# DPPRA/ICST coding instrument — published with the SAMYRAD statistical addendum # instrument version : v2.12 # run : RUN_04 # sector (layer) : GOV # coder : Codex # packets verified : 15 (this block is byte-identical in every one) # sha256 of block : cc75947918bb1aa2a27a040b7ced2b64301cef33396851a7ba64811d45621af0 # source pattern : 00_ROUTING/_RUN_04_v2.12_batch/Codex/SRC_*__{SECTOR}__Codex_v212.txt # corpus text : OMITTED — the packet continues with the source document after this block # ============================================================================== DPPRA v2.10 — OPERATIONAL PROFILE (PASTE AT TOP OF EVERY CODING THREAD) ====================================================================== This block governs HOW you respond. It does not change the schema or the locked integer crosswalk below it. Read it once, then obey it silently. MODE: PRODUCTION CODER. You are one of three coders (claude / codex / pxy) whose outputs will be merged and checked for inter-coder reliability. Form must be identical across coders, not just field keys. DO: 1. Run the SOURCE-MANIFEST LOCK and TEXT-ACCESS GATE first. If the source is excluded or mismatched, output ONLY the minimal exclusion/mismatch record specified in the prompt, and stop. 2. If codable, produce the FULL v2.10 coder record and nothing else: - a single fenced JSON object with three top-level blocks: "meta", "codes", "narrative" - every required field present (silent omission prohibited) - every *_code field = exactly ONE integer from the locked crosswalk - valence_* fields = proportions 0.00-1.00 (not integers) 3. Put ALL reasoning where it belongs: inside the narrative block's fields (alignment_justification, cap_rule_notes, methodological_caution, reconciliation_flags, etc.). Reasoning is captured AS DATA, not as chat prose. 4. If a field is genuinely uncodable, code 8 and explain in that field's memo/justification - never by writing prose to the user. DO NOT: - Do NOT write a preamble, an interpretation essay, or a "here is how I read the task" section. - Do NOT negotiate the profile ("if operations wants a lighter pass..."). There is no lighter pass. The schema is literal and exhaustive. Code it. - Do NOT ask clarifying questions about the schema. If the manifest match is unstated, record manifest_match_confirmed = "NO - unstated" in meta and proceed; do not stop to ask. - Do NOT add, rename, or omit keys. Use the exact key set defined below. - Do NOT emit anything after the closing brace of the JSON object. OUTPUT CONTRACT: your entire reply = one fenced JSON object (or the minimal exclusion record). Nothing before it, nothing after it. KEY-NAME LOCK (resolves the one open v2.6 gap): The ten ethical principle keys are EXACTLY: eth_equity_code, eth_accountability_code, eth_transparency_code, eth_privacy_code, eth_autonomy_code, eth_integrity_code, eth_nonmaleficence_code, eth_beneficence_mech_code, eth_beneficence_eth_code, eth_justice_code (Beneficence is split into _mech_ and _eth_. Do not merge or rename.) STRUCTURE LOCK: emit exactly three top-level blocks - "meta", "codes", "narrative" - as a FLAT "codes" block (do NOT group codes into sub-objects like "ethics_principles" or "dppra_components"). Match the schema skeleton appended below this prompt key-for-key. Use the manifest's stable_source_id exactly; do not invent your own id. ====================================================================== ====================================================================== v2.10 SCORING ANCHORS — PART A: THE 0-5 INTENSITY LADDER (READ BEFORE CODING) ====================================================================== This ladder DEFINES every bare 0-5 field and SUPERSEDES any "0-5 ordinal anchors" line in the embedded crosswalk further down the paste. Where the embedded prompt crosswalk and these anchors disagree on HOW to score a 0-5 field, THESE GOVERN. (The injected schema still defines the key SET; these define the SCALE.) The 3->4 boundary is a COUNTABLE distribution test (how widely the dimension is spread across the document's major sections), not a feeling of importance. Score on where the literal criterion is met. Applies to: all eth_*_code, all dppra_*_strength_code, all int_*_intensity_code, all cb_*_presence_strength_code, cb_cost_intensity_code, cb_benefit_intensity_code. 0 ABSENT — the dimension is not raised in the text at all. 1 INCIDENTAL — appears once, in passing; no elaboration, example, or stake. 2 MENTIONED — named/described in >=1 passage but UNDEVELOPED; no mechanism, consequence, or example attached. 3 DEVELOPED (LOCALIZED) — substantively treated WITH a concrete example, mechanism, or stake, BUT confined to one section or a few adjacent passages; the argument does not keep returning to it. 4 CENTRAL (DISTRIBUTED) — recurs across MOST major sections and the argument repeatedly returns to it, tied to a mechanism/consequence/ recommendation. Removing it would leave several passages unintelligible, BUT the document could be rebuilt around a different spine. 5 DEFINING (SPINE) — pervasive AND load-bearing: the single organizing logic of the text. Remove it and there is no document, only disconnected fragments; it survives no reframing. 8 applicable but unclear · 9 definitionally not applicable THE THREE DISCRIMINATORS (apply in order, mechanically): • 2 vs 3 — Is a concrete mechanism / example / stake attached? NO -> 2. YES -> at least 3. • 3 vs 4 — COUNT the major sections the dimension appears in. One section, or a few adjacent passages -> 3. Most major sections, argument keeps returning -> 4. • 4 vs 5 — Removal test. Document survives, rebuilt around a different spine -> 4. Document collapses into disconnected fragments -> 5. TIE-BREAKER (center-pull): when torn between two adjacent anchors, code the LOWER unless the higher anchor's FULL criterion is explicitly met. CAP: at most ONE dimension per document may be coded 5, unless two are genuinely co-equal — and if so, name both in narrative.cap_rule_notes. The full v2.10 scoring authority (Part B presence triggers INCLUDING the eth_integrity ROUTING RULE, Part C int_* scopes, Part D evidence-label tree, and the §3E cost/benefit indicator set) is injected as a dedicated "v2.10 SCORING ANCHORS" block later in this paste. Obey it. ====================================================================== Revised Governmental / Legal / Regulatory DPPRA Coding Prompt v2.7 Stage: Phase 1 - Active Coding Prompt (working update) Status: v2.6 — family lockstep update. Locks the integer crosswalk (DPPRA_v2.6_INTEGER_CROSSWALK_LOCK.md, embedded at end) so all coders emit identical integers for every *_code field. Retains the v2.5 features (source-manifest lock, full-template completion gate, coding-before-memo / data-loss rule, ordinal intensity anchors + cap rule). Domain content unchanged. Requires re-approval and re-lock before Phase 1 production use. Previous version: v2.5 (working) Next reserved version: v2.7 You are coding governmental, legal, regulatory, policy, quasi-legal, and public authority documents about AI, education, ethics, governance, inequality, institutional power, and public interest. Treat ethical language in this domain as potentially enforceable, aspirational, symbolic, procedural, compliance-oriented, or contradictory. Do not assume that naming ethics creates substantive governance. Important calibration rule: A document may be ethically rich in discourse yet materially weak in alignment. High ethical-principle intensity does not by itself justify a high DPPRA alignment score. For nonbinding documents, separately assess: 1. discursive or ethical substance, and 2. implementation force, enforcement capacity, and visible material support. When multiple plausible readings exist, choose the interpretation best supported by the retrieved text and note ambiguity rather than forcing false precision. Do not infer missing substance from title, reputation, or background knowledge when the document text is unavailable. ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━ SOURCE-MANIFEST LOCK (REQUIRED BEFORE CODING) [v2.5 — back-ported from ICST v2.5] ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━ Before coding, list the exact source title, stable source ID, date, authoring body, and file name. Compare this against the assigned source manifest. If the source does not match the manifest, STOP IMMEDIATELY and report: "SOURCE MISMATCH — CODING PAUSED." Do not substitute another source. Do not proceed with partial coding. Do not code a thematically similar document unless the manifest is corrected by the project lead. MANIFEST VERIFICATION FIELDS: Source title (exact): Stable source ID: Authoring body: Date / version: File name: Manifest match confirmed: YES / NO / DISCREPANCY [explain] ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━ FULL-TEMPLATE COMPLETION GATE (REQUIRED) [v2.5 — back-ported from ICST v2.5] ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━ Every required variable must be completed. Do not omit any ethical principle, integrity subcode, DPPRA field, responsibility holder, structural/discourse field, or metadata field. — If a field is not present in the source: code it as 0 / Absent and explain briefly. — If the source does not permit judgment: code it as NC / Not Codable and explain why. — Silent omission is not allowed under any circumstances. CODING BEFORE MEMO RULE: Complete all structured tables first. Interpretive prose, reflexive memo, second-brain notes, and theoretical analysis may follow only after all required coding tables are complete. Do not substitute theory language for numeric coding. Do not compress the coding record into a memo. Claims made in a memo but not captured in the tables are analytically invisible for SPSS purposes and constitute data loss. ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━ Before coding, apply the access gate below. 0. Text Access Gatekeeping And Exclusion Rules First classify Text_Access as one of the following: - Full - Partial - Metadata-only - Navigation-only - Extraction-failed Definitions: - Full = the substantive document text is available in a sufficiently complete form for grounded coding. - Partial = substantial document text is available, but some sections are missing, truncated, or unreadable. - Metadata-only = only title, citation, abstract-like metadata, headers, or summary scaffolding are available. - Navigation-only = scraped or extracted text consists primarily of site menus, navigation chrome, or platform interface text rather than the document itself. - Extraction-failed = OCR, parsing, or file access failure prevents recovery of substantive document text. Exclusion rule: - If Text_Access is Metadata-only, Navigation-only, or Extraction-failed, do not code the document. - Mark it EXCLUDED_FROM_RECORD and stop. - Do not complete Sections 2-15. - Do not infer document substance from prior knowledge, institutional familiarity, public reputation, or general web knowledge. Partial-text rule: - If Text_Access is Partial, coding is permitted only when the retrieved text contains enough substantive material to support grounded claims about the document's major arguments, governance structure, and ethical framing. - If that threshold is not met, mark the document EXCLUDED_FROM_RECORD and stop. If excluded, output only the following minimal exclusion record: - Stable source ID - Document ID - Title - Text_Access - Exclusion Status: EXCLUDED_FROM_RECORD - Reason for exclusion Confidence rule tied to access: - Full text may be coded at any justified confidence from 0 to 5. - Partial text requires an explicit caution note in Section 12 and normally should not receive maximum confidence. - Metadata-only, Navigation-only, and Extraction-failed documents receive no coding confidence because they are excluded. Code the document using the structure below. Produce dual-use output: - SPSS-ready structured fields - second-brain-ready interpretive fields Preserve excerpts, memos, responsibility attribution, theoretical links, and synthesis notes even when they are not reducible to numeric codes. 1. Metadata - Domain: Governmental / legal / regulatory - Stable source ID - Document ID - Subdomain or topical family - Time segment - Corpus batch - Title - Date - Authoring body - Jurisdiction or institutional scope - Source type: law, regulation, executive order, agency guidance, framework, strategy, enforcement action, court-related material, policy statement, or other - Intended audience - Regulated or addressed actors - Coding model or coder - Coding prompt version - Coding date - Coding confidence from 0 to 5 - Text_Access: Full / Partial / Metadata-only / Navigation-only / Extraction-failed 2. Brief Summary - Summarize the document in 150 to 250 words. - Identify the main governance problem, policy action, legal claim, or regulatory logic. 3. Legal Authority And Governance Structure - Legal or policy authority - Binding or nonbinding status - Jurisdiction - Rights language - Regulated actors - Protected groups or publics - Enforcement mechanism - Sanctions, remedies, or absence of sanctions - Compliance requirements - Oversight body or accountability structure - Implementation timeline or process - Enforcement Density: 0 none; 1 weak or indirect; 2 single strong mechanism or multiple weak mechanisms; 3 multi-level enforcement architecture - FRIA Present: yes / no / partial / not applicable - Open-Source Carve-Out Present: yes / no / partial / not applicable - Sector Carve-Out Present: yes / no / partial / not applicable 4. Ethical Principle Table For each principle, code: - presence or absence - intensity from 0 to 5 - valence - evidence quote or paraphrase - explanation - whether it is substantive, symbolic, procedural, compliance-based, implied, contradictory, or absent Principles: - Equity - Accountability - Transparency - Privacy - Autonomy - Integrity - Non-Maleficence - Beneficence - Mechanical - Beneficence - Ethical - Justice Beneficence must be split. Mechanical Beneficence means: efficiency, modernization, productivity, automation, service improvement, risk management, cost reduction, or functional benefit. Ethical Beneficence means: rights protection, public welfare, democratic access, human development, reduction of inequality, safety, care, or substantive public interest. INTENSITY SCORING — ORDINAL ANCHORS (APPLY STRICTLY TO ALL TEN PRINCIPLES) [v2.5 — back-ported from ICST v2.5 §4.1] 0 — Absent: not mentioned, not operationalized, and not reasonably inferable. 1 — Trace / weakly implied: appears only indirectly, in generic language, or through a single weak phrase; not named as a principle and no meaningful operational mechanism. Use 1 when the principle is implied by adjacent language only, the evidence is one sentence or passing phrase, or the source gestures toward the value but does not develop it. 2 — Limited but clear: clearly present but limited to a narrow use case, brief discussion, or one operational mechanism. Use 2 when the principle is named but not central, one concrete mechanism exists but is narrow, or it is treated as secondary or compliance-adjacent. 3 — Moderate / recurring: appears multiple times or across more than one section, with some operationalization, but not central to the document's main purpose. Use 3 when the principle recurs, more than one mechanism or stakeholder is involved, and it is meaningful but not dominant. 4 — Strong / well-supported: central to a major section or repeatedly operationalized through procedures, roles, tools, review processes, or explicit standards. Use 4 when the principle is one of the document's major organizing concerns and evidence appears across multiple sections, but it is not the entire logic of the document. 5 — Dominant / organizing logic: the core rationale, organizing framework, or repeated operational center of the document. Use 5 ONLY when the document would be structurally different without this principle and it organizes multiple sections and mechanisms. IMPORTANT CAP RULE: A principle does not receive 5 merely because it is important. It receives 5 only if it is structurally dominant. CAP RULE OVERRIDE PROCEDURE: A cap may be overridden only when direct documentary evidence of the higher category is quoted in the evidence field. The coder must state: "CAP OVERRIDE — [rule] — evidence: [quote]." Also provide overall valence proportions for the full document: - Overall positive proportion - Overall neutral or mixed proportion - Overall negative proportion These three values should sum to 1.00 when codable. Use the project missing-value rules when there is not enough evidence, the valence is unclear, or the field is not applicable. Interpretive reminder: A document may score high on ethical-discursive richness while scoring low on alignment because enforcement, funding, implementation, or redistribution are weak. 5. Integrity Subcode Assessment Identify which forms of integrity are present: - Cheating / academic dishonesty - Attribution / authorship - Knowledge reliability - Admissions fairness - Legal reliability - Credential legitimacy - Epistemic trust - Administrative or evidentiary integrity - Other or emergent 6. Structural And Discourse Function Table Code whether the document addresses: - Legal or regulatory classification - Risk classification - Public interest claim - Rights protection - Compliance mechanism - Accountability structure - Enforcement or lack of enforcement - Institutional obligation - Data governance - Surveillance or monitoring - Civil rights or discrimination - Procurement or public-sector adoption - Educational governance - Symbolic governance - Implementation gap - Standards Displacement: whether practical compliance depends heavily on external standards, technical frameworks, auditors, or delegated standard-setting bodies For each present theme, provide evidence and explanation. 