ACT: Product-grade data-efficiency operationalization protocol

[neuron7xLab]

ACT: Product-grade data-efficiency operationalization protocol

Date: 2026-06-17
Scope: GeoSync final product-operational readiness lane
Status: PROTOCOL_READY / EXECUTION_PENDING / EVIDENCE_REQUIRED

0. Purpose

Convert GeoSync from a verification-first research repository into a product-grade operational data-efficiency system without diluting scientific falsifiability, CI integrity, or provenance. This is not an alpha claim, not a trading-bot claim, and not a physics-law proof. It is an engineering protocol for turning validated market-structure hypotheses into reproducible, measurable, bounded operational data pipelines.

1. Current factual constraints

1. The fast-test oracle restoration PR (ci(pr-gate): fix vacuous fast-test shard — restore real node-id collection #1153) is still open. Product-readiness work must treat all prior vacuous-green critical lanes as weakened evidence until replayed under the restored oracle.
2. The revalidation-ledger gate PR (gov(revalidation): post-vacuous-oracle revalidation ledger gate (DEBT_5) #1167) is open and designed to prevent silent promotion from PENDING to REPLAYED without oracle_terminal_sha and in-repository replay evidence.
3. The X10R reconstruction capsule PR (docs: add X10R physics reconstruction execution capsule #1165) is a draft execution contract, not an implemented pipeline.
4. Productization must not promote documentation-only artifacts into runtime guarantees.
5. Market-output claims must remain falsifiable, bounded, and evidence-classed.

2. Operational model

ORACLE_RESTORATION
  -> REVALIDATION_LEDGER_REPLAY
  -> EXECUTABLE_X10R_PIPELINE
  -> DATA_EFFICIENCY_METRICS
  -> PRODUCT_API_CONTRACT
  -> OBSERVABILITY_AND_DRIFT_GATES
  -> RELEASE_CANDIDATE_EVIDENCE_BUNDLE

Any skipped state is a protocol violation.

3. Industrial engineering objective

Deliver a product-level GeoSync lane that can answer, reproducibly and auditably:

• What data was ingested?
• What transformations were applied?
• Which physical / geometric / synchronization operators ran?
• What evidence supports or falsifies the result?
• What is the computational cost per useful signal?
• What is the decision boundary for refusing output?
• What artifact proves the run happened under the correct oracle?

4. Data-efficiency target definition

Data efficiency is not model decoration. It is the ratio between consumed data, compute, and validated signal output.

Required metrics:

data_coverage_ratio
missingness_rate
feature_reuse_ratio
artifact_reproducibility_hash_match
compute_seconds_per_instrument_day
memory_mb_per_1k_nodes
signal_density_per_window
falsifier_pass_rate
oracle_replay_coverage
claim_promotion_error_rate

Minimum gate:

artifact_reproducibility_hash_match == 1.0
falsifier_pass_rate == 1.0
claim_promotion_error_rate == 0.0
oracle_replay_coverage >= declared_bound

5. Required implementation lanes

Lane A — Oracle finalization

• Merge or terminally resolve ci(pr-gate): fix vacuous fast-test shard — restore real node-id collection #1153.
• Preserve terminal CI evidence.
• Record the final oracle SHA.
• No data-product claims are allowed before this state.

Lane B — Revalidation ledger replay

• Merge gov(revalidation): post-vacuous-oracle revalidation ledger gate (DEBT_5) #1167 or equivalent gate.
• Flip seeded lanes from PENDING only with real replay evidence.
• Each replay entry must include oracle_terminal_sha, artifact path, status, residual, and bound.
• Any missing replay evidence keeps the lane non-promotable.

Lane C — X10R executable pipeline

Implement a single-command pipeline:

python -m geosync.reconstruction.x10r --config configs/x10r.yaml --out artifacts/x10r_reconstruction

Required outputs:

manifest.json
input_fingerprint.json
curvature_residuals.json
kuramoto_trajectory.parquet
energy_contracts.json
falsifier_results.json
metrics/data_efficiency.json
reports/product_readiness.md
RO-Crate metadata

Lane D — Data contract hardening

• Define schema for market input windows.
• Add synthetic + fixture-backed real-data adapters.
• Fail closed on missing timestamps, non-monotonic series, invalid adjacency, NaN/Inf, unit ambiguity, or empty graph.
• Add golden fixtures and hash-pinned expected outputs.

Lane E — Product API boundary

Expose read-only deterministic endpoints or CLI commands:

geosync ingest
geosync reconstruct
geosync verify
geosync report
geosync export-capsule

No endpoint may imply trade execution or profitability.

Lane F — Observability / drift / refusal gates

Add operational monitors:

input_distribution_drift
graph_topology_drift
curvature_residual_drift
sync_regime_drift
energy_contract_violation
falsifier_regression
oracle_mismatch

Every monitor must have a refusal state.

