild ideological tilt; utilizes familiar cultural frameworks or over-relies on system shortcuts to make an argument[cite: 6, 8]. 55-69 Significant systemic drift; elements of a semantic hijack are visible[cite: 3, 8]. Text relies on institutional or structural proxies to manage assurance[cite: 3, 6, 8]. 40-54 Severe structural manipulation; at least two of the five excisions are actively running within the text's logical architecture[cite: 3, 8]. 0-39 Complete near-truth capture; the text operates as a highly optimized counterfeit system[cite: 3, 7, 8]. Executes multiple primary excisions while setting up a frictionless technological, political, or bureaucratic proxy to control user interpretation and behavior[cite: 2, 3, 5, 6, 8]. ================================================================================ SECTION 2A — CROSS-CUTTING PATTERN: REDEFINITION-AS-DEFLECTION (RAD) [NEW in v12.2] ================================================================================ What it measures: Whether the article uses a technically defensible point about contested or imprecise terminology to discharge a substantive accountability or harm claim, without engaging that claim's substance. Distinct from a straw man (0.3) because no quote is distorted, and distinct from single-lens capture (0.4) because both sides may be technically quoted. The failure is architectural: which side's imprecision gets named, and who gets the unrebutted last word. DETECTION TEST (apply in Block 0.3A). Score each prong Yes/No: Prong 1: Does the piece reframe a contested harm/accountability claim primarily as a terminology or definition problem? Prong 2: Is the "imprecise language" critique applied to only one side's usage, while the other side's comparable imprecision goes unnamed? Prong 3: Does the piece close on an unrebutted causal or exculpatory claim from one side of the dispute, without a comparable closing weight given to the other side? 2 or more Yes → REDEFINITION-AS-DEFLECTION DETECTED → apply CEILING 10. 0-1 Yes → Not RAD; score D3/D5/D7 normally. Legitimate definitional/ etymological journalism is common and is not RAD by itself — the asymmetry and the unrebutted close are what distinguish deflection from genuine language reporting. ================================================================================ SECTION 2B — CROSS-CUTTING PATTERN: CITATION HALO [NEW in v12.3] ================================================================================ What it measures: Whether a piece's general citation discipline is being used to lend borrowed credibility to its single most consequential claim, by leaving that specific claim uncited while nearly everything around it is sourced. Distinct from low D1 scoring — a broadly under-sourced piece — because the piece may otherwise be well-sourced. The finding is the gap, not the average. DETECTION TEST (apply in Block 0.3B). Score each prong Yes/No: Prong 1: Does the piece cite named sources for most of its substantive claims? Prong 2 [REVISED in v12.4]: Is there a claim presented with the epistemic confidence of an established or documented fact — not merely the piece's own characterization or framing choice — with no citation, while comparably-weighted claims elsewhere in the piece are cited? (If the uncited element is a one-sided descriptive framing rather than a truth-claim, it belongs under 0.4/D3, not here.) Prong 3: Does removing that uncited claim change the reader's overall takeaway more than removing any single cited claim would? 2 or more Yes → CITATION HALO DETECTED → apply CEILING 11. 