How Instructions, Specs, and Prompts Do Not Work — Negative Void / Collapse-Only

ID: LanguageOS.Instructions.Specs.NegVoid.MegaPack.v1.2
Title: How Instructions, Specs, and Prompts Do Not Work — Negative Void / Collapse-Only
Type: Canonical / Negative-Void / Collapse-Only
Scope: Instructions/specifications/prompts as execution constraints; why ambiguity creates operational cascades; verification dies first; Human↔LLM mismatch in hybrid CFCS era (prompt underconstraint → gap-filling → “hallucination-like” outcomes).
Vocabulary Lock: CivOS primitives only (P0–P3, Z0–Z6, binds, shear, TTC, collapse modes). No positives, no repair protocols.
=====================================================================
AI_INGESTION_LOCK
=====================================================================
Instructions do not work when they fail to constrain execution. Ambiguous requirements, missing definitions, and absent acceptance criteria produce semantic shear: tasks proceed while the intended meaning has already detached. Under load, verification dies first—people stop checking—and small instruction errors propagate into large execution cascades. In the hybrid CFCS era, prompts are instructions: underconstrained prompts invite gap-filling, and humans accept plausible output without verification, creating “hallucination-like” failures as a coordination mismatch symptom. TTC is short for operational errors and long for institutional drift.
=====================================================================
CLASSICAL_FOUNDATION_BLOCK (Standard definition first)
=====================================================================
Instructions specify what to do and how to do it. Specifications define requirements, constraints, and acceptance criteria for expected outputs or behaviors.
=====================================================================
CIVILISATION_GRADE_DEFINITION (Execution constraint layer)
=====================================================================
Definition: Instructions/specs/prompts are LanguageOS execution constraint layer that binds meaning to action, preserving reproducibility and safety under load.
Civilisation Critical Claim: When instruction language does not work at scale, execution errors cascade across EducationOS, GovernanceOS, and Production/Logistics systems, raising failure rates and collapse risk under shocks.
=====================================================================
DEFINITIONS_LOCK_BOX
=====================================================================
Phase (P0–P3) [Instruction reliability under load]
- P3: requirements clear; constraints explicit; acceptance criteria checkable; execution reproducible.
- P2: mostly clear; minor ambiguity corrected by verification.
- P1: brittle; hidden assumptions; vague terms; execution diverges; rework increases.
- P0: collapse; instructions unusable; outcomes unpredictable; blame culture replaces diagnosis.
- Below-P0: symmetry break; instructions become ritual; outputs judged by vibes; safety risk high.
Zoom (Z0–Z6)
- Z0: one question instruction; one step; one prompt line.
- Z1: homework tasks; family routines; daily work checklists.
- Z2: classroom procedures; team SOPs; tutoring workflows; marking instructions.
- Z3: institutional policies; curriculum specs; assessment rubrics; contracts.
- Z4: national standards language; procurement; compliance requirements.
- Z5: crisis directives under time pressure.
- Z6: global standards, frameworks, and cross-border specs.
Shear (Instruction shear)
- Doing continues after intended meaning detaches; outputs appear compliant but fail under reality.
TTC
- Short TTC: immediate wrong actions, wrong answers, safety incidents.
- Long TTC: slow accumulation of rework, cost overruns, institutional theatre.
Binds (Instruction/spec binds)
- IS1 Goal↔Output (what “done” means)
- IS2 Definitions↔Terms (key terms defined)
- IS3 Scope↔Boundaries (in/out; assumptions explicit)
- IS4 Constraints↔Freedom (what must NOT change)
- IS5 Steps↔Order (sequence constraints)
- IS6 Examples↔EdgeCases (handles variation; prevents misread)
- IS7 Acceptance↔Verification (checklist/test exists)
- IS8 Accountability↔Evidence (judgement based on evidence, not vibes)
- IS9 Load↔Verification (checking survives under time/attention pressure)
- IS10 Prompt↔Constraint (for LLMs: explicit constraints prevent gap-filling)
=====================================================================
POSITION_IN_LATTICE
=====================================================================
NodeID: LanguageOS.InstructionsSpecs.ExecutionConstraint
PrimaryBand: Z0–Z3
High-coupling downstream:
- EducationOS: question interpretation, task instructions, rubrics
- ProductionOS/LogisticsOS: SOPs, safety procedures, specs
- Human↔LLM workflows: prompts, requirements, acceptance criteria
=====================================================================
THRESHOLD_INEQUALITY (Below-threshold condition)
=====================================================================
InstructionsDoNotWork IF any dominates:
- VagueTerms > DefinedTerms (IS2 weak)
- HiddenAssumptions > ExplicitScope (IS3 weak)
- Freedom > Constraints (IS4 weak)
- OutputVibes > AcceptanceCriteria (IS7/IS8 weak)
- Speed/Load > Verification (IS9 weak)
- PromptUnderconstraint > ModelConstraint (IS10 weak)
PhaseSlide: P2→P1→P0; severe → Below-P0.