Lane G — Release candidate evidence bundle

Create:

artifacts/release_candidate/VERDICT.md
artifacts/release_candidate/MANIFEST.json
artifacts/release_candidate/SHA256SUMS
artifacts/release_candidate/REPRODUCE.md
artifacts/release_candidate/METRICS.json

Verdict values:

PRODUCT_READY
RESEARCH_ONLY
BLOCKED_BY_EVIDENCE
BLOCKED_BY_ORACLE
BLOCKED_BY_DATA_CONTRACT

6. Acceptance criteria

The task is complete only when:

1. ci(pr-gate): fix vacuous fast-test shard — restore real node-id collection #1153 or equivalent oracle fix is terminal and evidence-preserved.
2. Revalidation ledger has no unjustified promotion.
3. X10R pipeline is executable by one command.
4. Artifacts are hash-stable across two independent runs.
5. Falsifiers fire on tampered curvature, wrong Kuramoto scaling, invalid energy sign, and missing oracle evidence.
6. Data-efficiency metrics are emitted and bounded.
7. Product API refuses invalid or under-evidenced output.
8. CI runs unit, integration, physics, governance, and artifact-reproducibility gates.
9. VERDICT.md states a bounded status without marketing inflation.

7. Stop rules

Stop immediately if:

• tests run under a vacuous oracle;
• documentation claims exceed executable evidence;
• market-profitability language appears without registered statistical proof;
• replay evidence is missing;
• artifact hashes are unstable;
• invalid data silently produces output;
• any gate downgrades fail-closed behavior to warning-only.

8. Final operational verdict target

TARGET_STATE=PRODUCT_GRADE_RESEARCH_OPERATIONS
TRADING_CLAIM=FORBIDDEN
PHYSICS_LAW_CLAIM=FORBIDDEN
DATA_EFFICIENCY=MEASURED_AND_BOUNDED
FALSIFIABILITY=EXECUTABLE
PROVENANCE=HASHED_AND_REPLAYABLE
RELEASE=BLOCKED_UNTIL_EVIDENCE_BUNDLE_GREEN

This issue is the execution protocol. Implementation must proceed as PRs bounded by the lanes above, not as one unreviewable mega-commit.

---

9. X11R post-critique reconstruction expansion

Status: POST_CRITIQUE_PROTOCOL_ADDED / EXECUTION_PENDING / EPISTEMIC_EVIDENCE_REQUIRED

This section extends X10R into X11R: rigorous epistemic reconstruction and falsification stress-test of the curvature-synchronization free-energy manifold under non-stationary market embedding with topological, causal, and uncertainty integrity enforcement.

The objective is not to intensify terminology. The objective is to convert the criticism of the prior ontology into executable research-engineering work: convergence analysis, embedding validation, causal robustness, uncertainty quantification, adversarial perturbation, and peer-review-grade artifact discipline.

9.1 Critical ontology constraints

X11R must explicitly preserve these boundaries:

• Kuramoto synchronization is a valid nonlinear-dynamics transfer, but Ricci-coupled Kuramoto requires stability, convergence, and perturbation analysis before any strong dynamical claim.
• Forman/Ollivier-Ricci curvature are valid network-geometric stress operators, but market graphs are not automatically Riemannian manifolds.
• Forman curvature can be unbounded below; any onset threshold such as -2.0 is policy unless derived, stress-tested, and evidence-classed.
• Gauss-Bonnet is a topological combinatorial integrity gate, not proof of market law.
• Fractal and neuro-symbolic layers are admissible only when falsifiers, uncertainty, and negative evidence are preserved.
• Market systems are adaptive, non-stationary, heteroskedastic, and agent-mediated; therefore all physics transfers must remain heuristic until falsified or bounded by evidence.

10. X11R required implementation lanes

Lane H — Graph construction and embedding validation

• Add order-book/trades-ready graph construction contracts for LOB, trades, returns, correlation, MST, PMFG, and weighted graph families.
• Add embedding distortion metrics: stress, distance preservation, curvature distortion, topology preservation, and sampling-regime metadata.
• Add multifractal frequency assignment path for natural frequencies omega_i, including MF-DFA or equivalent bounded estimator.
• Emit embedding_validation.json with declared graph family, distortion metrics, admissible interpretation class, and refusal reason if the graph is not geometrically interpretable.
• Fail closed on invalid embedding, undeclared graph family, non-stationary window without regime metadata, or distance metric ambiguity.

Lane I — Curvature layer with Ricci-flow analysis

• Preserve exact Knill-style Gauss-Bonnet rational residual checks for valid graph families.
• Compute dual Forman/Ollivier curvature where dependencies and graph scale allow.
• Add discrete Ricci-flow simulation lane with explicit convergence criteria, stopping rules, and topological-change logging.
• Track Euler-characteristic preservation or explicitly record topological transitions.
• Add witness families: tree, cycle, complete graph, random graph, scale-free, small-world, empirical graph, and tampered graph.
• Emit ricci_flow_convergence.json, topology_transition_log.json, and curvature_policy_bounds.json.
• Fail closed if curvature residual is non-zero, flow convergence is claimed without criterion, or threshold policy is promoted to theorem.

Lane J — Coupled Kuramoto on dynamic curvature manifold

• Implement curvature-modulated coupling options: K_local(kappa), omega_i(kappa), and bounded hybrid mode.
• Add Lyapunov/stability proxy tests: linearization spectrum, perturbation decay, phase-slip count, order-parameter convergence, and stochastic-noise sensitivity.
• Add stochastic Kuramoto extension for controlled noise injection, with deterministic seed and reproducible path.
• Track R(t), coherence spectra, phase slips, synchronization transitions, and regime-shift alignment.
• Emit kuramoto_curvature_coupling.json, stability_proxy.json, phase_slip_report.json, and sync_regime_transitions.parquet.
• Fail closed on negative coupling, unstable integration, unbounded local K, NaN phase, unseeded stochasticity, or undocumented scaling ownership.