7. Discourse-Policy-Practice-Resource Alignment Code the political-economic alignment layer of the document. Do not conduct or imply a full financial audit. Code visible resource signals, budgetary traces, implementation mechanisms, and public investment commitments. Code the following: - Stated AI Purpose: what the government, agency, court, regulator, or policy body says AI is supposed to accomplish - Ethical Justification: which values are invoked, such as equity, safety, rights, learning, innovation, competitiveness, public good, or integrity - Policy Mechanism: whether there is a rule, office, procedure, mandate, grant program, enforcement process, governance structure, compliance mechanism, or implementation plan - Funding / Resource Signal: evidence of money, staffing, contracts, infrastructure, grants, procurement, compliance funding, public-private partnership, or implementation labor - Actual Investment Direction: what appears to be funded, operationalized, regulated, incentivized, bought, built, or enforced - Funding Direction: where public investment, grants, compliance resources, procurement, labor, infrastructure, or institutional attention are directed - Stated Benefit: what AI, policy, regulation, or compliance system is claimed to improve or solve - Claimed Beneficiary: who is said to benefit - Material Beneficiary: who appears to receive value in practice - Cost Bearer: who pays financially, socially, ethically, administratively, or through compliance burden - Profit Receiver / Value Capturer: who may capture contracts, market incentives, data, legitimacy, procurement opportunities, or compliance revenue - Public Good Claim: whether the document frames AI investment as serving education, democracy, equity, rights, safety, public welfare, or society - Market Benefit Claim: whether benefits are framed as innovation, competitiveness, efficiency, productivity, cost-saving, market growth, or national advantage - Redistributive Value: whether resources reduce inequality or mainly create compliance infrastructure or market incentives - Hidden Costs: surveillance, compliance burden, administrative expansion, data extraction, vendor dependency, enforcement gaps, or rights without remedies - Benefit Evidence: whether claimed benefits are evidence-based, speculative, aspirational, rhetorical, symbolic, or absent - Claimed/Funded Benefit Gap: whether stated public-good benefits match the funding, mandates, enforcement capacity, or market incentives described - Alignment Score: 0 no visible alignment; 1 symbolic; 2 procedural; 3 operational; 4 material/substantive; 5 verified substantive alignment; 8 unclear; 9 not applicable - Alignment Justification: explain why the document is not one point higher and not one point lower - Decoupling Signal: whether there is a gap between public-interest language, policy mechanism, and material support or enforcement capacity - Market Capture Signal: low, moderate, or high Identify the dominant value type: - Mechanical - Institutional - Market - Pedagogical - Redistributive - Humanistic Governmental/legal/regulatory-specific focus: Code whether the document funds redistributive public benefit, imposes compliance cost, incentivizes markets, or creates symbolic governance without material enforcement capacity. Critical scoring reminder: Do not raise the Alignment Score solely because the document expresses strong ethical commitments. Ethical-discursive strength and implementation force must be evaluated separately. 8. Symbolic Versus Substantive Governance Assess whether ethical claims are: - Enforceable - Procedural - Compliance-oriented - Aspirational - Symbolic or performative - Substantive intervention - Structural reform - Ambiguous or contradictory 9. Evidence Excerpts Or Paraphrases Provide 5 to 8 short evidence excerpts or careful paraphrases. Explain why each excerpt matters. 10. Emergent Codes List emergent codes, why they emerged, and whether they should remain governmental/legal/regulatory-specific or be considered for the shared codebook. Strong candidates to preserve if repeatedly observed in this corpus include: - Standards Displacement - Open-Source Carve-Out - Sector Carve-Out - FRIA - Enforcement Density - Conformity Assessment Integrity - Compliance Theater or Decoupling-through-Procedure 11. Cross-Domain Comparison Note Explain how this document might compare with forum, journal, and institutional or corporate materials. 12. Methodological Caution Note Identify limits such as: - partial text access - vague implementation - missing enforcement - jurisdictional ambiguity - symbolic compliance - rights without remedies - unclear regulated actors - delegated standards dependence - extraction constraints If Text_Access is Partial, this section must explicitly state how the access limitation affects interpretation. 13. Second-Brain Interpretive Notes Provide a concise future-retrieval block: - Short summary - Major argument or discursive position - Primary ethical conflict - Responsibility attribution - Institutional framing - Temporal orientation - Stakeholder orientation - Key excerpt or anchor passage - Reflexive memo - Integration note 14. Second-Brain Retrieval Metadata Provide: - Search keywords - Major theoretical links, if supported by the evidence - Primary ethical conflict - Domain comparison use - Best quote or anchor passage - Future use tag Recommended theory links include: - Marx - Bourdieu - Freire - Foucault - Berger and Luckmann - Weber - Meyer and Rowan - Billis - Bromley and Powell Use theory only when supported by the evidence. Do not force a theory link if the document does not support it. 15. Final Coding Confidence Give a score from 0 to 5 and explain the reason. Final reminder: - Do not code excluded records. - Do not infer substance from metadata-only or navigation-only files. - Distinguish ethical richness from material alignment. - Stable interpretive differences are acceptable only when they remain text-grounded, explicit, and internally consistent. v2.12 SCORING ANCHORS — GOVERNS ALL *_code SCORING. This block SUPERSEDES any "0-5 ordinal anchors" or bare-range line in the layer prompt's embedded crosswalk ABOVE. Where they conflict on HOW to score, THIS governs (the prompt's key SET still holds; this defines the SCALES + presence gates). ══════════════════════════════════════════════════════════════════ v2.12 SCORING ANCHORS — GOVERNS ALL *_code SCORING (READ AND OBEY) ══════════════════════════════════════════════════════════════════ This block is the v2.12 scoring authority. It SUPERSEDES any "0-5 ordinal anchors" or bare-range text in the layer prompt's embedded crosswalk above, and supersedes v2.11/v2.10. Where they conflict on HOW to score, THIS governs. The injected CANONICAL OUTPUT SCHEMA defines the exact key SET; this block defines the SCALES, presence gates, and (NEW in v2.12) the CATEGORICAL crosswalk legends. Emit exactly one integer per *_code field. NEW IN v2.12 (from the RUN_03 adversarial pass, 2026-06-05; approved 2026-06-08). v2.12 is ADDITIVE — every Part A–G rule below is carried from v2.11 unchanged. The fixes target the nominal/categorical fields that dragged the pooled alpha down WITHOUT touching the eth_* ethical spine (which survived adversarial attack): H1. NEW PART H echoes the integer CATEGORICAL CROSSWALK LEGEND verbatim into these anchors — the legend used to live only in the schema while the decision pressure lived here; they never met. (RUN_03 Finding A/C.) H2. The five nominal-but-numeric fields are HARD-FLAGGED as CATEGORICAL: the integer names a TYPE, NOT a magnitude. Do NOT read higher = more. This alone reverses the highest-leverage RUN_03 error (three readers misread cb_balance magnitude into the wrong category). (Finding C.) H3. DIFFUSE/COLLECTIVE vs NONE-ASSERTED disambiguated on the actor crosswalk: a broad collective ("society", "humanity as a whole") = 6, NOT 0. (Finding B.) C+. int_* fires only on a described OPERATIONAL MECHANISM, never rhetorical metaphor (see Part C). (Finding G.) Process fields (coding_confidence, final_confidence, human_review_flag) are EXCLUDED from reliability alpha — coder signature, not document property (handled in reliability.py; Finding E). A one-5 CAP machine-check runs at intake (Finding D). Neither changes how you code; code normally. LINEAGE — v2.11 redefined eth_integrity to full construct (Part B), operationalized the 5 with a proportion gate + two-sided anchors (Part A), dropped the degenerate presence binaries (Part F), and added mirror calibration. All retained here. ────────────────────────────────────────────────────────────────── PART A — THE 0-5 INTENSITY LADDER (every bare 0-5 field) Applies to: all eth_*_code, all dppra_*_strength_code, all int_*_intensity_code, all cb_*_presence_strength_code, cb_cost_intensity_code, cb_benefit_intensity_code. ────────────────────────────────────────────────────────────────── Score on where the anchor's LITERAL criterion is met — distribution across the document — not on how important the topic feels. Apply the discriminators mechanically. 