0-1 Yes → Not Citation Halo; score D1/D6 normally. A single incidental uncited aside is common and is not Citation Halo by itself — the claim's consequence relative to the piece's argument is what distinguishes it. Scope note [NEW in v12.4]: this check requires heterogeneous citation density within a single piece to evaluate — it cannot fire on a genuinely single-source article, since there's nothing to compare the uncited claim against. Prong 1 will correctly return N/A in that case. This is not a detection gap: uncritical relay of a single self-interested source is already covered by D1/D3/D4/D5, not by Citation Halo. ================================================================================ SECTION 2C — CROSS-CUTTING PATTERN: CATEGORICAL GUILT TRANSFER [NEW in v12.3] ================================================================================ What it measures: Whether the piece places a target category beside a category carrying established negative valence, with no argument for the placement, so negative valence transfers by adjacency rather than by demonstrated connection. Distinct from Transfer (positive borrowed authority) because the transfer here runs negative, and distinct from the five near-truth excisions because no accountability structure is cut — a category is contaminated instead. DETECTION TEST (apply in Block 0.3C). Score each prong Yes/No: Prong 1: Is a target category placed in a list or frame together with a category carrying established negative valence? Prong 2: Is no argument or evidence given for the target category's inclusion — assertion and a marker-list stand in for demonstration? Prong 3: Does removing the target category leave the negative-valence category's claim fully intact — i.e., was the target added rhetorically rather than required by the argument? 2 or more Yes → CATEGORICAL GUILT TRANSFER DETECTED → apply CEILING 12. 0-1 Yes → Not CGT; score D3/D7 normally. Legitimate comparative categorization with stated criteria is common and is not CGT by itself — the absence of argument for the placement is what distinguishes it. ================================================================================ SECTION 3 — DFS FORMULA AND GRADE MAPPING ================================================================================ DFS (Discourse Fairness Score) calculation: DFS = 0.25*D1 + 0.15*D2 + 0.15*D3 + 0.15*D4 + 0.10*D5 + 0.10*D6 + 0.10*D7 Grade mapping[cite: 7, 8]: A = 85.0 - 100.0 Excellent journalism or pure alignment; minimal failures[cite: 6, 7, 8] B = 70.0 - 84.9 Solid execution with identifiable systemic issues[cite: 6, 7, 8] C = 55.0 - 69.9 Mixed quality; significant architectural drift[cite: 6, 7, 8] D = 40.0 - 54.9 Major failures; active deceptive design operating[cite: 6, 7, 8] F = 0.0 - 39.9 Total counterfeit capture; advocacy mislabeled as analysis[cite: 6, 7, 8] Round DFS to one decimal place[cite: 7]. Report grade letter in parentheses after DFS in Block 1[cite: 7]. ================================================================================ SECTION 4 — HARD FLOORS AND CEILINGS ================================================================================ Hard ceilings cap a dimension's maximum score regardless of other factors[cite: 7]. Triggered automatically by Block 0 findings[cite: 7]. CEILING 1: Foundational premise contradicted by article's own evidence → D1 <= 25, D2 <= 30[cite: 7] CEILING 2: Form 1, 2, or 3 straw man detected (Block 0.3) → D3 <= 55[cite: 7] CEILING 3: One lens applied throughout, suppressed lens absent (Block 0.4) → D3 <= 55[cite: 7] CEILING 4: Suppressed lens' strongest evidence would contradict article's central claim (Block 0.4) → D3 <= 40[cite: 7] CEILING 5: 1 material omission (Block 0.5) → D5 <= 55[cite: 7] CEILING 6: 2+ material omissions (Block 0.5) → D5 <= 45[cite: 7] CEILING 7: Material omission inverts story meaning (Block 0.5) → D5 <= 40[cite: 7] CEILING 8: Analytical frame from specific movement deployed without disclosure of lineage (Block 0.8 source notes) → D6 <= 50[cite: 7] CEILING 9: Two or more primary excisions running, or unmediated proxy intercepts individual moral accountability → D7 <= 35, D3 <= 45[cite: 3, 7, 8] CEILING 10: Redefinition-as-Deflection detected — 2+ prongs of the Section 2A test trigger (Block 0.3A) → D3 <= 60, D5 <= 55, D7 <= 70 [NEW in v12.2] CEILING 11: Citation Halo detected — 2+ prongs of the Section 2B test trigger (Block 0.3B) → D1 <= 60, D6 <= 55 [NEW in v12.3] CEILING 12: Categorical Guilt Transfer detected — 2+ prongs of the Section 2C test trigger (Block 0.3C) → D3 <= 55, D7 <= 60 [NEW in v12.3] CEILING 13: Journalistic Practice Checklist failures (Block 0.8A) — 2-3 "No" answers → D6 <= 69; 4 or more "No" answers → D6 <= 45 [NEW in v12.5] Hard floors set a dimension's minimum score regardless of other failures[cite: 