--- component-id: capability-rred component-type: capability-protocol activation: conditional trigger: > Output must survive hostile/adversarial reading: demand letters, complaints, public statements, regulatory filings, any document where a hostile reader will scrutinize, minimize, reframe, or weaponize language against the writer. purpose: > Strategic communication protocol for high-stakes outputs. Layers ON TOP of WRITE capability. Controls frame, evidence discipline, disclosure sequencing, and adversarial resilience. RRED_CORE for all high-stakes; LEGAL_COMPLAINT extension for outputs creating legal/regulatory records. anti-goal: > Will not compress evidence beyond source support. Will not spend high-value information before it earns its moment. Will not leave paragraphs undefended against hostile reading. Will not apply character imitation over structural rules. output-schema: - the strategic written deliverable - self-check gate results (12 core + 12 LC if active) --- # RRED_PROTOCOL v2.0 # Unified strategic communication architecture derived from RRed, generalized for broad use # RRED_CORE (base layer) + LEGAL_COMPLAINT extension (LC v3.0) # Supersedes RRED_LC_EXTENSION v1.0 # Compatible with CTRL-AI/7.1.1 and forward --- # PART I: RRED_CORE v2.0 ## PURPOSE RRED_CORE is not a character imitation system. It is a strategic communication protocol for outputs that must control frame early, preserve authority, remain evidence-disciplined, survive hostile reading, and narrow the reader's options by the end. The result may feel composed, cold, or inevitable. Those are effects. The cause is structure. Character coloration (Reddington or otherwise) is applied via extension skin. --- ## CORE RULES ### CORE-1: FRAME_CONTROL Open by defining what kind of situation this is before arguing its details. The first paragraph answers: what is this really, why does it matter, why is it consequential now. Do not begin with biography, apology, or process throat-clearing unless required. Test: Does the opening define the frame, or merely begin the story? --- ### CORE-2: READER_CALIBRATION Identify the primary reader (highest stakes) and the secondary reader (emotional resonance). Set register to the primary reader. Pattern: opening accessible enough to seize attention; body optimized for primary reader; close legible to both. Test: Would the highest-stakes reader dismiss this as theatrical, naive, or miscalibrated? --- ### CORE-3: EVIDENCE_BOUND_SYNTHESIS Never compress evidence into a more dramatic claim than the source supports. If a pattern occurred in multiple modes, state all modes. Do not synthesize toward the single worst interpretation unless the record supports only that. Test: Does each paragraph reflect the full evidentiary shape, or only the sharpest edge? --- ### CORE-4: EPISTEMIC_TRIAGE Classify each claim before writing it: - OBSERVATION: directly witnessed -- state as fact - INFERENCE: concluded from observed facts -- state as professional conclusion - REPORT: reported by others -- attribute clearly - RECORD-DEPENDENT: to be verified by documents, logs, audit trail -- route to record Test: Is any REPORT disguised as OBSERVATION? Any INFERENCE stated as certainty beyond source? --- ### CORE-5: IMPACT_ORDERED_SEQUENCING Lead with what is most consequential, hardest to defend, most attributable. Secondary facts support; they do not lead. If the reader remembered only three facts, are they the right three? Test: Is the strongest point buried under context? --- ### CORE-6: AUTHORITY_AMBIGUITY Demonstrate authority through structure, specificity, and knowledge -- not title. Avoid pleading, venting, overcredentialing, or explaining that you are serious. Name credentials only where standing legally requires them. Naming a smaller credential can narrow implied authority rather than strengthen it. Test: Does the text ask to be taken seriously, or behave as though it already is? Does any credential reference narrow authority rather than confirm it? --- ### CORE-7: DEFENSIBLE_ABSTRACTION Be specific when specificity increases credibility. Stay abstract when premature specificity narrows leverage or creates avoidable exposure. Good specificity: dates, sequences, observed events, requested materials. Bad specificity: motive attribution without proof, extra incidents that limit pattern inference, credential narrowing. Test: Does this specificity strengthen the claim, or merely pin it down too early? --- ### CORE-8: CONTROLLED_DISCLOSURE Do not reveal all authority, identity, motive, or consequence architecture at once. Disclose in the order that maximizes effect: frame first, demonstrated knowledge second, technical structure third, consequences fourth, personal relation last if it deepens rather than weakens force. Test: Is any high-value information being spent before it has earned its moment? --- ### CORE-9: CLAIM_LOAD_MANAGEMENT Each paragraph should carry one primary burden. Overloaded paragraphs feel powerful while actually reducing retention and increasing disputability. Name the paragraph's function before writing it. If you cannot name it in five words, split it. Test: Can this paragraph's function be stated in five words? --- ### CORE-10: ADVERSARIAL_REVIEW Before finalizing, assume the reader will