Lane K — Free-energy and thermodynamic closure

• Preserve canonical thermo_free_energy = U - T*S separately from operational cost/entropy penalties.
• Add boundedness checks via epsilon-cap, delta-growth, Monte-Carlo tail stress, and finite-domain declarations.
• Add transfer-entropy or lagged cross-correlation falsifier for neuro-modulation claims.
• Add explicit units/dimensionless declaration for every energy-like quantity.
• Emit free_energy_closure.json, tail_stress_report.json, neuromodulation_falsifier.json, and unit_contracts.json.
• Fail closed on sign ambiguity, unit ambiguity, entropy penalty reversal, unbounded energy trajectory, or positive/flat correlation where inverse invariant is required.

Lane L — Causal robustness and counterfactual falsification

• Add causal robustness tests: Granger-style predictability tests, synthetic controls, placebo interventions, and counterfactual graph perturbations.
• Add Pearl-style do-intervention terminology only where the implemented test genuinely supports intervention semantics; otherwise mark as observational only.
• Add curvature intervention tests: modify curvature while preserving topology and verify whether synchronization/energy outputs respond as claimed.
• Add regime split validation: train/calibrate on one regime, falsify or bound on another.
• Emit causal_robustness.json, placebo_interventions.json, counterfactual_graph_tests.json, and regime_split_validation.json.
• Fail closed on causal language without intervention evidence, correlation-only claims promoted to causal claims, or placebo passing as signal.

Lane M — Bayesian uncertainty and epistemic ledger

• Add quantitative uncertainty layer for invariants: bootstrap, Bayesian posterior, MCMC, variational approximation, or explicitly bounded deterministic uncertainty where appropriate.
• Add evidence class per claim: THEOREM, EXACT_INVARIANT, EMPIRICAL_OBSERVATION, HEURISTIC_POLICY, FAILED_FALSIFIER, NOT_EVIDENCED.
• Extend contradiction ledger with prior claim, new evidence, falsifier result, posterior/evidence score, and demotion/promotion state.
• Emit epistemic_ledger.json, uncertainty_quantification.json, claim_state_transitions.json, and negative_evidence.json.
• Fail closed on claim promotion without evidence, missing uncertainty denominator, deleted negative evidence, or posterior reported without sampling/config provenance.

Lane N — Adversarial and non-stationary stress tests

• Add adversarial graph perturbations: edge rewiring, weight shock, node dropout, timestamp shuffle, market-regime injection, heavy-tail noise, and liquidity-gap simulation.
• Add non-stationarity injection fixtures with declared shock windows and expected refusal or bounded degradation behavior.
• Add agent-based synthetic falsifier lane for intentional-agent market dynamics where graph-physics assumptions should weaken.
• Emit adversarial_graph_stress.json, nonstationary_injection_report.json, and agent_based_falsifier.json.
• Fail closed if adversarial perturbations preserve unsupported claims, if timestamp shuffle still produces strong causal claims, or if regime injection does not change uncertainty.

Lane O — X11R capsule generator

• Extend the single-command interface:

python -m geosync.reconstruction.x11r \
  --config configs/x11r.yaml \
  --out artifacts/x11r_epistemic_reconstruction

• Required X11R outputs:

manifest.json
input_fingerprint.json
embedding_validation.json
curvature_residuals.json
ricci_flow_convergence.json
kuramoto_curvature_coupling.json
stability_proxy.json
free_energy_closure.json
causal_robustness.json
uncertainty_quantification.json
epistemic_ledger.json
adversarial_graph_stress.json
falsifier_results.json
metrics/data_efficiency.json
reports/x11r_epistemic_verdict.md
ro-crate-metadata.json
SHA256SUMS

• Fail closed if any required artifact is missing, unhashable, non-reproducible, or not referenced in the manifest.

11. X11R data-efficiency additions

Add the following metrics to metrics/data_efficiency.json or metrics/x11r_data_efficiency.json:

embedding_distortion_score
curvature_flow_convergence_rate
stability_proxy_pass_rate
causal_placebo_rejection_rate
counterfactual_sensitivity_ratio
uncertainty_coverage_ratio
negative_evidence_preservation_rate
adversarial_degradation_bound
nonstationary_refusal_accuracy
claim_evidence_alignment_score

Minimum gates:

negative_evidence_preservation_rate == 1.0
claim_evidence_alignment_score == 1.0
nonstationary_refusal_accuracy >= declared_bound
causal_placebo_rejection_rate >= declared_bound
uncertainty_coverage_ratio >= declared_bound

No X11R verdict may use PRODUCT_READY unless all required X10R gates and X11R epistemic gates are green.

12. X11R acceptance criteria

The X11R expansion is complete only when:

1. X10R oracle, ledger, replay, reproducibility, and data-efficiency gates are already satisfied or explicitly bounded.
2. Graph embedding validity is quantified and interpretation class is declared.
3. Ricci-flow convergence is tested or explicitly bounded as non-convergent / unsupported.
4. Curvature-modulated Kuramoto coupling has stability proxy evidence.
5. Free-energy contracts preserve unit/sign separation and boundedness.
6. Causal claims are separated from observational claims and protected by placebo/counterfactual tests.
7. Uncertainty is quantified for empirical invariants or explicitly declared as unsupported.
8. Negative evidence is preserved in the ledger.
9. Adversarial and non-stationary stress tests execute and alter verdicts when assumptions break.
10. The X11R capsule is reproducible, hashed, and linked through RO-Crate metadata.