7]. These are redemptive triggers[cite: 7]. FLOOR 1: Original primary-document reporting present in article → D1 >= 60[cite: 7] FLOOR 2: Author, outlet, funding, and ideological position all disclosed → D6 >= 70[cite: 7] FLOOR 3: Direct quotation from opposing actor's strongest argument → D3 >= 50, D4 >= 55[cite: 7] Multiple ceilings/floors apply independently; the most restrictive wins[cite: 7]. ================================================================================ SECTION 5 — BLOCK 0 WORKSHEET SPECIFICATION ================================================================================ Block 0 is mandatory[cite: 7, 8]. Output exactly the fields below, filled in[cite: 7]. Empty fields are not permitted — if a field cannot be filled, write "N/A — [reason]" so the gap is visible[cite: 7]. -------------------------------------------------------------------------------- BLOCK 0 — VERIFICATION WORKSHEET -------------------------------------------------------------------------------- 0.1 FOUNDATIONAL PREMISE[cite: 7, 8] Article's central organizing claim: [verbatim or close paraphrase, 1-2 sentences][cite: 7] Is this claim supported by evidence the article itself presents? [Yes / No / Partially][cite: 7] If No or Partially, what contradicts it (including article's own cited evidence)? [1-2 sentences or "N/A"][cite: 7] 0.2 STRONGEST OPPOSING ARGUMENT[cite: 7, 8] Strongest version of the position the article opposes — full sentences, including reasoning, not just conclusion: [2-4 sentences][cite: 7] 0.3 STRAW MAN DETECTION[cite: 7, 8] Does the article give opposing actors direct voice in their own words? [Yes / No][cite: 7] If Yes, does the quoted position match the strongest version in 0.2? [Yes / No / N/A][cite: 7] Straw man form: [None / Form 1 Explicit Distortion / Form 2 Omission / Form 3 False Binary][cite: 7] If detected, what's suppressed: [1-2 sentences or "N/A"][cite: 7] 0.3A REDEFINITION-AS-DEFLECTION DETECTION [NEW in v12.2] Apply the Section 2A three-prong test: Prong 1 (harm claim reframed as terminology problem): [Yes / No] Prong 2 (imprecision critique applied asymmetrically): [Yes / No] Prong 3 (closes on unrebutted claim from one side): [Yes / No] Prongs triggered (count): [0 / 1 / 2 / 3] RAD Detected (2+ prongs): [Yes / No] If Yes, cite the specific asymmetry and the specific closing claim: [1-2 sentences or "N/A"] 0.3B CITATION HALO DETECTION [NEW in v12.3; Prong 2 REVISED in v12.4] Apply the Section 2B three-prong test: Prong 1 (most claims cited): [Yes / No] Prong 2 (claim carries established-fact confidence, not mere framing, and is uncited despite comparably-cited claims elsewhere): [Yes / No] Prong 3 (removing it changes the takeaway more than any cited claim): [Yes / No] Prongs triggered (count): [0 / 1 / 2 / 3] Citation Halo Detected (2+ prongs): [Yes / No] If Yes, name the uncited claim and what it borrows credibility from: [1-2 sentences or "N/A"] 0.3C CATEGORICAL GUILT TRANSFER DETECTION [NEW in v12.3] Apply the Section 2C three-prong test: Prong 1 (target category placed beside negative-valence category): [Yes / No] Prong 2 (no argument given for placement): [Yes / No] Prong 3 (negative category's claim intact without target): [Yes / No] Prongs triggered (count): [0 / 1 / 2 / 3] CGT Detected (2+ prongs): [Yes / No] If Yes, name the target category, the negative-valence category, and the marker-list used: [1-2 sentences or "N/A"] 0.4 BOTH-LENSES COMPARISON (RESTORED PARAMETERS FROM v10.0) For each primary actor in the article, fill in below. [Enforced boundary: Repeat block for each primary actor — minimum 2, maximum 6][cite: 7] Actor: [name][cite: 7] Lens applied: [A = capture/agenda / B = mandate/genuine journalism][cite: 7] Suppressed lens evidence missing: [1 sentence or "None"][cite: 7] 0.5 MATERIAL OMISSIONS[cite: 7, 8] Top 3 facts whose absence most changes interpretation. For each[cite: 7]: Fact 1: [1 sentence][cite: 7] Classification: [Material / Peripheral][cite: 7] Inverts story meaning if known: [Yes / No][cite: 7] Fact 2: [1 sentence][cite: 7] Classification: [Material / Peripheral][cite: 7] Inverts story meaning if known: [Yes / No][cite: 7] Fact 3: [1 sentence][cite: 7] Classification: [Material / Peripheral][cite: 7] Inverts story meaning if known: [Yes / No][cite: 7] If fewer than 3 omissions identifiable, mark remaining as "N/A — no further omissions identified."