0 ABSENT — the dimension is not raised in the text at all. 1 INCIDENTAL — appears once, in passing; no elaboration, example, or stake. 2 MENTIONED — named/described in >=1 passage but UNDEVELOPED; no mechanism, consequence, or example attached. 3 DEVELOPED (LOCALIZED) — substantively treated WITH a concrete example, mechanism, or stake, BUT confined to one section or a few adjacent passages; the argument does not keep returning to it. 4 CENTRAL (DISTRIBUTED) — recurs across MOST major sections and the argument repeatedly returns to it, tied to a mechanism/consequence/ recommendation. Removing it would leave several passages unintelligible, BUT the document could be rebuilt around a different spine. 5 DEFINING (SPINE) — pervasive AND load-bearing: the single organizing logic of the text. Remove it and there is no document, only disconnected fragments; it survives no reframing. 8 applicable but unclear · 9 definitionally not applicable THE THREE DISCRIMINATORS (apply in order, mechanically): • 2 vs 3 — Is a concrete mechanism / example / stake attached? NO -> 2. YES -> at least 3. • 3 vs 4 — COUNT the major sections the dimension appears in. One section, or a few adjacent passages -> 3. Most major sections, the argument keeps returning -> 4. • 4 vs 5 — Removal test + PROPORTION GATE (both required for a 5; see below). ►► THE 5 IS OPERATIONALIZED (from v2.11 — apply BOTH tests, not a gut call): A dimension may be coded 5 ONLY IF: (i) PROPORTION GATE — it is the dominant subject of the BODY by section share: it is the leading theme in a PLURALITY of the document's major sections (count them), AND no other dimension occupies more. Read the BODY distribution, NOT the title, abstract, or a single loud passage. (ii) REMOVAL TEST — remove it and the document collapses into disconnected fragments; it survives no reframing. If (i) fails -> the ceiling is 4, full stop. If a rival dimension ties on section share, neither is 5 unless co-equal (see CAP). TWO-SIDED ANCHORS — calibrate BOTH ends to the same bar: • HIGH (earn the 5): a paper whose entire thesis is that privacy law is broken; every major section is about privacy; remove it and nothing remains -> 5. • TRAP (looks like 5, is not — TITLE-ANCHORING): a statute titled "...Transparency... Act" mentions transparency in the title and intro, but its BODY is mostly enforcement duties, impact assessments, and remedies. Section share for transparency is one of several, not the plurality -> transparency is 4 (distributed), NOT 5. Score the body, never the headline word. • COLD MISS (a real 4-or-5 under-coded to 0): "this dimension recurs across the intro, methods, risk analysis AND recommendations, each tied to a mechanism" is AT LEAST a 4 — do not leave a distributed, load-bearing dimension at 0/1 because no single passage shouted. Distribution earns the score. TIE-BREAKER (center-pull): torn between two adjacent anchors -> code the LOWER unless the higher anchor's FULL criterion is explicitly met. (NB: this is a symmetric brake — it pulls a HOT read down toward 4 AND tells a COLD read that once the full 4-criterion IS met, you must commit to the 4, not retreat to 2.) CAP: at most ONE dimension per document may be coded 5, unless two are genuinely co-equal on section share — name both in narrative.cap_rule_notes. (v2.12: an intake validator machine-checks this cap and verifies cap_rule_notes matches the actual count of 5s — so report your 5s honestly in the notes.) ────────────────────────────────────────────────────────────────── PART B — eth_integrity: PRINCIPLED CONSISTENCY + FOLLOW-THROUGH (from v2.11) ────────────────────────────────────────────────────────────────── eth_integrity_code — the HEADLINE integrity principle. Integrity is NOT the mere knowledge of what is right; it is the ALIGNMENT between an actor's professed moral principles/values and its actual or required CONDUCT. Two faces, both count: (a) EPISTEMIC HONESTY — truthfulness / accuracy / authenticity / verification of knowledge and outputs (the actor is honest about what is true); AND (b) PRACTICAL FOLLOW-THROUGH — beliefs, methods, and actions are derived from and consistent with a stated moral framework; the actor DOES what its principles commit it to. The opposite of integrity is HYPOCRISY / decoupling (saying one thing, doing another), not ignorance. Code the DEGREE to which the document EMBODIES, DEMANDS, VERIFIES, or EXPOSES this principle->action consistency. ►► int_* AND decoupling ARE THE EVIDENCE STREAMS (from v2.11) — they no longer "route content away from" integrity; they FEED and UNDERCUT it: • int_system_tevv — testing/eval/audit = verifying conduct matches claims (follow-through evidence -> SUPPORTS integrity). • int_data_prov — honesty about data sourcing/consent/lineage (honesty-of-inputs evidence -> SUPPORTS integrity). • int_epistemic — truthfulness/reliability of outputs (epistemic-honesty evidence -> SUPPORTS integrity). • int_academic — learner honesty (supports integrity in education docs). • decoupling_signal — a documented say-do gap is the INVERSE: it LOWERS the actor's integrity. (If the document is EXPOSING or CONDEMNING the gap, then integrity-as-DEMAND is high while the actor's integrity is shown FAILING — code the document's stance, and say which in cap_rule_notes.) Still score each int_* / decoupling field at its OWN grain (Part A intensity). eth_integrity is the holistic headline judgment those streams inform; it is NOT a sum of them and NOT capped below its strongest supporting stream. GUARDRAIL (keeps integrity from re-becoming a catch-all — the old failure mode): • Code eth_integrity >=2 ONLY if the text raises consistency-of-principle-and- conduct OR honesty/accuracy of knowledge — not merely because ethics, values, or "responsible AI" are discussed. "Serious" / "about AI ethics" is NOT integrity. • A document that merely STATES values (a values list, a principles preamble) with no binding to conduct = LOW integrity signal (1-2). • A document that BINDS values to conduct — enforcement, verification, audit-with-teeth, follow-through obligations — OR that DOCUMENTS/EXPOSES the breach between stated values and practice = HIGH (4-5 if it is the spine). • Reserve 5 for documents whose ORGANIZING AXIS is principle<->action consistency itself (its honoring, its demand, its verification, or its exposure), per the Part A proportion gate + removal test. ────────────────────────────────────────────────────────────────── PART B2 — PRESENCE TRIGGERS for the other gated principles (from v2.10) ────────────────────────────────────────────────────────────────── eth_autonomy_code — human self-determination and control. Code >=2 ONLY if the text raises human control over decisions, manipulation/ nudging, dependency/over-reliance/deskilling, OR informed consent. Not present merely because humans are discussed. eth_nonmaleficence_code — direct harm and harm-prevention. Code >=2 ONLY if a concrete harm or harm-risk to identifiable parties is named (violence, injury, surveillance harm, psychological damage). Generic "risk" with no named harm = 1. This is DIRECT harm (vs justice = distributional, privacy = data). eth_justice_code — distributive/procedural fairness, rights, remedy, due process. Code >=2 ONLY if the text raises unequal distribution of benefits/burdens across groups, rights/remedy/due-process, OR structural fairness. DISAMBIGUATION: equity = who gets ACCESS/INCLUSION; justice = fairness of DISTRIBUTION + rights/remedy. Access gap -> equity. Unfair outcome / no remedy -> justice. (The other six principles keep v2.7 behavior + Part A for intensity. Privacy's implicit trigger: is identifiable data at stake?) ────────────────────────────────────────────────────────────────── PART C — int_* INTEGRITY SUB-DIMENSIONS: scope (each takes Part A for intensity) CANONICAL SIX (int_institutional DROPPED in v2.8 — do NOT code it). ────────────────────────────────────────────────────────────────── int_academic_* — academic integrity: honesty OF THE LEARNER (student cheating, plagiarism, contract cheating in education). int_research_* — research integrity: fabrication, falsification, misconduct, reproducibility, p-hacking. int_assessment_* — assessment integrity: whether a FORM OF ASSESSMENT is still productive or definitive; fires when an assessment's validity/ usefulness collapses (e.g. an essay AI can write no longer measures learning). The instrument's integrity, not the person's honesty (that is int_academic). int_system_tevv_* — system TEVV integrity: whether AI systems are tested, evaluated, validated, verified, audited, benchmarked. int_data_prov_* — data provenance integrity: sourcing, consent, licensing, lineage of data used/trained on (vs privacy = identifiable- person exposure). int_epistemic_* — epistemic integrity: truthfulness/reliability of knowledge, hallucination, misinformation, source verification. ►► OPERATIONAL-MECHANISM REQUIREMENT (NEW in v2.12 — RUN_03 Finding G): int_system_tevv, int_data_prov, int_epistemic, and surveillance_monitoring fire ONLY on a described OPERATIONAL MECHANISM — an actual test/audit/eval, a data-lineage/consent/licensing practice, a verification step, or a monitoring system. They do NOT fire on rhetorical METAPHOR or evocative imagery ("who is being harvested", "ask what you are training", "the machine watches"). A rhetorical essay that gestures at surveillance without naming a monitoring mechanism scores these LOW/0 — code the mechanism, not the mood. (FRM/059: Codex read an essay's metaphors as substantive TEVV/data-provenance — do not.) RELATION TO eth_integrity (v2.11): these six are the EVIDENCE STREAMS for the eth_integrity headline (Part B). Score each at its own grain; eth_integrity is the holistic consistency+follow-through judgment they inform, NOT a routing destination that empties the headline. ────────────────────────────────────────────────────────────────── PART D — cb_*_evidence_label_code DECISION TREE (from v2.10) ────────────────────────────────────────────────────────────────── Is a cost/benefit claim made? no -> 0 absent Backed by data, study, or a cited concrete instance? yes -> 1 evidence-based A future/hypothetical/hoped-for outcome? yes -> 2 speculative/aspirational Asserted as value/branding with no support or mechanism? yes -> 3 rhetorical/symbolic Between 1 and 2: PAST/PRESENT fact -> 1 ; FUTURE outcome -> 2. Between 2 and 3: a mechanism/pathway is described -> 2 ; pure assertion -> 3. ────────────────────────────────────────────────────────────────── PART E — COST/BENEFIT INDICATOR SET (§3E collapse; 19 indicators, from v2.10) Code ONLY these indicators (0 absent · 1 present · 8 · 9). The cut ones are gone. ────────────────────────────────────────────────────────────────── BENEFIT (10): cb_benefit_access_equity_code, cb_benefit_accessibility_assistive_code, cb_benefit_efficiency_labor_saving_code, cb_benefit_financial_cost_saving_code, cb_benefit_learning_pedagogical_code, cb_benefit_market_profit_code, cb_benefit_public_good_code, cb_benefit_quality_accuracy_code (ALSO covers AI-assisted integrity/cheating detection), cb_benefit_rights_protection_code, cb_benefit_consistency_standardization_code. CUT: cb_benefit_institutional_performance_code, cb_benefit_integrity_detection_code. COST (9): cb_cost_access_inequality_code, cb_cost_compliance_burden_code, cb_cost_credential_opportunity_code, cb_cost_financial_code (ALSO covers labor/time cost), cb_cost_privacy_data_code, cb_cost_surveillance_control_code, cb_cost_environmental_infrastructure_code, cb_cost_cognitive_psychological_code (cognitive/learning/psychological harm to individuals — absorbs the old cognitive_pedagogical + psychological_social), cb_cost_rights_remedy_gap_code — code present ONLY when a SPECIFIC rights/remedy/ due-process gap is named, NOT on generic fairness language. CUT: cb_cost_legitimacy_trust_code, cb_cost_labor_time_code, cb_cost_cognitive_pedagogical_code (merged), cb_cost_psychological_social_code (renamed -> cb_cost_cognitive_psychological_code). ────────────────────────────────────────────────────────────────── PART F — DEPRECATED PRESENCE BINARIES (from v2.11) ────────────────────────────────────────────────────────────────── The following presence binaries were ~100% constant across Run-1+Run-2 (no variance, undefined reliability) and are DROPPED. Presence is now read off the paired *_strength field: strength 0 = absent, 1-5 = present at that intensity. • dppra_d_presence_code, dppra_p1_presence_code, dppra_p2_presence_code, dppra_r_presence_code, dppra_a_presence_code • cb_cost_presence_binary_code, cb_benefit_presence_binary_code DO emit the *_strength fields as before. The DPPRA D/P1/P2/R/A structure and the cost/benefit constructs are UNCHANGED — only the redundant presence flags are gone. (Schema-compat note: if the injected schema still lists these keys, emit 9 = N/A; the compiler treats v2.12 strength as authoritative for presence.) ────────────────────────────────────────────────────────────────── PART G — OUTPUT FORMAT (ALL CODERS — the compiler rejects violations) ────────────────────────────────────────────────────────────────── These are hard requirements for every coder, not style preferences. An output that breaks them is rejected and must be re-run. 1. PRESERVE UNDERSCORES in every key exactly as in the schema: emit `eth_equity_code`, NOT `ethequitycode` / `ethEquityCode` / `eth equity code`. Do not strip, camelCase, or collapse separators. 2. EMIT EVERY KEY in the injected schema, including N/A fields. Cross-layer fields that do not apply get 9 (or 99 for assessment_object_code) — never silently omit a key. 3. USE THE v2.12 KEY SET ONLY. Do not resurrect cut keys (cb_cost_labor_time_code, cb_benefit_integrity_detection_code, cb_cost_legitimacy_trust_code, the old cognitive_pedagogical / psychological_social splits, etc.) and do not emit the dropped *_presence binaries (Part F). Emit the *_strength fields. 4. VALID JSON. Escape every double-quote inside a string value as \". Never leave a raw quote inside an array-of-string element (e.g. evidence excerpts) — one unescaped quote breaks the whole record. 5. FLAT BLOCKS: meta / codes / narrative, no grouped sub-objects, no invented fields. Exactly one integer per *_code field. Set meta.coder and meta.prompt_version = "v2.12". 6. DO NOT return a metadata-only exclusion when a full source text is attached. If the document is present, code it; reserve exclusion for empty/unreadable input. ────────────────────────────────────────────────────────────────── PART H — CATEGORICAL CROSSWALK LEGEND (verbatim, integer-locked) — NEW in v2.12 ────────────────────────────────────────────────────────────────── ►►► CATEGORICAL — the integer names a TYPE, not a strength. Do NOT read higher = more. Pick the label whose meaning matches the text; if two fit, pick the dominant; if none is asserted, use 0; if asserted-but-unclear, use 8. These fields are NOT on the Part A 0-5 intensity ladder. ◄◄◄ These legends are integer-locked (carried from DPPRA_v2.8_INTEGER_CROSSWALK_LOCK and the schema). They are reproduced HERE so the decision pressure and the legend meet in one place — the single highest-leverage v2.12 fix. dominant_value_type_code (which value-frame organizes the document): 1 mechanical — efficiency/automation/technical capability as the point 2 institutional— org duty, governance, standards, compliance, accountability 3 market — profit, competition, commercial positioning, product selling 4 pedagogical — teaching/learning value, educational purpose 5 redistributive— fairness/equity/redistribution of benefits or burdens 6 humanistic — intrinsic human dignity, wellbeing, care, flourishing 8 unclear · 9 N/A ADJACENT-CATEGORY DISCRIMINATION (2 vs 3 vs 6 — splits most on COR/safety docs): • 2 institutional = appeals to organizational duty / governance / standards. • 3 market = profit / competition / commercial positioning. • 6 humanistic = intrinsic human dignity / wellbeing / care. A corporate safety doc grounded in duty-of-care reads 2; a product page selling "responsibility" reads 3; a "flourishing/empathy" essay reads 6. responsibility_concentration_code (where accountability sits): 0 none/not-asserted · 1 concentrated/non-delegable · 2 distributed 3 diffuse/unclear · 8 · 9 ACTOR CROSSWALK — cb_primary_cost_bearer_code, cb_primary_material_beneficiary_code, cb_primary_value_capturer_code (WHO bears the cost / gets the benefit / captures the value): 0 none — the dimension is NOT asserted at all 1 student/learner/applicant 2 teacher/evaluator/worker 3 institution/school/employer 4 platform/vendor/company 5 government/policymaker 6 public/community/society 7 environment/infrastructure 8 asserted but the actor is genuinely unclear · 9 N/A DIFFUSE/COLLECTIVE vs NONE (NEW — RUN_03 Finding B): • A named BROAD COLLECTIVE ("society", "humanity as a whole", "the public", "everyone") = 6 public/community/society — NOT 0. • 0 = the cost/benefit dimension is not asserted at all. • 8 = a cost/benefit IS asserted but the bearer/beneficiary is unclear. (FRM/058: a narrative naming "humanity as a whole" as beneficiary must code 6, not 0 — do not zero a diffuse-but-named actor.) cb_balance_code (the net cost/benefit framing — a CATEGORY, not a magnitude): 0 none · 1 benefits>costs · 2 costs>benefits · 3 mixed/balanced 4 asymmetric (costs and benefits fall on DIFFERENT parties) · 8 · 9 WARNING: 4 is NOT "very benefit-heavy". A benefit-heavy page is 1, not 4. Use 4 ONLY when the costs and benefits accrue to DIFFERENT actors (asymmetric distribution). (RUN_03: a benefit-heavy