13. X11R stop rules

Stop immediately if:

• a market graph is called a Riemannian manifold without embedding evidence;
• a policy threshold is presented as a theorem;
• Kuramoto-Ricci coupling is claimed stable without stability proxy;
• causal language appears without causal test evidence;
• uncertainty is reported without method/config/sample provenance;
• negative evidence is removed or hidden;
• adversarial perturbation does not affect confidence yet claims remain strong;
• non-stationary injection produces unchanged product verdict;
• X11R attempts to bypass X10R oracle and replay gates.

14. X11R final verdict target

TARGET_STATE=PRODUCT_GRADE_EPISTEMIC_RESEARCH_OPERATIONS
X10R_DEPENDENCY=REQUIRED
GRAPH_EMBEDDING=QUANTIFIED
CURVATURE_FLOW=TESTED_OR_BOUNDED
KURAMOTO_STABILITY=PROXIED_AND_HASHED
FREE_ENERGY=UNIT_SAFE_AND_BOUNDED
CAUSALITY=SEPARATED_FROM_CORRELATION
UNCERTAINTY=QUANTIFIED_OR_REFUSED
NEGATIVE_EVIDENCE=PRESERVED
ADVERSARIAL_ROBUSTNESS=MEASURED
RELEASE=BLOCKED_UNTIL_X10R_AND_X11R_EVIDENCE_GREEN

X11R is not a license to make larger claims. It is a stricter machine for demoting unsupported claims, preserving uncertainty, and forcing every high-status physical metaphor to either become executable evidence or shut up politely.

[neuron7xLab]

X11R Final Verification Checklist

Status: FINAL_VERIFICATION_CHECKLIST_ADDED / EXECUTION_PENDING / EVIDENCE_REQUIRED

This block extends Issue #1168 with the final X11R verification checklist. A checkbox is valid only when supported by executable code, CI output, reproducible artifact, hash, replay evidence, or explicit bounded refusal. Documentation-only completion is not accepted.

1. Curvature and topology integrity

• Enforce exact ricci.gauss_bonnet: rational residual exactly zero for valid graph families.
• Cover tree, cycle, K5, small-world, scale-free, empirical, and tampered graph witnesses.
• Compute dual Ollivier/Forman curvature where supported.
• Harden DualTrackRicciMonitor; policy thresholds must stay evidence-classed as policy.
• Quantify market-graph embedding distortion before geometric interpretation.
• Simulate discrete Ricci flow with convergence criteria and topology-transition log.

2. Kuramoto on curved dynamic manifold

• Preserve canonical FP-1 Kuramoto equation and single-owner scaling.
• Reject negative, unbounded, or undocumented coupling.
• Emit deterministic RK4/JAX trajectory artifacts.
• Add controlled stochastic extension only with deterministic seed.
• Track R(t), coherence spectra, phase slips, sync transitions, and early-stop reason.
• Add curvature feedback modes: K_local(kappa), omega_i(kappa), bounded hybrid.
• Add stability proxies: spectrum, perturbation decay, boundedness, phase-slip count.

3. Free-energy and thermodynamic closure

• Keep thermo_free_energy = U - T*S separate from operational cost objectives.
• Verify sign-correct entropy penalty and unit or dimensionless contracts.
• Add epsilon-cap, delta-growth, finite-domain, and tail-stress checks.
• Emit free_energy_closure.json, tail_stress_report.json, and unit_contracts.json.

4. Neuromodulation and control

• Validate dopamine/serotonin trajectory contracts when neuro-control is enabled.
• Enforce C-NEURO-003: inverse correlation must fail on positive or flat relation.
• Add transfer-entropy or lagged-correlation falsifier for neuro-control claims.
• Clamp controller verdict by weakest supporting evidence.

5. Governance and epistemic layer

• Pass physics_contracts/catalog.yaml and new contracts for convergence, causality, uncertainty, embedding validity, and perturbation robustness.
• Extend VerificationProtocol with evidence weights without compensating over critical failures.
• Extend contradiction ledger with claim, evidence, warrant, rebuttal, falsifier result, evidence class, and state transition.
• Preserve negative evidence in repository artifacts.
• Emit RO-Crate metadata with workflow-run provenance.
• Quantify uncertainty for empirical invariants or declare unsupported status.

6. Reproducibility and engineering standards

• Produce deterministic artifacts under declared seeds.
• Generate SHA256SUMS for all X11R artifacts.
• Compare deterministic fixtures against golden references.
• Reach ECC >= 0.95 or explicitly bound lower coverage.
• Add CI gates for physics, governance, reproducibility, data contracts, and CLI smoke tests.

7. Falsifiability and stress tests

• Tampered curvature must fire a gate.
• Wrong Kuramoto scaling must fire a gate.
• Invalid energy sign must fire a gate.
• Missing oracle evidence must fire a gate.
• Non-stationary input must weaken, refuse, or bound the verdict.
• Perturbed graph input must alter confidence or trigger refusal.
• Separate causal claims from observational claims.
• Maintain limitations ledger: markets are adaptive systems, not closed physical systems.