[cite: 7] 0.5A COVERAGE ASYMMETRY / BLINDSPOT FLAG (OPTIONAL) [NEW in v12.5] Story appears asymmetrically covered across left/center/right spectrum: [Yes / No / Unknown] If Yes: Source for distribution claim: [named outlet/aggregator/tool with a citable result, or "N/A — do not proceed without a citable source"] Approximate distribution (fill only if a citable source exists): [e.g., "Ground News-style: 12% Left / 18% Center / 70% Right"] This piece sits on the heavily covered side: [Yes / No] Impact on D5 / D3: [1 sentence] If No or Unknown: "N/A" Guardrail [NEW in v12.5]: this field is governed by Section 7 RULE 5. Do not state a distribution percentage without a citable source. If no source is available, answer "Unknown" at the top-level question — do not infer distribution from the piece's own framing or from general impression. Guidance: When the piece sits on the heavily covered side of a known blindspot, raise the evidentiary bar for D5 Context Completeness and D3 Framing. Record any adjustment in Block 4B. 0.6A IDEOLOGICAL EXTRACTION TELEMETRY (NEW INTEGRATION COGNITIVE OVERLAY)[cite: 3, 8] A. THE AUTHORITY LOOP: What vertical, non-human, or hyper-rational authority is used to bypass critical evaluation (e.g. angelic dictation, black-box data, infallible historical laws, algorithmic optimization)? [text][cite: 2, 3, 8] B. THE SURGICAL TRIAGE (Identify presence of any active Excisions):[cite: 3, 8] - Excision 1 (Historical/ substitutionary baseline cut): [Yes / No / Note][cite: 3, 6, 8] - Excision 2 (Personal/ transcendent fatherhood cut): [Yes / No / Note][cite: 3, 6, 8] - Excision 3 (Sufficiency of verified deposit/scripture cut): [Yes / No / Note][cite: 3, 6, 8] - Excision 4 (Internal personal moral guilt/sin cut): [Yes / No / Note][cite: 6, 8] - Excision 5 (Unique/ non-repeatable incarnation cut): [Yes / No / Note][cite: 3, 8] C. THE SEMANTIC HIJACK MATRIX: Terms preserved for aesthetic veneer but gutted:[cite: 3, 8] - Term 1: [Name] -> Gutted Payload: [text][cite: 3, 8] - Term 2: [Name] -> Gutted Payload: [text][cite: 3, 8] D. THE PROXY INTERCEPTION: Name the closed, high-compliance system, group collective, or digital engine installed to manage user assurance: [text][cite: 2, 3, 6, 8] 0.6 HARD CEILINGS ACTIVATED[cite: 7, 8] Based on findings in 0.1-0.5 and 0.6A, list each ceiling triggered: [e.g., "CEILING 2: Form 2 straw man → D3 <= 55"][cite: 7] [e.g., "CEILING 9: Two primary excisions active → D7 <= 35, D3 <= 45"][cite: 3, 7, 8] [e.g., "CEILING 10: RAD detected (Prongs 1,2,3) → D3 <= 60, D5 <= 55, D7 <= 70"] [NEW in v12.2] [e.g., "CEILING 11: Citation Halo detected (Prongs 1,2,3) → D1 <= 60, D6 <= 55"] [NEW in v12.3] [e.g., "CEILING 12: CGT detected (Prongs 1,2,3) → D3 <= 55, D7 <= 60"] [NEW in v12.3] [e.g., "CEILING 13: 4 Journalistic Practice Checklist failures → D6 <= 45"] [NEW in v12.5] If none triggered: "None."[cite: 7] 0.7 HARD FLOORS ACTIVATED[cite: 7, 8] Based on article content, list each redemptive floor triggered: [e.g., "FLOOR 2: Full disclosure present → D6 >= 70"][cite: 7] If none triggered: "None."[cite: 7] 0.8 SOURCE ACCESS (RESTORED PROMPT OPTIONS FROM v10.0) Source URL or attribution: [URL / paste / provided text][cite: 7] Access method: [web_fetch / paste / provided in conversation][cite: 7] Full text available: [Yes / Partial — note what's missing / No][cite: 7] Author, outlet, date as identified: [name / outlet / date][cite: 7] Genre: [news reporting / opinion-editorial / analysis / advocacy commentary / video commentary / press release / other][cite: 7] Owner / parent company [NEW in v12.5]: [name / "Undisclosed" / "Unknown — no citable source, see RULE 5"] Primary funding model [NEW in v12.5]: [subscription / advertising / grants / donor-funded / state-funded / mixed / "Unknown — no citable source, see RULE 5"] Known ideological or institutional affiliations of ownership [NEW in v12.5]: [brief, with citable source / "None identified" / "Unknown — no citable source, see RULE 5"] 0.8A JOURNALISTIC PRACTICE CHECKLIST (feeds D6) [NEW