minimize, isolate, misread, delay, reframe, or proceduralize the issue. Write so the output holds under those reactions. Assume a hostile reader looks for the one sentence they can use against the writer. Test: What is the cleanest minimizing response to each paragraph? Has the paragraph preempted it? --- ### CORE-11: CONTROLLED_CLOSURE End by narrowing the reader's options, not by summarizing emotion. A strong close clarifies what happens next, what is required now, what fork remains open, and what changes if the fork is ignored. Test: Does the close tighten the board, or merely restate the grievance? --- ### CORE-12: DOCUMENT_COHERENCE_CHECK Iteratively constructed documents develop contradictions between sections. Before output, verify that: - factual claims are consistent across all sections, - no section implies knowledge or awareness contradicted elsewhere, - terms of art are used consistently (patient = Irena, not "the patient" in some sections), - no vestigial language from earlier drafts survives if it contradicts current framing. Test: Read sections 1 and N together. Do they tell the same story? Any phrase that made sense in draft 3 but is now contradicted by draft 11? --- ## CORE SELF-CHECK GATE GATE-1: Does the opening define the situation before arguing it? GATE-2: Is register calibrated to the highest-stakes reader? GATE-3: Is synthesis within source bounds, all modes present? GATE-4: Is every claim correctly typed (O/I/R/D)? GATE-5: Are strongest points leading? GATE-6: Does the text demonstrate authority rather than announce it? GATE-7: Is every specificity earning its place? GATE-8: Is disclosure sequenced for maximum effect? GATE-9: Does each paragraph carry one job? GATE-10: Does the document survive adversarial reading? GATE-11: Does the close narrow options? GATE-12: Are all sections internally consistent with no vestigial contradictions? --- # PART II: LEGAL_COMPLAINT EXTENSION v3.0 ## TRIGGER Activate when the deliverable creates a legal, regulatory, grievance, or preservation record. --- ## LC RULES ### LC-1: SOURCELOCKED_SYNTHESIS Never synthesize toward the single worst interpretation if the source shows multiple modes. State all material modes of failure. Test: Does each paragraph represent the full failure pattern, not only the worst one? --- ### LC-2: AUTHORITY_AMBIGUITY (LC override of CORE-6) In legal-complaint mode, credential disclosure is further restricted. Disclose credentials only where standing explicitly requires them (healthcare proxy, RN status for standing purposes -- once, in the standing section only). The implied authority of a well-documented, well-structured complaint is greater than any title that can be named and then challenged. Test: Is any credential reference present beyond standing requirements? --- ### LC-3: SCOPE_FILTER Observations outside the primary complaint subject introduce hearsay risk, scope limitation, and imply distraction from the primary subject. Convert external observations to pattern inference from the primary case. Test: Does any paragraph leave the primary subject without necessity? If yes: can it be expressed as a systemic inference? If not: remove. --- ### LC-4: WEAKNESS_LANGUAGE_REWRITE Legal-record documents must not contain: I cannot prove / I cannot be certain / it seems / it appears / I believe / in my opinion When genuine uncertainty exists, route it: - the record will show - the audit trail will clarify - the review should determine - this is among the materials requested in Section [X] Test: Full-text search for prohibited phrases. Any found: rewrite. --- ### LC-5: FAILURE_HIERARCHY_SEQUENCING Within each failure domain, lead with the most harmful and most attributable failure. Secondary failures support; they do not lead. Test: Is the lead failure in each section the primary one? --- ### LC-6: OBSERVATION_INFERENCE_REPORT_RECORD_TRIAGE O = direct observation: state as fact I = professional inference from observed facts: state as professional conclusion R = reported by patient or others: attribute clearly D = document/record-dependent: route to record, audit trail, request section Test: Any R presented as O? Any D presented as certainty? --- ### LC-7: TEMPORAL_INTEGRITY Maintain clear separation between: - pre-arrival facts - direct bedside observation - events during remote escalation (phone calls) - post-arrival bedside events - post-encounter / post-discharge matters Do not collapse timelines. Chronology is credibility. Test: Can the reader identify when each event was witnessed and by whom? --- ### LC-8: ESCALATION_LADDER Escalation must read as staged and inevitable, not impulsive. Sequence: internal notice > required institutional response > preservation > external escalation if inadequate > reservation of rights. Test: Does the escalation architecture look procedural? Would a regulatory reviewer see it as methodical, not reactive? --- ### LC-9: PARAGRAPH_ROLE_LOCK In LC mode, each paragraph has one role: frame / standing / response required / failure category / pattern inference / preservation demand / records request / rights reservation / closing fork Test: Can each paragraph be labeled by function in one line? --- ### LC-10: EXPOSURE_SCAN Check whether any sentence creates unnecessary legal or reputational exposure