8. Required capsule artifacts

manifest.json
input_fingerprint.json
embedding_validation.json
curvature_residuals.json
ricci_flow_convergence.json
kuramoto_curvature_coupling.json
stability_proxy.json
free_energy_closure.json
causal_robustness.json
uncertainty_quantification.json
epistemic_ledger.json
perturbation_stress.json
falsifier_results.json
metrics/data_efficiency.json
reports/x11r_epistemic_verdict.md
ro-crate-metadata.json
SHA256SUMS

9. Final acceptance gates

• X10R oracle, ledger, replay, reproducibility, and data-efficiency gates are green or explicitly bounded.
• Graph embedding validity is quantified.
• Ricci-flow convergence is tested or explicitly declared unsupported.
• Curvature-modulated Kuramoto coupling has stability proxy evidence.
• Free-energy contracts preserve unit/sign separation and boundedness.
• Causal language is evidence-classed.
• Uncertainty is quantified or declared unsupported.
• Negative evidence is preserved.
• Stress tests execute and influence verdicts.
• X11R capsule is reproducible, hashed, manifest-linked, and RO-Crate referenced.

10. Target verdict

TARGET_STATE=PRODUCT_GRADE_EPISTEMIC_RESEARCH_OPERATIONS
TRADING_CLAIM=FORBIDDEN
PHYSICS_LAW_CLAIM=FORBIDDEN
MARKET_MANIFOLD_CLAIM=CONDITIONAL_ON_EMBEDDING_EVIDENCE
CAUSAL_CLAIM=CONDITIONAL_ON_EVIDENCE
UNCERTAINTY=QUANTIFIED_OR_DECLARED_UNSUPPORTED
FALSIFIABILITY=EXECUTABLE
PROVENANCE=HASHED_AND_REPLAYABLE
RELEASE=BLOCKED_UNTIL_X10R_AND_X11R_EVIDENCE_GREEN

This checklist is an execution gate for future PRs, not a completion claim.

[neuron7xLab]

Cross-Margin Validation Specification

Status: SPEC_ADDED / EXECUTION_PENDING / PRODUCT_CLAIM_FORBIDDEN

Purpose: add a cross-margin validation layer as a risk-integrity and data-efficiency gate, not as a trading or profitability module.

Scope

Validate portfolio-level margin consistency across instruments, accounts, collateral pools, and stress regimes.

Allowed claims:

• margin accounting integrity
• collateral sufficiency verification
• exposure aggregation correctness
• liquidation-boundary reproducibility
• stress-test falsifiability

Forbidden claims:

• trade execution
• alpha generation
• profit prediction
• exchange-specific liquidation guarantees without adapter evidence

Required inputs

1. positions: instrument, side, size, entry_price, mark_price, contract_multiplier, quote_currency
2. collateral: asset, quantity, haircut, valuation_source, timestamp
3. margin model: initial_margin, maintenance_margin, leverage cap, cross/isolated mode
4. risk parameters: volatility shock, correlation shock, liquidity haircut, funding cost
5. provenance: source, timestamp, hash, adapter version

Core invariants

1. total_equity = collateral_value + unrealized_pnl - fees - funding
2. used_margin >= sum(position_margin_after_haircuts)
3. free_margin = total_equity - used_margin
4. margin_ratio = total_equity / maintenance_margin_required
5. liquidation_boundary must be deterministic for identical inputs
6. cross-margin exposure must not silently net incompatible instruments
7. collateral haircut must be explicit and non-negative
8. missing mark price, stale timestamp, NaN, Inf, or negative collateral must fail closed

Validation gates

• single-position parity: isolated result must match cross-margin result when portfolio size = 1
• netting guard: long/short offsets allowed only inside declared compatible risk buckets
• collateral stress: haircut increase must not improve margin ratio
• price shock monotonicity: adverse shock must not improve equity for exposed side
• timestamp gate: stale collateral or mark data blocks PRODUCT_READY verdict
• currency conversion gate: FX conversion must be source-pinned and hash-recorded
• liquidation reproducibility: same input hash => same liquidation boundary hash

Required artifacts

• artifacts/margin/input_fingerprint.json
• artifacts/margin/collateral_valuation.json
• artifacts/margin/exposure_matrix.parquet
• artifacts/margin/margin_report.json
• artifacts/margin/stress_results.json
• artifacts/margin/falsifier_results.json
• artifacts/margin/SHA256SUMS

Tests and falsifiers

1. negative collateral -> fail closed
2. stale price -> fail closed
3. missing haircut -> fail closed
4. incompatible instrument netting -> fail closed
5. adverse price shock improves margin ratio -> fail closed
6. hash mismatch on repeated run -> fail closed
7. isolated/cross single-position mismatch -> fail closed

Data-efficiency metrics

• margin_input_coverage_ratio
• collateral_freshness_seconds
• exposure_matrix_density
• stress_scenarios_per_second
• reproducible_margin_hash_match
• incompatible_netting_rejection_rate
• stale_input_rejection_rate

Integration order

1. Add pure calculation module with no exchange execution.
2. Add fixtures for single-position, multi-asset, multi-currency, stale data, and incompatible netting.
3. Add falsifiers.
4. Add artifact writer.
5. Add CI gate.
6. Wire into X10R/X11R verdict only as risk-integrity evidence.

Final target: CROSS_MARGIN_VALIDATION_EXECUTABLE / FAIL_CLOSED / HASH_REPRODUCIBLE / NON_TRADING_RISK_GATE.

[neuron7xLab]

Cross-Margin Validation Operationalization #7

Status: OPERATIONAL_LAYER_ADDED / EXECUTION_PENDING / EVIDENCE_REQUIRED

Purpose: convert cross-margin validation from abstract risk need into an executable, role-bound, metric-driven implementation lane. This remains a non-trading risk-integrity gate: no buy/sell signals, no profitability claims, no silent margin assumptions.