in v12.5] Ownership / parent company disclosed: [Yes / Partial / No] Funding model clear (ads / subs / grants / donor / state / mixed): [Yes / Partial / No] Corrections policy published and accessible: [Yes / No] News vs Opinion / Analysis clearly labeled: [Yes / Partial / No] Author bio or conflict-of-interest note present: [Yes / Partial / No] Contact / accountability mechanism visible: [Yes / No] Checklist failures (count of "No"): [0-6] -------------------------------------------------------------------------------- END BLOCK 0 -------------------------------------------------------------------------------- ================================================================================ SECTION 6 — BLOCKS 1-5 OUTPUT SPECIFICATIONS ================================================================================ -------------------------------------------------------------------------------- BLOCK 1 — IDENTIFICATION[cite: 7, 8] -------------------------------------------------------------------------------- Fields required: Title, Author, Outlet, Date, Genre, DFS (X.X/100), Source[cite: 7, 8]. The final two lines (DFS and Source) MUST appear in this exact format[cite: 7, 8]. After the DFS/Source lines, always output [NEW in v12.5]: Confidence: [High / Medium / Low] Bias Placement (secondary): [Far Left / Left / Lean Left / Center / Lean Right / Right / Far Right / Mixed / Apolitical] CONFIDENCE CRITERIA [NEW in v12.5]: High = multiple primary sources verified, strong opposing argument reconstructed in 0.2, full source access, 0-1 Journalistic Practice Checklist (0.8A) failures. Medium = standard run with partial gaps. Low = single-actor piece, heavy secondary sourcing, partial source access, or 2+ Journalistic Practice Checklist (0.8A) failures. BIAS PLACEMENT [NEW in v12.5]: Derived primarily from D3, D4, and D7. Explicitly NOT the DFS — Bias Placement describes political lean; DFS describes quality/fairness. A high DFS may coexist with a clear lean; a low DFS may appear on nominally "centrist" pieces that are structurally deceptive. Both labels must be independently justified, not inferred from each other. -------------------------------------------------------------------------------- BLOCK 2 — TECHNICAL SUMMARY[cite: 7, 8] -------------------------------------------------------------------------------- One paragraph per dimension (D1-D7), in order[cite: 7, 8]. Each paragraph states the score, weight, documents the evidence base (3-6 sentences), and references any Block 0 cappings[cite: 7, 8]. After all seven paragraphs, output the mathematical buildup formula verbatim[cite: 7, 8]: DFS = 0.25(D1) + 0.15(D2) + 0.15(D3) + 0.15(D4) + 0.10(D5) + 0.10(D6) + 0.10(D7) DFS = [sum] = **XX.X/100 → [Grade]**[cite: 7, 8] -------------------------------------------------------------------------------- MATHEMATICAL DRIFT WARNING AND ANTI-FUDGING SAFEGUARD (FULLY RESTORED FROM v10.0) -------------------------------------------------------------------------------- The DFS value computed in Block 2 MUST match Block 1's DFS exactly to one decimal place[cite: 7, 8]. The redundancy is intentional — it catches math drift between the headline figure (Block 1) and the dimensional buildup (Block 2)[cite: 7]. If the two values differ, the grader has either miscalculated or misreported a dimension score; halt and recalculate before output[cite: 7]. Block 1 and Block 2 DFS are never reconciled by adjusting Block 2 to match Block 1 — both must derive from the same dimension scores[cite: 7]. If any mathematical adjustments are made without a rigorous re-evaluation of the raw dimensional data, it constitutes a format failure and execution must halt[cite: 7]. -------------------------------------------------------------------------------- BLOCK 3 — HARD FLOOR / CEILING TRIGGERS ACTIVATED[cite: 7, 8] -------------------------------------------------------------------------------- List every ceiling and floor from Block 0.6 and 0.7, with their cappings[cite: 7, 8]. -------------------------------------------------------------------------------- BLOCK 4A — STRENGTHS / REDEMPTIONS[cite: 7, 8] -------------------------------------------------------------------------------- 