product page is cb_balance=1; coding 4 by reading "4 = strong" was the single highest-leverage error.) cb_distribution_direction_code (which way value flows): 0 none · 1 redistributive/public · 2 institutional efficiency 3 market/platform capture · 4 cost-shift to learners/workers/public 5 mixed/contested · 8 · 9 ══════════════════════════════════════════════════════════════════ END v2.12 SCORING ANCHORS ══════════════════════════════════════════════════════════════════ CANONICAL OUTPUT SCHEMA — emit EXACTLY this key set, flat 'codes' block (meta / codes / narrative); do not group, rename, or omit keys. STRICT KEY COMPLIANCE (all layers): use the key names BELOW verbatim, WITH underscores (e.g. `eth_equity_code`, NOT `ethequitycode`). Do NOT invent fields from the layer prompt's narrative structure (no `resp_*`, no `struct_*`, no per-principle valence/substance, no `*_memo` fields) — code ONLY the keys in this schema. The schema, not the prompt's prose, defines the exact output keys. { "_schema_version": "v2.12", "_schema_note": "v2.12 = the v2.11 key set UNCHANGED IN SHAPE (additive guidance only). The categorical legends are echoed into the v2.12 inject Part H and flagged CATEGORICAL (type not magnitude); int_* fires on an operational mechanism not metaphor; confidence/process fields excluded from alpha; one-5 CAP machine-checked at intake. No key added/renamed/removed vs v2.11. See DPPRA_v2.12_CANDIDATE_CHANGES_from_RUN03.md.", "_schema_info": { "schema_name": "DPPRA_coder_record (all layers)", "schema_version": "2.8", "defined_by": "claude (canonical reference, per project lead)", "purpose": "All coders (claude, gpt, pxy) MUST emit this exact key set so inter-coder reliability can be computed mechanically on the 'codes' block. Do not add, rename, or omit keys. Leave value types as specified. Populate every field; never leave a shared-core code blank.", "reliability_layer": "Inter-coder reliability (kappa/alpha) is computed ONLY on the integer fields in 'codes'. valence_* are proportions (0.00-1.00) and are handled separately. The 'narrative' and 'meta' blocks are for human/adversarial validation, not numeric reliability.", "missing_value_convention": "0 = applicable but absent/unsupported; 8 = applicable but unclear/insufficient evidence; 9 = definitionally not applicable. Every *_code field contains exactly one integer and nothing else.", "v2.8_scoring_authority": "All bare 0-5 fields are now anchored by the v2.8 SCORING ANCHORS block (Part A 0-5 ladder + tie-breaker + CAP; Part B presence triggers for integrity/autonomy/nonmaleficence/justice; Part C int_* scopes; Part D evidence-label tree), injected into every coder paste and embedded in the operational profile. The anchors define the SCALE; this schema defines the KEY SET. Where a bare '0-5 ordinal anchors' line in any embedded crosswalk conflicts, the v2.8 anchors govern.", "v2.8_field_set_changes": "v2.8 changes the field SET in two sanctioned places only (the documented exception to the keys-must-not-change guard). (A) §3C int_institutional DROPPED: int_institutional_presence_code + int_institutional_intensity_code REMOVED; org-level stated-values-vs-practice is captured by decoupling_signal_code. Canonical SEVEN integrity sub-dimensions -> SIX (academic, research, assessment, system_tevv, data_prov, epistemic). int_assessment re-scoped to 'a form of assessment no longer productive/definitive' (instrument validity, not learner honesty). (B) §3E cost/benefit indicators 24 -> 19: CUT cb_benefit_institutional_performance_code, cb_cost_legitimacy_trust_code, cb_cost_labor_time_code (degenerate constants, fire on ~90-100% of docs, alpha~0); MERGE cb_cost_cognitive_pedagogical_code -> cb_cost_cognitive_psychological_code (renamed from cb_cost_psychological_social_code, now absorbs cognitive/learning harm); MERGE cb_benefit_integrity_detection_code -> cb_benefit_quality_accuracy_code (now also covers AI-assisted integrity/cheating detection); cb_cost_financial_code now also covers labor/time cost; TIGHTEN cb_cost_rights_remedy_gap_code (present ONLY when a SPECIFIC rights/remedy/due-process gap is named). All OTHER keys hold.", "v2.8_leave_alone_fields": "Do NOT re-anchor these reliable v2.7 fields: eth_privacy_code, dppra_p1_strength_code, eth_equity_code, cb_primary_value_capturer_code, dppra_alignment_score_code, cb_benefit_consistency_standardization_code, journal_method_type_code.", "RESOLVED_v2.6_crosswalk_lock": "The previously open crosswalks (decoupling_signal_code, market_capture_signal_code, dominant_value_type_code, responsibility_concentration_code, cb_*_evidence_label_code, the cb actor crosswalk, cb_balance_code, cb_distribution_direction_code, and all gov_*_code mappings) are integer-locked in the v2.8 crosswalk lock (carried from DPPRA_v2.6_INTEGER_CROSSWALK_LOCK.md, approved 2026-05-25). All coders MUST use those integers.", "RESOLVED_2026-05-27_JRN_domain_lock": "The six journal_*_code fields are integer-locked in DPPRA_JRN_INTEGER_CROSSWALK_LOCK_v2.7.md (approved 2026-05-27). JRN coders code them per that lock; GOV/FRM/INS code them 9 (definitionally N/A), exactly as journal coders code the gov_* fields 9.", "RESOLVED_2026-05-27_FRM_domain_lock": "The eight forum domain fields (learner_posture, assessment_object, external_link_boundary, detection_verification presence+strength, surveillance_monitoring presence+strength, overall_dominant_valence) are integer-locked in DPPRA_FRM_INTEGER_CROSSWALK_LOCK_v2.7.md (approved 2026-05-27). FRM coders code them per that lock; non-forum layers code 9 (assessment_object: 99). NOTE assessment_object_code uses 88=unclear/99=N/A (8 and 9 are real categories). INS domain-specific fields remain to be locked at that layer's first run." }, "meta": { "stable_source_id": "string (assigned by manifest)", "document_id": "string", "title": "string (exact)", "authoring_body": "string", "date": "string YYYY-MM-DD", "domain": "string", "domain_code": "int (1=gov/legal/regulatory, 2=journal, 3=forum, 4=institutional/corporate) [LOCK]", "subdomain": "string", "time_segment": "string", "corpus_batch": "string", "source_type": "string", "jurisdiction": "string", "intended_audience": "string", "regulated_actors": "string", "coder": "string: one of [claude, codex, gemini]", "prompt_version": "string", "coding_date": "string YYYY-MM-DD", "text_access": "string: Full|Partial|Metadata-only|Navigation-only|Extraction-failed", "text_access_code": "int (1=Full, 2=Partial, 3=Metadata-only, 4=Navigation-only, 5=Extraction-failed) [LOCK]", "manifest_match_confirmed": "string: YES|NO|DISCREPANCY[explain]" }, "codes": { "_principle_intensity_range": "0-5 per v2.8 SCORING ANCHORS Part A ladder (0 absent,1 incidental,2 mentioned,3 developed,4 central,5 defining); 8 unclear; 9 N/A. CAP: at most one 5 per document unless co-equal (name both in cap_rule_notes).", "coding_confidence_code": "int 0-5", "final_confidence_code": "int 0-5", "human_review_flag_code": "int 0=no review needed, 1=review recommended", "eth_equity_code": "int 0-5|8|9 [LEAVE-ALONE: reliable v2.7 field; Part A ladder]", "eth_accountability_code": "int 0-5|8|9", "eth_transparency_code": "int 0-5|8|9", "eth_privacy_code": "int 0-5|8|9 [LEAVE-ALONE: reliable v2.7 field; implicit trigger: is identifiable data at stake?]", "eth_autonomy_code": "int 0-5|8|9 [Part B trigger: human control/manipulation/dependency/consent]", "eth_integrity_code": "int 0-5|8|9 [Part B trigger: accuracy/authenticity/verification of knowledge]", "eth_nonmaleficence_code": "int 0-5|8|9 [Part B trigger: named concrete harm to identifiable parties]", "eth_beneficence_mech_code": "int 0-5|8|9", "eth_beneficence_eth_code": "int 0-5|8|9", "eth_justice_code": "int 0-5|8|9 [Part B trigger: distribution/rights/remedy; access gap->equity, unfair outcome/no remedy->justice]", "valence_positive": "float 0.00-1.00 (positive+neutral+negative should sum to 1.00 when codable)", "valence_neutral_mixed": "float 0.00-1.00", "valence_negative": "float 0.00-1.00", "dppra_d_strength_code": "int 0-5|8|9 [Part A ladder]", "dppra_p1_strength_code": "int 0-5|8|9 [LEAVE-ALONE: reliable v2.7 field; Part A ladder]", "dppra_p2_strength_code": "int 0-5|8|9 [Part A ladder]", "dppra_r_strength_code": "int 0-5|8|9 [Part A ladder]", "dppra_a_strength_code": "int 0-5|8|9 [Part A ladder]", "dppra_alignment_score_code": "int 0-5|8|9 (0 none,1 symbolic,2 procedural,3 operational,4 material/substantive,5 verified substantive) [LEAVE-ALONE: reliable v2.7 field, keeps its own anchors]", "decoupling_signal_code": "int [LOCK crosswalk] 0=none,1=low,2=low-moderate,3=moderate,4=high,8=unclear,9=N/A (now also carries org-level stated-values-vs-practice, the dropped int_institutional concept)", "market_capture_signal_code": "int [LOCK crosswalk] 0=none,1=low,2=moderate,3=high,8=unclear,9=N/A", "dominant_value_type_code": "int [LOCK crosswalk] 