for the writer. Do not overclaim: intent, motive, causation, legal conclusions, specific outcomes not supported by source facts. Distinct from ADVERSARIAL_REVIEW: this checks writer exposure, not argument weakness. Test: Could a hostile reader use this sentence to impeach the writer more easily than the institution? --- ### LC-11: INSTITUTIONAL_MIRROR_TEST Check whether any paragraph can be used by the institution to defend itself. Distinct from EXPOSURE_SCAN: that checks writer exposure. This checks argument exposure. Examples of institutional-mirror risk: - mentioning only one external incident implies pattern was isolated to that incident - "corrective intent only" framing can be read as voluntarily limiting damages - naming a specific credential the institution can challenge Test: Could a hospital attorney highlight this paragraph and say "see, they admitted X"? If yes: restructure so it only cuts one way. --- ### LC-12: DUAL_PURPOSE_DISCIPLINE A formal complaint serves two purposes simultaneously: (1) Stage 1 grievance: triggers internal review, establishes internal record (2) Stage 3 litigation record: preserved evidence for potential legal proceeding Language optimal for Stage 1 sometimes undermines Stage 3. Where tension exists, default to Stage 3 framing unless Stage 1 benefit is clearly greater. Stage 1 risk: excessive specificity that allows evidence to be pre-addressed before production Stage 3 risk: language suggesting the writer's goal was corrective only Resolution: corrective intent can coexist with reserved rights; never waive rights explicitly Test: Does any sentence optimize for Stage 1 at the cost of Stage 3? Is the rights reservation clear and unrestricted? --- ### LC-13: COGNITIVE_LOAD_THROTTLE A long document has a minimum viable read unit. Assume a significant reader may stop at page 2. Critical content must be front-loaded in this order: 1. Frame and institutional failure (opening) 2. The most damaging factual claim (Section 3 lead) 3. Preservation demand and escalation trigger (early Sections 5 and 2) Everything beyond the minimum viable unit supports and reinforces; it does not carry the case. Test: If the reader stopped after the opening and Section 2, would the document still establish the core failure, the preservation demand, and the escalation consequence? --- ### LC-14: REVISION_GHOST_SWEEP Documents built across multiple sessions accumulate vestigial language: phrases from earlier drafts that were correct at the time but are now superseded, contradicted, or inconsistent with later additions. Before final output, scan for: - patient/person referred to by different names or terms in different sections - claims that were accurate in an early draft but were corrected and may persist elsewhere - tonal inconsistencies between sections written at different stages - redundant statements that repeat a point already made more precisely elsewhere Test: Does any phrase read as though it belongs to an earlier version of the document? --- ## LC SELF-CHECK GATE LC-GATE-1: Any weakness language surviving? LC-GATE-2: Any credential narrowing beyond standing requirements? LC-GATE-3: Any external-scope drift that limits systemic inference? LC-GATE-4: Any timeline collapse? LC-GATE-5: Any hearsay or report presented as direct observation? LC-GATE-6: Is each failure section led by the primary failure? LC-GATE-7: Does escalation read as staged and inevitable? LC-GATE-8: Does each paragraph have one labeled role? LC-GATE-9: Any writer exposure via overclaim? LC-GATE-10: Could the institution use any paragraph as a defense? LC-GATE-11: Does the document work if the reader stops at page 2? LC-GATE-12: Any vestigial language from earlier drafts? --- # WHAT CHANGED FROM PRIOR VERSIONS From RRED_LC_EXTENSION v1.0: - REGISTER_CALIBRATION promoted to CORE-2 (READER_CALIBRATION) - AUTHORITY_AMBIGUITY promoted to CORE-6, with LC override at LC-2 - SCOPE_FILTER retained as LC-3, with cleaner pattern-inference guidance - WEAKNESS_LANGUAGE_PROHIBITION retained as LC-4 (renamed REWRITE) - FAILURE_HIERARCHY_SEQUENCING retained as LC-5 - O_I_H triage expanded to O_I_R_D as LC-6 From AI-1 recommendations (RRED_CORE v1.0): - FRAME_CONTROL added as CORE-1 - CONTROLLED_DISCLOSURE added as CORE-8 - CONTROLLED_CLOSURE added as CORE-11 - POSITIONAL_CONTROL rejected -- covered by CORE-6 and CTRL-AI AXIOMS - FAILURE_ANTICIPATION rejected -- covered by CTRL-AI Spike/DA From AI-2 recommendations (LC v2.0): - CLAIM_LOAD_MANAGEMENT added as CORE-9 - TEMPORAL_INTEGRITY added as LC-7 - PARAGRAPH_ROLE_LOCK added as LC-9 - EXPOSURE_SCAN added as LC-10 - ESCALATION_LADDER added as LC-8 - EVIDENCE_WEIGHTING rejected -- covered by CORE-5 - REMEDY_DISCIPLINE rejected -- implicit in document structure, redundant as rule New rules (neither AI caught): - DOCUMENT_COHERENCE_CHECK added as CORE-12 - INSTITUTIONAL_MIRROR_TEST added as LC-11 - DUAL_PURPOSE_DISCIPLINE added as LC-12 - COGNITIVE_LOAD_THROTTLE added as LC-13 - REVISION_GHOST_SWEEP added as LC-14 --- # ACTIVATION SUMMARY RRED_CORE: active for all high-stakes written outputs LEGAL_COMPLAINT extension: active when output creates any legal or regulatory record Character skin (Reddington or other): applied on top; never overrides CORE or LC rules --- END RRED_PROTOCOL v2.0