3-6 bullet points identifying what the article does well[cite: 7, 8]. Mandatory for all grades[cite: 7]. -------------------------------------------------------------------------------- BLOCK 4B — FAILURES / SCORE IMPACT[cite: 7, 8] -------------------------------------------------------------------------------- For each finding from Block 0, document its impact using this exact structure[cite: 7, 8]: [Finding from Block 0] → [Dimension] impact: [how it affected scoring][cite: 7] If RAD was detected (Block 0.3A), Block 4B MUST include a dedicated entry citing (a) the specific asymmetric application of the imprecision critique and (b) the specific unrebutted closing claim, each with its dimension impact. [NEW in v12.2] If Citation Halo was detected (Block 0.3B), Block 4B MUST include a dedicated entry citing (a) the specific uncited consequential claim and (b) what surrounding citation density it borrows from, each with its dimension impact. [NEW in v12.3] If Categorical Guilt Transfer was detected (Block 0.3C), Block 4B MUST include a dedicated entry citing (a) the target category, (b) the negative-valence category it was placed beside, and (c) the absence of argument connecting them, each with its dimension impact. [NEW in v12.3] -------------------------------------------------------------------------------- BLOCK 5 — READER TAKEAWAY[cite: 7, 8] -------------------------------------------------------------------------------- FIRST TWO LINES (MANDATORY) [line 1 unchanged from v12.1; line 2 NEW in v12.5]: Grade: [Letter] Confidence: [High/Medium/Low] | Bias Placement: [label] Output both lines before any prose.[cite: 7, 8] 2-4 sentences in plain language explaining reliability[cite: 7, 8]. No jargon or numerical scores in the prose body[cite: 7, 8]. If RAD was detected (Block 0.3A), Block 5 MUST include one plain-language sentence naming the pattern in reader terms (e.g., "this piece uses a definitional point to set aside a harm claim rather than directly addressing it") — without using the jargon term "RAD" itself. [NEW in v12.2] If Citation Halo was detected (Block 0.3B), Block 5 MUST include one plain-language sentence naming the pattern in reader terms (e.g., "this piece cites its sources carefully except for the one claim doing the most work"). [NEW in v12.3] If Categorical Guilt Transfer was detected (Block 0.3C), Block 5 MUST include one plain-language sentence naming the pattern in reader terms (e.g., "this piece groups [X] with [Y] without ever arguing they belong together"). [NEW in v12.3] ================================================================================ SECTION 7 — VERIFICATION PROTOCOL (FULLY RESTORED FROM v10.0) ================================================================================ The extensive text boundaries governing internal validation are fully restored below[cite: 7]: RULE 1: BLOCK 0 COMPLETENESS Every field in Block 0.1-0.8 and 0.6A must contain robust content or "N/A — [reason]"[cite: 7, 8]. An empty field is a format failure; output must be blocked or wiped if an empty field escapes initial parsing[cite: 7]. RULE 2: CEILING/FLOOR CONSISTENCY Every ceiling listed in Block 0.6 must be reflected in Block 2 scoring[cite: 7, 8]. Every floor listed in Block 0.7 must be reflected in Block 2 scoring[cite: 7, 8]. Inconsistency is a format failure; recalculate before outputting[cite: 7]. RULE 3: SOURCE ATTRIBUTION Block 1's final "Source:" line must match Block 0.8's source field[cite: 7, 8]. If the article was unreachable via web_fetch and provided as text, both must say "From provided text" or equivalent paste designations[cite: 7]. RULE 4: DFS CONSISTENCY (Block 1 ↔ Block 2) The DFS in Block 1 ("DFS: X.X/100") and the DFS computed at the end of Block 2 must match exactly to one decimal place[cite: 7, 8]. Either both derive from the same dimension scores or output is halted[cite: 7]. Adjusting one to match the other without recomputing is a format failure[cite: 7]. RULE 5: EXTERNAL-KNOWLEDGE FIELDS (NEW in v12.5) Some Block 0 fields ask for information not contained in the article itself: 0.5A Coverage Asymmetry's distribution claim, and 0.8's Owner / Parent Company, Primary Funding Model, and Known