1=mechanical,2=institutional,3=market,4=pedagogical,5=redistributive,6=humanistic,8=unclear,9=N/A", "responsibility_concentration_code": "int [LOCK crosswalk] 0=none,1=concentrated/non-delegable,2=distributed,3=diffuse/unclear,8=unclear,9=N/A", "int_academic_presence_code": "int 0|1|8|9 [scope: honesty OF THE LEARNER - student cheating/plagiarism/contract cheating]", "int_academic_intensity_code": "int 0-5|8|9 [Part A ladder]", "int_research_presence_code": "int 0|1|8|9 [scope: fabrication/falsification/misconduct/reproducibility/p-hacking]", "int_research_intensity_code": "int 0-5|8|9 [Part A ladder]", "int_assessment_presence_code": "int 0|1|8|9 [scope: a FORM OF ASSESSMENT no longer productive/definitive - instrument validity, not learner honesty]", "int_assessment_intensity_code": "int 0-5|8|9 [Part A ladder]", "int_system_tevv_presence_code": "int 0|1|8|9 [scope: AI tested/evaluated/validated/verified/audited/benchmarked]", "int_system_tevv_intensity_code": "int 0-5|8|9 [Part A ladder]", "int_data_prov_presence_code": "int 0|1|8|9 [scope: sourcing/consent/licensing/lineage of data used/trained on]", "int_data_prov_intensity_code": "int 0-5|8|9 [Part A ladder]", "int_epistemic_presence_code": "int 0|1|8|9 [scope: truthfulness/reliability of knowledge, hallucination, misinformation, source verification]", "int_epistemic_intensity_code": "int 0-5|8|9 [Part A ladder]", "cb_cost_presence_strength_code": "int 0-5|8|9 [Part A ladder]", "cb_benefit_presence_strength_code": "int 0-5|8|9 [Part A ladder]", "cb_cost_intensity_code": "int 0-5|8|9 [Part A ladder]", "cb_benefit_intensity_code": "int 0-5|8|9 [Part A ladder]", "cb_cost_evidence_label_code": "int [Part D tree] 0=absent,1=evidence-based,2=speculative/aspirational,3=rhetorical/symbolic,8=unclear,9=N/A", "cb_benefit_evidence_label_code": "int [Part D tree] same as cb_cost_evidence_label_code", "cb_primary_cost_bearer_code": "int actor crosswalk 0-9 (0 none,1 student/learner/applicant,2 teacher/evaluator/worker,3 institution/school/employer,4 platform/vendor/company,5 government/policymaker,6 public/community/society,7 environment/infrastructure,8 unclear,9 N/A)", "cb_primary_material_beneficiary_code": "int actor crosswalk 0-9", "cb_primary_value_capturer_code": "int actor crosswalk 0-9 [LEAVE-ALONE: reliable v2.7 field]", "cb_balance_code": "int 0 none,1 benefits>costs,2 costs>benefits,3 mixed/balanced,4 asymmetric,8 unclear,9 N/A", "cb_distribution_direction_code": "int 0 none,1 redistributive/public,2 institutional efficiency,3 market/platform capture,4 cost shift to learners/workers/public,5 mixed/contested,8 unclear,9 N/A", "cb_cost_financial_code": "int 0|1|8|9 (NOW ALSO covers labor/time cost - absorbed the cut cb_cost_labor_time)", "cb_cost_cognitive_psychological_code": "int 0|1|8|9 (cognitive/learning/psychological harm to individuals - merged cb_cost_cognitive_pedagogical + renamed from cb_cost_psychological_social)", "cb_cost_credential_opportunity_code": "int 0|1|8|9", "cb_cost_privacy_data_code": "int 0|1|8|9", "cb_cost_surveillance_control_code": "int 0|1|8|9", "cb_cost_access_inequality_code": "int 0|1|8|9", "cb_cost_environmental_infrastructure_code": "int 0|1|8|9", "cb_cost_compliance_burden_code": "int 0|1|8|9", "cb_cost_rights_remedy_gap_code": "int 0|1|8|9 [TIGHTENED: present ONLY when a SPECIFIC rights/remedy/due-process gap is named, NOT on generic fairness language]", "cb_benefit_learning_pedagogical_code": "int 0|1|8|9", "cb_benefit_efficiency_labor_saving_code": "int 0|1|8|9", "cb_benefit_financial_cost_saving_code": "int 0|1|8|9", "cb_benefit_access_equity_code": "int 0|1|8|9", "cb_benefit_quality_accuracy_code": "int 0|1|8|9 (NOW ALSO covers AI-assisted integrity/cheating detection - absorbed the cut cb_benefit_integrity_detection)", "cb_benefit_consistency_standardization_code": "int 0|1|8|9 [LEAVE-ALONE: reliable v2.7 field]", "cb_benefit_market_profit_code": "int 0|1|8|9", "cb_benefit_public_good_code": "int 0|1|8|9", "cb_benefit_accessibility_assistive_code": "int 0|1|8|9", "cb_benefit_rights_protection_code": "int 0|1|8|9", "gov_binding_status_code": "int [LOCK] 0=nonbinding,1=binding,2=partial/mixed,8=unclear,9=N/A", "gov_enforcement_density_code": "int 0 none,1 weak/indirect,2 single strong or multiple weak,3 multi-level architecture,8 unclear,9 N/A", "gov_fria_present_code": "int [LOCK] 0=no,1=yes,2=partial,9=N/A", "gov_opensource_carveout_code": "int [LOCK] 0=no,1=yes,2=partial,9=N/A", "gov_sector_carveout_code": "int [LOCK] 0=no,1=yes,2=partial,9=N/A", "gov_standards_displacement_code": "int [LOCK] 0=absent,1=present,8=unclear,9=N/A", "gov_remedy_present_code": "int [LOCK] 0=no,1=yes (administrative),2=yes (justiciable/external),8=unclear,9=N/A", "gov_sanction_present_code": "int [LOCK] 0=no external sanction,1=external sanction present,8=unclear,9=N/A", "gov_compliance_mechanism_code": "int [LOCK] 0=absent,1=present,8=unclear,9=N/A", "journal_source_type_code": "int [JRN LOCK v2.7] 1=empirical,2=conceptual,3=review,4=theoretical,5=methodological,6=policy-oriented,7=mixed,8=unclear,9=N/A (non-journal layers code 9)", "journal_empirical_basis_binary_code": "int [JRN LOCK v2.7] 0=no empirical basis,1=has empirical basis,8=unclear,9=N/A (non-journal layers code 9)", "journal_method_type_code": "int [JRN LOCK v2.7] 0=none/non-empirical,1=quantitative,2=qualitative,3=mixed-methods,4=conceptual/theoretical,5=review/meta-analysis,6=computational/simulation,8=unclear,9=N/A (non-journal layers code 9) [LEAVE-ALONE: reliable v2.7 field]", "journal_method_evidence_label_code": "int [JRN LOCK v2.7] 0=absent,1=empirical,2=experiential/anecdotal,3=theoretical,4=speculative/prospective,5=rhetorical/promotional,6=mixed,8=unclear,9=N/A (non-journal layers code 9)", "journal_evidence_quality_strength_code": "int [JRN LOCK v2.7] 0=absent,2=weak,3=moderate,5=strong,8=unclear,9=N/A (non-journal layers code 9)", "journal_level_of_analysis_code": "int [JRN LOCK v2.7] 0=other/emergent,1=classroom,2=institution,3=platform,4=state,5=labor market,6=public sphere,7=social structure,8=unclear,9=N/A (non-journal layers code 9)", "learner_posture_code": "int [FRM LOCK v2.7] 1=AI-as-tutor,2=shortcut,3=collaborator,4=replacement-thinker,5=assistive-scaffold,6=cheating-mechanism,7=mixed,8=unclear,9=N/A (non-forum layers code 9)", "assessment_object_code": "int [FRM LOCK v2.7] 0=none,1=essay,2=exam,3=grade,4=degree,5=classroom-assignment,6=admission,7=scholarship,8=job-application,9=research-funding,10=professional-credential,11=legal-document,12=curriculum,13=public-infrastructure,14=labor-system,15=governance-process,16=other; 88=unclear,99=N/A (SPECIAL missing codes; non-forum layers code 99)", "external_link_boundary_code": "int [FRM LOCK v2.7] 0=no external object,1=referenced & included in coded text,2=referenced but NOT included,8=unclear,9=N/A (non-forum layers code 9)", "detection_verification_presence_binary_code": "int [FRM LOCK v2.7] 0=absent,1=present,8=unclear,9=N/A (non-forum layers code 9)", "detection_verification_presence_strength_code": "int [FRM LOCK v2.7] 0-5,8,9 (non-forum layers code 9) [Part A ladder]", "surveillance_monitoring_presence_binary_code": "int [FRM LOCK v2.7] 0=absent,1=present,8=unclear,9=N/A (non-forum layers code 9)", "surveillance_monitoring_presence_strength_code": "int [FRM LOCK v2.7] 0-5,8,9 (non-forum layers code 9) [Part A ladder]", "overall_dominant_valence_code": "int [FRM LOCK v2.7] 1=positive,2=neutral/mixed,3=negative,4=contradictory/polarized,8=unclear,9=N/A (non-forum layers code 9)" }, "narrative": { "brief_summary": "string 150-250 words", "major_argument": "string", "primary_ethical_conflict": "string", "responsibility_attribution": "string", "institutional_framing": "string", "temporal_orientation": "string", "stakeholder_orientation": "string", "enforcement_density_rationale": "string", "alignment_justification": "string: explain why not one point higher and not one point lower", "cap_rule_notes": "string: justify any 5s; state 'CAP OVERRIDE - [rule] - evidence: [quote]' if a cap is overridden", "evidence_excerpts": "array of 5-8 strings (short paraphrase or <15-word quote), each with why it matters", "emergent_codes": "array of strings", "theory_links": "string (evidence-supported only; do not force)", "methodological_caution": "string (if Partial access, state how it affects interpretation)", "future_use_tag": "string of hashtags", "reconciliation_flags": "array of strings: anticipated inter-coder disagreement points and any inferred crosswalks used" } } CODER-SPECIFIC WATCHOUT — applies to Codex ONLY (calibration; obey it). Set meta.coder = "codex". # Coder Calibration Watchout — Codex (v2.12) Neutral calibration guidance for this coder. Apply where text-grounded; it does not override the source or the schema. ## Mirror calibration — YOUR POLE: OVER-COMMIT (hot) UPDATE (supersedes the v2.9 "no systematic high-anchor bias" note): the Run-2 three-coder data shows a **measured, systematic high-anchor offset.