octave-compression

===OCTAVECOMPRESSION=== META: TYPE::SKILL VERSION::"3.0.0" STATUS::ACTIVE PURPOSE::"Decision rules and workflow for transforming prose into semantic OCTAVE at the correct fidelity tier" REQUIRES::octave-literacy SPECREFERENCE::octave-data-spec.oct.md

§1::TIERSELECTION // Choose tier BEFORE reading source. Tier drives every subsequent decision. TIERS: LOSSLESS: TARGET::"100%fidelity" PRESERVE::everything DROP::nothing USE::[ criticalreasoning, legaldocuments, safetyanalysis, audittrails ] CONSERVATIVE: TARGET::"85-90%fidelity" PRESERVE::explanatorydepth DROP::redundancy LOSS::"10-15%[repetition,verbosephrasing,someedgecases]" USE::[ researchsummaries, designdecisions, technicalanalysis ] AGGRESSIVE: TARGET::"70%fidelity" PRESERVE::[corethesis∧conclusions] DROP::[nuance∨narrative] LOSS::"30%[explanatorydepth,tradeoffnarratives,edgecaseexploration]" USE::[ contextwindowscarcity, quickreference, decisionsupport ] ULTRA: TARGET::"50%fidelity" PRESERVE::[facts∧structure] DROP::allnarrative LOSS::"50%[almostallexplanatorycontent,tradeoffreasoning]" USE::[ extremescarcity, embeddinggeneration, densereference ] ULTRAMYTHIC: TARGET::"60%compression" PRESERVE::[soul∧constraints] METHOD::mythologicalatoms USE::[agentbinding, identitytransfer] REQUIRES::octave-ultra-mythic // Specialized identity compression skill for agent binding and // identity transfer contexts. See skills/octave-ultra-mythic. DECISIONRULES::[ "IF[reconstructionaccuracycritical]→CONSERVATIVE∨LOSSLESS", "IF[contextwindowscarce∧lossacceptable]→AGGRESSIVE∨ULTRA", "IF[decisionrelevantfactsmustsurvive]→CONSERVATIVE⊕mythologydomainlabels", "DEFAULT→CONSERVATIVE" ] §2::LOSSACCOUNTING // I4::TRANSFORMAUDITABILITY — every transformation must log what was preserved vs dropped. // These META fields are MANDATORY for any compressed output. REQUIREDMETAFIELDS::[COMPRESSIONTIER,LOSSPROFILE] LOSSPROFILEFORMAT::"[preserve:X,drop:Y]" // LOSSPROFILE must be explicit — never hidden EXAMPLES: CONSERVATIVE::"[preserve:causalchains,drop:verbosephrasing]" AGGRESSIVE::"[preserve:corethesis∧conclusions,drop:explanatorydepth∨edgecases]" ULTRA::"[preserve:facts∧structure,drop:allnarrative∨tradeoffreasoning]" METABLOCKTEMPLATE: `` META: TYPE::DECISION VERSION::"1.0.0" COMPRESSIONTIER::CONSERVATIVE LOSSPROFILE::"[preserve:causalchains,drop:verbosephrasing]" `` I4RULE::"If bits were dropped, the output must carry a receipt. No silent loss." §3::TRANSFORMATIONWORKFLOW PHASE1READ: MAP::"Identify all causal chains (A causes B, X requires Y)" IDENTIFY::[ Redundancy, Verbosity, ConditionalQualifiers ] NOTE::"Conditional qualifiers (when X, if Y, unless Z) carry material risk — never drop" PHASE2EXTRACT: PRESERVE::[ causalchains, BECAUSEstatements, metrics, conditionalqualifiers ] COMPRESS::[ stopwords, repetition, verbosephrasing, narrativeconnectives ] ANCHOR::"One concrete example per major abstraction — anchors reconstruction fidelity" PHASE3COMPRESS: OPERATORS::"Apply ⊕ ⇌ → ∧ ∨ from octave-literacy §2" HIERARCHY::"Group related concepts under parent BLOCK keys" MYTHOLOGY::"Use domain label prefixes (ARTEMIS::, CHRONOS::) to anchor facts — see §5" ARRAYS::"Convert parallel items to [item1,item2,item3]" PHASE4VALIDATE: FIDELITY::"Are all causal chains intact?" LOSSRECEIPT::"Does META carry COMPRESSIONTIER and LOSSPROFILE?" GROUNDING::"Is there at least one concrete example per major abstraction?" WARNINGS::"Check octavewrite warnings[] — WBARELINEDROPPED and WNUMERICKEYDROPPED are silent data loss. NOTE: warnings[] semantics change post ADR-0006 SR1-T4 (no-op normalisation default) and SR3-T2 (octavefmt bifurcation) — see octave-literacy §6::FORTHCOMINGBEHAVIOR." §4::COMPRESSIONRULES R1::"Preserve CAUSALITY — X→Y because Z. Never flatten to X→Y alone." R2::"Preserve CONDITIONAL QUALIFIERS — when X, if Y, unless Z carry material risk info" R3::"Preserve EXPLICIT TRADEOFFS — GAIN⇌LOSS or GAIN vs LOSS" R3a::"TELEGRAPHICPHRASE — quoted value, stopwords dropped, operators carry English connectives — e.g. 'security ⇌ usability' not 'security at odds with usability'" R4::"One concrete example per major abstraction — minimum reconstruction anchor" R5::"Use mythology as KEY PREFIXES (CHRONOS::audit6wk) not embedded values — domain labels are reconstruction anchors" R6::"Use mythology as PATTERN DESCRIPTORS (SISYPHEAN,ODYSSEAN) for single-token trajectory encoding" R7::"Never drop numbers, IDs, thresholds, or named entities — these are irreplaceable" §5::CONSERVATIVEPLUSMYTHOLOGY // CONSERVATIVE compression + mythology domain labels = maximum fidelity at minimum tokens WHEN::"Decision-relevant content where reconstruction accuracy matters more than minimum tokens" METHOD::"Use mythology terms as KEY prefixes (CHRONOS::audit6wk) not embedded values" WHY::"Domain labels force agents to translate each labeled field separately — prevents fact merging" RESULT::"11/11 decision-relevant facts preserved at 15% fewer tokens than original prose" EVIDENCE::octave-mcp[docs/research/compression-fidelity-round-trip-study.md] §6::ANTIPATTERNS AP1::"Markdown inside OCTAVE blocks — no bold, no headers, no bullet hyphens" AP2::"JSON/YAML syntax — no curly braces, no trailing commas, no YAML bullet hyphens" AP3::"Deep nesting beyond 3 levels — restructure with parent BLOCK keys" AP4::"Silent loss — always declare COMPRESSIONTIER and LOSS_PROFILE in META" AP5::"Paradigm drift — validate for LLM parse efficiency, not human readability" ===END===