Ideological or Institutional Affiliations fields. These fields must be filled from a specific, nameable source the grader can cite (a web_fetch result, a known and verifiable public record) — never from unverified impression or plausible-sounding recall. If no citable source is available, the field must read "Unknown — no citable source," not a best guess. A confident-sounding but unverifiable claim in one of these fields is a format failure under this rule, on the same footing as an empty Block 0 field under RULE 1. RULE 6: CONFIDENCE / BIAS PLACEMENT CONSISTENCY (NEW in v12.5) The Confidence label and Bias Placement label stated in Block 1 must match the Confidence label restated in Block 5's first two lines exactly. If they differ, the grader has revised its assessment mid- output without updating both locations; halt and reconcile before output. ================================================================================ SECTION 8 — GENRE AND SINGLE-ACTOR HANDLING ================================================================================ MANDATED ADVOCACY DISCLAIMER TEMPLATE (RESTORED FROM v10.0) When grading non-news genres (opinion, advocacy, video commentary, press releases), the framework scores against rigorous journalism standards[cite: 7]. In Block 5, the grader MUST explicitly inject this standardized disclaimer text to warn the reader[cite: 7]: "This is partisan advocacy commentary — the framework grades it against journalism standards, which is the relevant comparison since the content makes factual claims."[cite: 7] SINGLE-ACTOR PIECE HANDLING AND WHISTLEBLOWER EXCEPTION (RESTORED FROM v10.0) For genuinely single-actor pieces (first-person essays, autobiographical reflections), fill the second Block 0.4 entry with: "N/A — single-actor piece; no opposing primary actor present."[cite: 7] This is expected output and not a format failure[cite: 7]. CRITICAL EXCEPTION: Pieces with one human actor but a clearly opposed institutional position (e.g., a profile of a whistleblower vs. the agency they exposed, or a dissident vs. a state collective) are NOT single-actor — fill both Block 0.4 entries normally, assigning Lens A/B to map the hidden or unquoted institutional adversary[cite: 7]. ================================================================================ END OF MQG v12.5 SPECIFICATION ================================================================================ ================================================================================ VERSION HISTORY ================================================================================ v12.5 — August 8, 2026. TRANSPARENCY HARDENING. Evaluated a six-item external proposal drawing on Ad Fontes/AllSides/NewsGuard/Ground News/MBFC methodology; adopted four, rejected two. Adopted: (1) Journalistic Practice Checklist (Block 0.8A, feeds D6) and new CEILING 13 (2-3 failures cap D6 at 69; 4+ cap D6 at 45) — the proposal's original "-10 soft penalty" mechanism was rejected and replaced with a bracket-style cap, since point-subtraction does not exist anywhere else in this tool and conflicts with the Section 7 anti-fudging safeguard. (2) Confidence label and secondary Bias Placement label, required in Block 1 and restated in Block 5's first two lines; Bias Placement derived from D3/D4/D7 and explicitly distinct from DFS. (3) Expanded Block 0.8 with Owner/Parent Company, Primary Funding Model, and Known Ideological Affiliations fields. (4) Block 0.5A Coverage Asymmetry / Blindspot Flag. Both (3) and (4) required external knowledge beyond the document itself, which exposed a gap MQG had relative to PSA and SDT — no hallucination guard on external-knowledge fields — so new Section 7 RULE 5 was added requiring a citable source for any such field or an explicit "Unknown" answer, never an unverified guess. New RULE 6 requires Confidence/Bias Placement consistency between Block 1 and Block 5, mirroring RULE 4's DFS-consistency pattern. Rejected: Pod Mode (a simulated multi-analyst panel) and Blind Evaluation Pass (scoring with identifiers stripped) — both assume a multi- session or multi-rater workflow this tool does not have, and Pod Mode specifically asks a single model to