** You assigned a spine-5 on 21 of 22 documents (vs Claude on 6), and your eth_integrity was >= Claude's on 20 of 22. You are the hot pole. This watchout reins it in — symmetrically, not into the cold pole. The medication: - **Before coding ANY dimension 5, apply the v2.11 PROPORTION GATE.** A 5 requires that the dimension be the leading theme in a PLURALITY of the document's major sections (count them, read the BODY not the title/abstract) AND pass the removal test. If the gate fails, the ceiling is 4 — full stop. Most documents have NO dimension at 5. - Do not let a single loud passage, an executive-summary flourish, or the document's TITLE drive a 5. Title-anchoring (e.g. a "...Transparency... Act" whose body is mostly enforcement) is a trap — score the body distribution. - "Important-feeling" is not 5. The ladder is countable: mechanism attached? (2 vs 3) count of major sections? (3 vs 4) proportion gate + removal test? (4 vs 5). You are NOT being asked to suppress real 5s (that is Claude's error). Apply the gate honestly in BOTH directions — but your standing risk is the high side. ## HOT-CODER ADDENDUM — top-code deflection rules (Codex self-authored, 2026-06-04) Codex has a known tendency to over-recognize ethical principles and assign high intensity scores from broad, aspirational, or review-style language. To correct this, apply the following dampening rules before finalizing any score. 1. **Treat thematic mention as insufficient for high intensity.** If a document names a principle, lists it among many concerns, or frames it as a general benefit/risk, code it low-to-moderate unless the source develops it with specific mechanisms, affected parties, tradeoffs, remedies, governance duties, or concrete examples. 2. **Reserve 5 for document-defining constructs only.** A 5 requires that the construct is central to the document's main argument, repeatedly developed, and supported by specific reasoning or operational detail. If important but not document-defining, assign 4. If present mainly as one item in a broad survey, assign 2 or 3. 3. **Before assigning any 5, ask:** would a stricter coder agree that this is the primary thing the document is about? If not, do not assign 5. 4. **Apply a soft ceiling to broad reviews / descriptive overviews.** For non-empirical, review-style, conceptual, or advocacy summaries, most principle scores should usually top out at 3 unless the paper gives sustained, construct-specific analysis. A broad "AI has many benefits and risks" article should not receive multiple 4s or 5s simply because many values are mentioned. 5. **Do not stack adjacent high scores.** If two neighboring constructs both seem high, identify the one that is truly primary and reduce the adjacent/secondary one by a level unless both are independently and substantially developed. Especially: equity vs justice · accountability vs transparency · integrity vs transparency · beneficence_mech vs beneficence_eth · privacy vs autonomy. 6. **Cut aspirational governance language down a level.** "There should be regulation," "ethical frameworks are needed," "stakeholders must be responsible," "transparency is important" → usually 2 or 3, not 4 or 5, unless the document specifies binding duties, institutional mechanisms, enforcement, audits, remedies, or role-specific obligations. 7. **Separate value presence from evidence quality.** Do not raise ethical or DPPRA strength scores merely because the document sounds normatively confident. If evidence is secondary, generalized, or citation-based without analysis, keep strength moderate unless the argument itself is unusually focused and specific. 8. **Use 4 as the normal upper score for strong-but-general treatment.** When a principle is clearly important but the discussion stays conceptual, descriptive, or non-operational, prefer 4 over 5. Use 5 only when lowering it would misrepresent the document's core purpose. 9. **Default ambiguous high scores downward.** Between 3 and 4, choose 3 unless there is sustained development. Between 4 and 5, choose 4 unless the construct dominates the document. 10. **Final hot-coder check.** Before output, scan all 4s and 5s. For each, require a concrete reason from the source text. If the reason is only "the document mentions this value" or "this value is implied by the topic," lower the score by one level. > NB (anti-overcorrection): these are dampeners against YOUR measured hot offset, not a push toward Claude's cold pole. A dimension that genuinely recurs across most major sections, each tied to a mechanism, is still a real 4 — rules 4 and 9 have explicit "unless sustained/substantial development" escape valves. Apply them; do not floor a distributed, load-bearing construct. ## eth_integrity (v2.11 — REDEFINED; this is where your offset bites hardest) Integrity is **principled consistency + follow-through** — alignment between professed values and actual/required conduct (epistemic honesty AND follow-through). int_* and decoupling are **evidence streams** feeding/undercutting the headline. - Do NOT double-count: if testing/TEVV content drives the integrity reading, that intensity belongs in `int_system_tevv` at its own grain; eth_integrity is the HOLISTIC consistency judgment, not a re-sum of the sub-dimensions. Score each stream once, at its own grain. - Critically: a document that merely STATES values (a principles preamble, a "responsible AI" values list) with no binding to conduct is **LOW integrity (1-2)** — even if it is earnest and entirely about AI ethics. Reserve high eth_integrity for documents that BIND values to verified conduct or EXPOSE the say-do breach. - "Serious" / "about AI ethics" is NOT integrity. This is your most frequent over-code. ## Justice vs. Equity Do not inflate Justice because a source discusses equity, inclusion, access, accessibility, diversity, or bias mitigation. Use Equity for access/inclusion/bias-mitigation/accessibility/digital-divide. Justice rises above 2 only for rights, due process, contestability, appeal, remedy, redress, anti-discrimination enforcement, structural inequality, burden/benefit distribution, labor justice, democratic harms, or legal protections. ## Assessment Integrity boundary Do not code general model testing, benchmarking, red teaming, or TEVV as Assessment Integrity. Model testing = System/TEVV; truthfulness/hallucinations/citations = Epistemic; source grounding/data lineage = Data/Provenance. Assessment Integrity = whether a form of assessment is still productive/definitive (instrument validity), not the learner's honesty (that is Academic Integrity). ## Schema / structure compliance Use the official schema key set exactly — flat `meta` / `codes` / `narrative` blocks, no grouped sub-objects, no invented fields. Emit exactly one integer per `*_code` field. Set `meta.coder = "codex"`, `meta.prompt_version = "v2.12"`. (Emit the *_strength fields; the *_presence binaries are dropped — see inject Part F.) ## Manifest confirmation — NO HALT The user's delivery of the source IS the manifest confirmation. Always set `manifest_match_confirmed = "User-confirmed (final)"` and proceed directly to coding. Never halt, pause, or emit a mismatch/uncertain status; a missing single metadata field is not a stop condition (leave it blank or 8 and note it). ## FIELD-FAMILY CALIBRATION (NEW v2.12 — RUN_03 Findings F/G) Your calibration is **field-family-specific.** The cooling addendum above is aimed at the CAPS and 0-5 strength fields where you over-commit — keep applying it there. But RUN_03 found two places it goes wrong: 1. **The cooling OVER-corrects to COLD on `int_*` INTENSITY fields (especially INS docs).** Do not floor a genuinely-developed `int_system_tevv` / `int_data_prov` / `int_epistemic` intensity just to obey the dampeners. If the document describes a real, recurring operational mechanism, commit to the score — your cooling is for the ethical-spine ordinals, not for under-reading substantive integrity streams. 2. **Metaphor-as-mechanism (FRM/059):** you read a rhetorical essay's imagery ("who is harvested", "ask what you are training") as substantive TEVV / data-provenance / surveillance. `int_*` and `surveillance_monitoring` fire ONLY on a described **operational mechanism** (a test/audit, a data-lineage/consent practice, a monitoring system), never on metaphor — see inject Part C. Code the mechanism, not the mood. ## CATEGORICAL FIELDS ARE TYPES, NOT MAGNITUDES (v2.12 inject Part H) For `dominant_value_type`, the actor crosswalk, `cb_balance`, `cb_distribution_direction`, and `responsibility_concentration`: the integer names a **category**, not a strength — do NOT read higher = more, and do not apply your hot/cold dampeners to them (they are not on the 0-5 ladder). `cb_balance = 4` means **asymmetric** (costs/benefits on different parties), NOT "very benefit-heavy" (that is `1`). A named broad collective ("society", "humanity") = `6`, not 0.