LMeta/LCDL PhysicsGenesis Prompt Pack v1

What this pack supports

It deliberately separates two programs:

1. Reference physics library construction — ingest established quantities, units, constants, principles, laws, derivations, models, experiments, datasets, simulations, and historical lineages into typed, provenance-preserving objects.
2. Clean-room physical discovery — expose a learner only to minimal mathematical, measurement, world, and intervention primitives; stage its invented quantities, relations, models, experiments, and invariants; then falsify, validate, deduplicate, safety-check, measure, and govern promotion.

The pack is designed for Forge Intelligence problem-class policies, LMeta units and library flows, FI-local LCDL contracts, artifact envelopes, maturity gates, session dual-wikis, and the deterministic-to-bounded-model reasoning ladder.

Physics-specific additions beyond MathGenesis

PhysicsGenesis treats the following as first-class rather than optional metadata:

• physical quantity, quantity kind, unit, dimension, and conversion semantics;
• coordinate frame, geometry, metric, orientation, sign, gauge, basis, and normalization conventions;
• system boundary, environment, initial conditions, and boundary conditions;
• governing, constitutive, closure, constraint, conservation, and fitted equations;
• approximations, omitted effects, dimensionless groups, validity regimes, and failure regimes;
• measurand definition, apparatus limits, calibration, traceability, uncertainty budget, covariance, preprocessing, and detection limits;
• competing hypotheses, interventions, controls, confounders, stopping rules, null results, and replication;
• train, held-out, intervention, and replication evidence as separate roles;
• numerical residuals, convergence, invariants, code and environment hashes, and model-form error;
• laboratory hazard and dual-use detail controls;
• clean-room worlds with hidden governing truth and learner-visible observation/intervention interfaces.

Start here

1. Read SITE_ANALYSIS.md and integration/TARGET_LAYOUT.md.
2. Choose reference or clean-room mode and read integration/GOVERNANCE.md.
3. Use prompts/00_system_library_governor.md as the system message for every task.
4. Run 01_partition_and_ingestion_plan for each source collection or discovery generation.
5. Use prompts 02–24 for extraction, metrology, theories, experiments, domain intelligence, simulations, inverse problems, structural links, curriculum, and candidate mining.
6. Run 25–29 in independent contexts for falsification, validation, ontology, safety, and contamination.
7. Run 30_promotion_gate only after all required reports exist.
8. Run 33_repository_assembler to produce a write plan; ordinary version-controlled tooling applies approved changes.
9. Run 34_batch_release_auditor before merge.
10. Run 35_historian_equivalence_novelty_assessor only after a learner generation is frozen.

Most important rule

Do not expose the modern reference library to a learner expected to rediscover its concepts. Build the complete modern library under reference; expose only learner_seed, curriculum, approved prior discoveries, and declared world interfaces during a clean-room run.

Native mode — recommended

Use:

• schemas/physics-knowledge-object.schema.json;
• schemas/physics-library-batch.schema.json;
• proposed storage under knowledge/physics/;
• separate object roots for reference, learner_seed, curriculum, discovery_staging, promoted_discovery, and rejected;
• source, quantity, unit, model, experiment, dataset, snapshot, and dependency registries.

Native mode distinguishes mathematical derivation, computational verification, source grounding, dimensional validity, training fit, held-out validation, intervention support, regime bounding, and replication.

Compatibility mode

Use templates/compatibility_knowledge_module.yaml to wrap native object references in the current generic knowledge-module shape. Keep the native object canonical and validate it separately. Suggested jurisdictions:

• physical-model/universal-foundation for quantity and metrology foundations;
• physical-model/<regime> for theories and models;
• experiment/<apparatus-or-world> for protocols and observations;
• clean-room/physica0 for learner seed and curriculum objects.

Prompt groups

Recommended first reference release

The backlog proposes twelve initial modules:

• quantity kinds and dimensions;
• units and conversions;
• measurement models;
• uncertainty and covariance;
• kinematics;
• momentum and collisions;
• oscillation foundations;
• circuit and network foundations;
• thermodynamic states and processes;
• numerical verification and validation;
• experiment design;
• safety and dual-use governance.

Recommended first clean-room experiment

Physica-0 starts from minimal logic, finite mathematics, geometry, comparison, reference marks, unit labels, position/time markers, observation, repeat, control, and intervention operations.

It includes controlled worlds for tracks, slopes, collisions, pendulums, waves, rays, static sources, circuits, gases, heat networks, orbits, moving observers, and two-path count statistics. Hidden modern target mappings remain historian-only.

The learner is evaluated on whether its inventions:

• improve held-out prediction;
• compress recurring observations or procedures;
• transfer across worlds;
• improve experiment efficiency;
• expose regime boundaries and anomalies;
• survive independent falsification;
• remain uncontaminated and safe;
• can be operationally measured and reproduced.

Key files

• PROMPT_PACK.md — all 36 prompts in one document.
• prompts/ — individual system and task prompts.
• prompts.jsonl — batch-friendly records with full system and user templates.
• prompt_catalog.yaml — prompt registry.
• schemas/ — 21 physics, planning, assurance, promotion, repository, training, and historian schemas.
• curriculum/ — Physica-0 seed, 35 problem families, 13 worlds, taxonomy, experimental ladder, and reference backlog.
• examples/ — schema-valid objects, batches, cards, and assurance artifacts.
• templates/ — policy, LMeta, envelope, registry, wiki, session, and receipt templates.
• lcdl_contracts/ — eight proposed FI-local governed contracts.
• integration/ — architecture, governance, migration, runbook, safety, clean-room, and implementation guidance.
• tools/ — local prompt rendering and pack validation.

Verification posture

The generating model never self-certifies. Strong statuses require the corresponding independent receipt. A solved equation is not a validated model; a converged simulation is not an empirically confirmed theory; a training fit is not held-out prediction; an observational regularity is not necessarily causal; and a source report is not replication.

Implementation status

This is a complete design and prompt pack. Its proposed physics task IDs are not claimed to be registered or deployed. Promote capabilities through the existing Forge Intelligence contract, allowlist, maturity, fixture, semantic-validation, and review process.