1. Action sequence

1. Define input schemas for positions, collateral, balances, prices, haircuts, liabilities, account tiers, liquidation rules, and timestamp provenance.
2. Build deterministic exposure normalization: instrument -> currency -> account -> portfolio.
3. Implement margin invariant engine: equity >= maintenance_margin, initial_margin >= required_initial_margin, exposure <= configured caps, stale marks rejected.
4. Implement cross-margin netting validator: allow only declared compatible instruments; reject incompatible netting, circular collateral, negative collateral, missing haircuts, stale FX, and NaN/Inf.
5. Add liquidation-boundary calculator: distance_to_liquidation, margin_ratio, stress_margin_ratio, worst_case_collateral_shortfall.
6. Add scenario stress tests: price shock, volatility shock, FX shock, collateral haircut shock, correlation break, liquidity gap.
7. Emit artifacts: cross_margin_report.json, cross_margin_metrics.json, cross_margin_falsifiers.json, cross_margin_manifest.json, SHA256SUMS.
8. Add CI gate: invalid portfolio must fail closed; valid fixture must reproduce stable hashes.
9. Add CLI path: geosync risk validate-cross-margin --input fixtures/risk/cross_margin_valid.json --out artifacts/cross_margin/.
10. Integrate into product-readiness verdict as BLOCKED_BY_MARGIN_VALIDATION when any critical invariant fails.

2. Responsible roles

• Risk systems architect: owns margin model boundaries, invariant definitions, and forbidden claims.
• Quant/market-structure engineer: owns stress scenarios, exposure normalization, and numerical tolerances.
• Verification engineer: owns fail-closed tests, falsifiers, artifact hashes, and CI integration.
• Data engineer: owns schemas, provenance, timestamps, FX sources, and missingness gates.
• Release owner: owns final verdict, acceptance checklist, and non-trading disclaimers.

3. Required resources

• Schema files: schemas/risk/cross_margin.schema.json, schemas/risk/position.schema.json, schemas/risk/collateral.schema.json.
• Fixtures: valid portfolio, under-margined portfolio, stale marks, invalid netting, missing haircut, FX shock, liquidity gap.
• Config: configs/risk/cross_margin.yaml with caps, haircut policy, stale-price threshold, stress shocks, precision rules.
• Modules: geosync/risk/cross_margin.py, geosync/risk/margin_models.py, geosync/risk/stress.py.
• Tests: tests/risk/test_cross_margin_validation.py, tests/risk/test_cross_margin_falsifiers.py, tests/ci/test_cross_margin_artifact_reproducibility.py.

4. Metrics

• margin_ratio
• maintenance_margin_coverage
• initial_margin_coverage
• distance_to_liquidation
• portfolio_leverage
• netting_compatibility_ratio
• stale_mark_rate
• missing_haircut_rate
• stress_margin_ratio_min
• collateral_shortfall_max
• artifact_hash_reproducibility
• falsifier_pass_rate

Minimum gates:

• artifact_hash_reproducibility == 1.0
• falsifier_pass_rate == 1.0
• missing_haircut_rate == 0.0
• stale_mark_rate == 0.0
• stress_margin_ratio_min >= configured_minimum
• invalid fixture must fail closed.

5. Checkpoints

• C1: schemas accepted and invalid schemas rejected.
• C2: deterministic exposure normalization passes fixture parity.
• C3: margin invariants pass valid fixture and fail invalid fixture.
• C4: cross-margin netting rejects incompatible/circular collateral cases.
• C5: stress tests emit bounded metrics.
• C6: artifacts are SHA256-stable across two runs.
• C7: CLI emits VALID, INVALID, or BLOCKED_BY_DATA_CONTRACT; no ambiguous state.
• C8: CI gate integrated into product-readiness workflow.

6. Stop rules

Stop and block merge if:

• any invalid portfolio produces a valid report;
• stale prices are accepted silently;
• missing haircut defaults to zero risk;
• incompatible instruments are netted;
• liquidation boundary is undefined or non-finite;
• artifacts are not reproducible;
• report contains trading/profitability claims;
• any critical risk failure is downgraded to warning-only.

Final target: CROSS_MARGIN_VALIDATION_OPERATIONAL / FAIL_CLOSED / HASH_REPRODUCIBLE / NON_TRADING_RISK_GATE.

[neuron7xLab]

Scale Normalization Layer

Status: NORMALIZATION_SPEC_ADDED / EXECUTION_PENDING / EVIDENCE_REQUIRED

Goal: bring heterogeneous GeoSync data, risk signals, physics metrics, evidence states, and PR-control states onto a single comparable operational scale without converting verification infrastructure into a trading engine.

1. Normalization targets

• Market data: timestamps, sampling intervals, missingness, instrument units, return windows, volatility windows.
• Graph data: node count, edge weights, adjacency normalization, curvature ranges, embedding distortion.
• Dynamics: Kuramoto order parameter R, phase slips, convergence residuals, Lyapunov/stability bands.
• Energy/risk: thermo_free_energy, operational_cost_energy, cross-margin exposure, collateral, haircut, liquidation distance.
• Governance: claim state, evidence class, quarantine count, replay coverage, falsifier pass rate.
• PR operations: draft/non-draft, mergeability, CI state, required rebase, blocked-by-evidence.