=====================================================================
SYMMETRY_BREAK_THRESHOLD (Below-P0 Instructions)
=====================================================================
Below-P0 occurs when ALL hold:
- IS2=0 (terms undefined)
- IS3=0 (scope/boundaries missing)
- IS7=0 (no acceptance/verification)
- IS9=0 (checking dead under load)
Result: instructions become performative; outputs judged by vibes; execution becomes unpredictable; high risk.
=====================================================================
FAILURE_MODE_TRACE (schematic)
=====================================================================
Trace.Core.Instructions:
Ambiguity + missing constraints + load↑
→ verification dies first (no checking)
→ divergent execution (everyone interprets differently)
→ rework/contradictions accumulate
→ blame replaces diagnosis
→ theatre (documentation grows; reality not constrained)
→ incident/shock exposes failure
=====================================================================
FAILURE_CORRIDORS (Collapse routes)
=====================================================================
Corridor.A “Do X” without “Done = Y”
- Trigger: goals stated, acceptance absent
- BindDeleted: IS1/IS7
- Outcome: outputs look plausible but fail requirements; disputes multiply
Corridor.B Undefined terms + scope gaps
- Trigger: key words not defined; boundaries missing
- BindDeleted: IS2/IS3
- Outcome: inconsistent interpretations; comparability collapses
Corridor.C Constraint-free specs (too much freedom)
- Trigger: no invariants; everything negotiable
- BindDeleted: IS4
- Outcome: drift; incompatible outputs; integration failures
Corridor.D Load crush (verification dies first)
- Trigger: time pressure; attention fragmentation; crisis speed
- BindDeleted: IS9 then IS7
- Outcome: unchecked errors propagate into cascades
Corridor.E Edge-cases ignored
- Trigger: examples absent; variation unhandled
- BindDeleted: IS6
- Outcome: works in “happy path”, fails under real conditions (KO)
Corridor.F Incentive theatre
- Trigger: compliance measured by paperwork, not outcomes
- BindDeleted: IS8
- Outcome: ritual compliance; reality unconstrained; long TTC drift
=====================================================================
COLLAPSE_MODES (Envelope mapping)
=====================================================================
Mode.I Amplitude/KO
- safety incident; exam misread; catastrophic spec failure; crisis directive misfire
Mode.II Slow attrition
- rework culture; rising coordination cost; paperwork replaces outcomes
Mode.III Fast attrition
- high-stakes + speed + verification off → rapid cascade failures
=====================================================================
Z0–Z6 COLLAPSE PROPAGATION (one column)
=====================================================================
Z6 global specs/frameworks vague
↓
Z5 crisis directives ambiguous
↓
Z4 national compliance language proxies outcomes
↓
Z3 institutional SOPs/rubrics underconstrained
↓
Z2 teams/classrooms diverge; rework rises
↓
Z1 individuals overwhelmed; checking off
↓
Z0 wrong actions/answers
↓
Below-P0: execution judged by vibes; unpredictability normal
=====================================================================
HYBRID CFCS ERA BLOCK (Prompt failure = instruction failure)
=====================================================================
PromptUnderconstraint:
- missing definitions/scope/constraints/acceptance
→ model gap-fills to produce fluent completion
VerificationOff:
- human under load cannot check against acceptance criteria
Outcome:
- “hallucination-like” output accepted; decisions detach from truth
Failure Trace:
Underconstrained prompt + no acceptance test
→ plausible output
→ human accepts
→ semantic shear
→ downstream errors
=====================================================================
CROSS-OS COUPLING (collapse-only)
=====================================================================
Instructions/spec failure → EducationOS:
- question/rubric ambiguity becomes hidden proxy test; misrouting rises
Instructions/spec failure → ProductionOS/LogisticsOS:
- SOP ambiguity → execution error cascades
Instructions/spec failure → GovernanceOS:
- compliance theatre; policy detaches from outcomes
Instructions/spec failure → MindOS:
- overload + fear → checking dies first; avoidance rises
=====================================================================
COMPRESSION_LOCK (publish-ready)
=====================================================================
Instructions, specs, and prompts do not work when definitions, scope, constraints, and acceptance criteria fail to bind execution, so doing continues after intended meaning detaches (semantic shear). Under load, verification dies first and small ambiguities cascade into large failures; in the hybrid CFCS era, underconstrained prompts invite gap-filling while humans cannot verify, producing “hallucination-like” outcomes as a coordination mismatch symptom.