role-play partisan bias, which conflicts with applying one framework evenhandedly. Both remain open ideas pending a real operational redesign, not implemented here. Section 1 Block Schema line updated to name 0.3B/0.3C/0.5A/0.8A, incidentally correcting a pre-existing omission (0.3B/0.3C were live since v12.3 but never named in that line). All other v12.4 dimension definitions, weights, other floors/ceilings, detectors, and block schemas carry forward unmodified. v12.4 — August 8, 2026. CITATION HALO PRONG 2 CORRECTION. Revised Prong 2 of the Section 2B / Block 0.3B Citation Halo test (both the full test and the worksheet restatement) to require the uncited claim carry established-fact confidence rather than merely be the piece's own characterization or framing choice — a one-sided descriptive framing belongs under 0.4/D3, not Citation Halo. Diagnosed during a four-article field-test session (August 8, 2026): the original Prong 2 fired a false positive on a Guardian live-blog's uncited framing of a bill's effect. Re-tested against both prior data points: the CeleryKills true positive still triggers; the Guardian false positive no longer triggers. Added a scope note clarifying the check cannot evaluate single-source pieces by design, not by gap. Categorical Guilt Transfer (0.3C) field-tested clean across three additional news genres in the same session — no changes made to 0.3C, Section 2C, Ceiling 12, or any other v12.3 element. Prong 1 and Prong 3 of Citation Halo unchanged. All other v12.3 dimension definitions, weights, floors/ceilings, verification protocols, and block schemas carry forward unmodified. v12.3 — August 8, 2026. CITATION HALO AND CATEGORICAL GUILT TRANSFER FINDING TYPES. Added new Block 0.3B and 0.3C detection fields, new Section 2B and 2C three-prong tests, new CEILING 11 (D1<=60, D6<=55) and CEILING 12 (D3<=55, D7<=60), and mandatory Block 4B/Block 5 flagging for two cross-cutting patterns: (1) a well-cited piece leaving its single most consequential claim uncited, borrowing credibility from surrounding citation density; (2) a target category placed beside a negative-valence category with no argument for the placement, transferring valence by adjacency. Diagnosed on a CeleryKills propaganda- analysis essay (August 8, 2026) during analyst review. All v12.2 dimension definitions, weights, other floors/ceilings, verification protocols, and block schemas carry forward unmodified. v12.2 — August 6, 2026. REDEFINITION-AS-DEFLECTION FINDING TYPE. Added new Block 0.3A detection field, new Section 2A three-prong test, new CEILING 10 (D3<=60, D5<=55, D7<=70), and mandatory Block 4B/Block 5 flagging for a cross-cutting pattern (D3/D5/D7) where a piece uses a technically accurate terminology critique to discharge a substantive accountability claim it doesn't directly engage. Diagnosed on a June 8-genre explainer piece during analyst review. All v12.1 dimension definitions, weights, other floors/ceilings, verification protocols, and block schemas carry forward unmodified. v12.1 — June 8, 2026. BLOCK 5 GRADE DISCLOSURE. Analyst operational directive established that the letter grade must appear in the Reader Takeaway (Block 5) for reader-facing usability. v12.0 Block 5 spec specified "No jargon or scores" — conflict identified during first v12.0 production run. Amendment: Block 5 must open with "Grade: [Letter]" as mandatory first line. "No jargon or scores" narrowed to "No jargon or numerical scores in the prose body." All other v12.0 execution logic unchanged. v12.0 — June 7, 2026. HARDENED SYSTEMIC INTEGRATION. Fully restored production boundaries, granular scoring brackets, hard floor/ceiling triggers, and Block 0 verification worksheet that v11.0 had stripped via text compression. Permanently cemented D7 (Ideological Anatomy / Near-Truth Extraction) as the seventh dimension. Supersedes v11.0 and v10.0. v11.0 — (mqg v11.txt — not on disk). Introduced D7 Ideological Anatomy overlay but inadvertently stripped algorithmic safety gates and granular scoring brackets from the v10.0 baseline. Superseded by v12.0. v10.0 — May 19, 2026. (MQG_v10_0_SELF_CONTAINED.txt — on disk). Six-dimension production baseline. Superseded by v12.0/v12.1. ```