2. Canonical scale contract

Every emitted metric must declare:

• raw_value
• normalized_value
• unit
• denominator
• window
• source_artifact
• transformation
• acceptable_band
• fail_closed_condition

Allowed normalized domains:

• ratio: [0, 1]
• signed score: [-1, 1]
• z-score: finite real with declared reference population
• ordinal state: deterministic enum only
• hash equality: 0 or 1

3. Required gates

• No metric without denominator.
• No score without declared transformation.
• No cross-asset comparison without unit/window normalization.
• No margin/risk comparison without collateral currency and haircut basis.
• No physics claim from normalized convenience score alone.
• No PRODUCT_READY verdict if normalized evidence coverage is below declared threshold.

4. Artifacts

Generate:

• artifacts/normalization/scale_manifest.json
• artifacts/normalization/normalized_metrics.json
• artifacts/normalization/unit_contracts.json
• artifacts/normalization/transformation_audit.json
• artifacts/normalization/refusal_report.md

5. Tests and falsifiers

• Missing denominator -> fail.
• Mixed units without conversion -> fail.
• NaN/Inf normalized values -> fail.
• Cross-asset aggregation without sampling-window alignment -> fail.
• Margin metric without haircut/collateral basis -> fail.
• Evidence score derived from docs-only claim -> fail.

6. Target verdict

NORMALIZED_OPERATIONAL_SCALE_READY only when all major data, physics, risk, governance, and PR-control metrics are bounded, unit-declared, denominator-declared, hash-reproducible, and fail-closed on invalid transformation.

[neuron7xLab]

Quantize / Adapt / Calibrate Control Layer

Status: QAC_LAYER_SPEC_ADDED / EXECUTION_PENDING / EVIDENCE_REQUIRED

Purpose: convert continuous measurements into bounded operational states, adapt model behavior under dynamic regimes, and calibrate thresholds/sensitivity without turning GeoSync into a trading engine.

1. Quantize

Convert continuous inputs into explicit states.

Required state domains:

coherence_state: LOW | MID | HIGH | SATURATED
curvature_state: STABLE | STRESSED | CRITICAL | INVALID
energy_state: BOUNDED | ELEVATED | VIOLATED
margin_state: HEALTHY | WARNING | LIQUIDATION_NEAR | INVALID
oracle_state: GREEN | RED | PENDING | VACUOUS
claim_state: UNSUPPORTED | BOUNDED | VERIFIED | REJECTED

Rules:

• every state must declare source metric, threshold, denominator, time/window, and artifact path;
• no raw float may directly promote a claim;
• state transition must be deterministic and replayable;
• ambiguous values enter INVALID or PENDING, never silent success.

Artifacts:

artifacts/qac/state_quantization.json
artifacts/qac/state_transitions.json

2. Adapt

Change behavior under regime drift without changing scientific claims.

Required adaptive inputs:

input_distribution_drift
graph_topology_drift
curvature_residual_drift
synchronization_regime_shift
cross_margin_stress_shift
oracle_replay_gap

Allowed adaptations:

• widen diagnostic sampling window;
• increase falsifier coverage;
• lower claim confidence;
• switch verdict from VERIFIED to BOUNDED or RESEARCH_ONLY;
• require revalidation before merge/release.

Forbidden adaptations:

• raising confidence under drift;
• changing thresholds without calibration evidence;
• suppressing failing falsifiers;
• converting research output into buy/sell action.

Artifacts:

artifacts/qac/adaptation_decisions.json
artifacts/qac/drift_response_report.md

3. Calibrate

Tune sensitivity, thresholds, and tolerances using evidence, not vibes wearing a lab coat.

Required calibrated thresholds:

coherence_R_thresholds
curvature_stress_thresholds
energy_violation_bounds
margin_haircut_sensitivity
liquidation_distance_bands
claim_promotion_bands
oracle_replay_coverage_threshold

Calibration protocol:

• declare baseline fixture set;
• run grid or Bayesian calibration;
• report false positive / false negative tradeoff;
• pin threshold version;
• require regression test for each calibrated boundary.

Artifacts:

artifacts/qac/calibration_report.json
artifacts/qac/threshold_registry.yaml
artifacts/qac/calibration_verdict.md

Acceptance gates

quantization_is_deterministic == true
state_transitions_are_hash_replayable == true
drift_response_is_fail_closed == true
calibration_thresholds_are_versioned == true
claim_promotion_requires_calibrated_state == true
invalid_or_ambiguous_state_never_promotes == true

Final target

QAC_LAYER_OPERATIONAL
DISCRETE_STATES_REPLAYABLE
ADAPTATION_FAIL_CLOSED
CALIBRATION_VERSIONED
PRODUCT_READY_BLOCKED_UNTIL_QAC_ARTIFACTS_EXIST

[neuron7xLab]

Filtering Layer — Signal and Data Hygiene Gate

Status: FILTERING_SPEC_ADDED / EXECUTION_PENDING / FAIL_CLOSED_REQUIRED

Purpose: remove redundant, stale, invalid, contradictory, duplicated, contaminated, or non-actionable data/signals before they enter graph construction, curvature metrics, synchronization dynamics, energy/risk metrics, cross-margin validation, QAC state transitions, or governance evidence.

This is a hygiene/integrity gate, not a trading signal generator.

1. Inputs

Required input classes:

• raw market observations
• normalized market windows
• graph candidates
• curvature measurements
• Kuramoto trajectories
• energy/risk metrics
• cross-margin records
• CI/governance evidence
• replay ledger entries
• artifact manifests

Every input must carry:

• source
• timestamp
• unit
• schema version
• provenance hash
• freshness window
• transformation lineage
• declared denominator where applicable

2. Filter categories

Filtering must classify each rejected item into exactly one primary reason:

• MISSING_REQUIRED_FIELD
• INVALID_UNIT
• NON_MONOTONIC_TIMESTAMP
• STALE_MARKET_DATA
• DUPLICATE_RECORD
• NAN_OR_INF
• OUT_OF_RANGE_VALUE
• SCHEMA_VERSION_MISMATCH
• PROVENANCE_HASH_MISSING
• UNDECLARED_TRANSFORMATION
• LOW_SIGNAL_DENSITY
• CONTRADICTORY_EVIDENCE
• ORACLE_UNSUPPORTED
• UNNORMALIZED_METRIC
• UNQUANTIZED_STATE
• UNSUPPORTED_MARGIN_INPUT

No silent drop is allowed.

3. Core invariants

• No metric enters downstream layers without unit + denominator.
• No graph is built from stale, duplicated, or non-monotonic observations.
• No curvature value is accepted if derived from an invalid graph.
• No Kuramoto trajectory is accepted if phase input contains NaN/Inf or undeclared scaling.
• No cross-margin record is accepted without collateral basis, haircut basis, and mark freshness.
• No governance evidence is accepted without artifact path and provenance hash.
• No claim-state promotion is allowed from filtered or oracle-unsupported evidence.

4. Required artifacts

Filtering must emit:

• artifacts/filtering/input_inventory.json
• artifacts/filtering/rejected_records.jsonl
• artifacts/filtering/accepted_records.jsonl
• artifacts/filtering/filter_summary.json
• artifacts/filtering/signal_density_report.json
• artifacts/filtering/provenance_filter_report.md
• artifacts/filtering/filter_verdict.md

Artifacts must be hash-stable and replayable.

5. Metrics

Required metrics:

• raw_input_count
• accepted_count
• rejected_count
• rejection_rate
• duplicate_rate
• stale_rate
• schema_error_rate
• nan_inf_rate
• low_signal_density_rate
• contradiction_rate
• oracle_unsupported_rate
• accepted_signal_density
• filter_replay_hash_match

Required gates:

• filter_replay_hash_match == 1.0
• nan_inf_rate == 0 after filtering
• accepted_count > 0
• accepted_signal_density >= declared_minimum
• oracle_unsupported_rate == 0 for promoted claims

6. Tests and falsifiers

Must include tests for:

• NaN/Inf rejection
• stale timestamp rejection
• duplicate collapse
• missing unit rejection
• schema mismatch rejection
• invalid margin input rejection
• invalid graph input rejection
• contradiction ledger unsupported evidence rejection
• replay hash stability
• no silent drop invariant

Inverse falsifiers:

• inject stale market mark -> filter fires
• inject duplicate records -> duplicate collapse + report
• inject undeclared unit -> fail closed
• inject invalid graph adjacency -> no curvature downstream
• inject unsupported evidence -> no claim promotion

7. Integration order

Filtering must run after raw ingestion and before:

1. scale normalization
2. QAC state quantization
3. graph construction
4. curvature computation
5. Kuramoto reconstruction
6. energy/risk metrics
7. cross-margin validation
8. replay ledger promotion
9. product-readiness verdict

8. Stop rules

Stop pipeline if:

• accepted_count == 0
• any NaN/Inf survives filtering
• stale records enter graph construction
• rejected records are not logged
• provenance hash is missing
• downstream layer consumes unfiltered input
• filtering thresholds are changed without calibration record

9. Target verdict

FILTERING_LAYER_OPERATIONAL
NO_SILENT_DROP
INVALID_SIGNALS_REJECTED
ACCEPTED_SIGNAL_SET_HASHED
PRODUCT_READY_BLOCKED_UNTIL_FILTER_ARTIFACTS_EXIST

[neuron7xLab]

Measurement & Profiling Layer

Status: MEASUREMENT_PROFILING_SPEC_ADDED / EXECUTION_PENDING

Goal: convert behavior into numeric, replayable evidence.

Metric contract

Every metric requires: metric_id, raw_value, normalized_value, unit, denominator, window, sample_count, source_artifact, method, threshold, status, sha256.

Required metrics: data quality, filter rejection rate, normalization coverage, state transition rate, curvature residual, Kuramoto order parameter, energy residual, cross-margin exposure, collateral coverage, oracle replay coverage, quarantine count, CI pass rate, artifact hash match rate, runtime seconds, peak memory MB.

Fail closed if: denominator missing, unit missing, source artifact missing, NaN/Inf emitted, threshold missing for gated metric, unsupported metric promoted as PASS.

Profiling contract

Required profiles: data profile, signal profile, graph profile, curvature profile, synchronization profile, energy profile, margin profile, CI profile, governance profile, resource profile.

Required artifacts: artifacts/profile/measurement_registry.json, artifacts/profile/*_profile.json, artifacts/profile/profile_verdict.md.

Acceptance: MEASUREMENT_LAYER_OPERATIONAL / PROFILE_ARTIFACTS_EMITTED / NO_METRIC_WITHOUT_DENOMINATOR / NO_PROFILE_WITHOUT_SOURCE_ARTIFACT / PRODUCT_READY_BLOCKED_UNTIL_PROFILE_VERDICT_GREEN.