v0.2.0-alpha.3 — Partner-ready review package snapshot

This prerelease packages hrm-weight-to-phase-validation as a partner-ready and reviewer-ready simulation-only HRM photonic AI planning toolkit.

New:

• Added Partner Readiness documentation
• Added Lab Data Request
• Added Foundry Data Request
• Added Hardware Evidence Checklist
• Added External Review Checklist
• Added Reproducibility Capsule
• Added Solo Completion Audit
• Added v0.2.0-alpha.3 readiness report
• Added model cards for portfolio candidates
• Added model portfolio explainer
• Improved CLI with doctor, portfolio, and model-card commands
• Improved static dashboard with partner-ready review sections

Current solo-completion estimates:

• Software completeness: 0.96
• Planning toolkit completeness: 0.93
• Partner readiness: 0.90
• Hardware validation completeness: 0.0
• Chip readiness: 0.05

Current status:

• Ready for external review: true
• Ready for partner discussion: true
• Ready for hardware claims: false

Blocked hardware gates:

• Stage 5: no_foundry_calibrated_device_model
• Stage 6: no_measured_hrm_transfer_matrix
• Stage 7: no_end_to_end_hardware_benchmark

Claim boundary:
This release remains simulation-only.
It does not claim hardware validation, foundry calibration, measured transfer matrices, production inference readiness, real hardware latency, real hardware energy efficiency, physical accuracy, quantum advantage, hardware-native intelligence, or power-free computation.

v0.2.0-alpha.2 — Review and usability hardening snapshot

This prerelease hardens the review and usability surface of the simulation-only HRM photonic AI planning toolkit.

New:

• Added simulation-only disclaimers to hrmwtp CLI commands
• Added repository-local CLI wrapper at scripts/hrmwtp.py
• Preserved machine-readable JSON stdout for summary and decision commands while printing disclaimers on stderr
• Added safe CLI smoke path with hrmwtp check --dry-run
• Expanded the static dashboard with what-this-is / what-this-is-not sections, key report links, stage status, model portfolio summary, hardware gate blockers, and a claim-boundary box
• Added docs/QUICKSTART.md
• Added docs/REVIEWER_GUIDE.md
• Updated README first-minute orientation and report links
• Added tests for CLI smoke paths, dashboard claim boundary, guide docs, and artifact hash coverage

Verification:

• make check passed
• 177 tests passed
• JSON validation passed
• SHA-256 artifact verification passed
• artifact hash coverage passed
• local-path hygiene passed
• claim-boundary guard passed

Claim boundary:
This release remains simulation-only. It does not claim hardware validation, foundry calibration, measured transfer matrices, production inference readiness, real hardware latency, real hardware energy efficiency, physical accuracy, quantum advantage, hardware-native intelligence, or power-free computation.

Stage 5, Stage 6, and Stage 7 remain blocked.

v0.2.0-alpha.1 — Simulation-only planning toolkit snapshot

This prerelease extends hrm-weight-to-phase-validation from a stable simulation-only planning framework into the first v0.2 planning-toolkit snapshot.

New:

• Added model portfolio benchmark and ranking for deterministic fixtures: tiny_mlp, projection_chain, low_rank_adapter_demo, sparse_linear_demo, and transformer_block_manifest_only
• Added model-portfolio-decision-summary.md
• Added optional model-export adapter protocol without a hard PyTorch dependency
• Added generated manifest validation demo for external exporter output
• Added hardware design-space sweep over phase bits, insertion loss, phase noise, and calibration interval proxies
• Added design-space analysis and Pareto summary
• Added synthetic transfer-matrix ingestion sandbox with normalization and target-vs-observed comparison logic
• Added hrmwtp CLI commands: check, regenerate, summary, decision, and list-reports
• Added static dashboard at dashboard/index.html
• Updated README, v0.2 roadmap, validation summary, evidence ledger, ARTIFACTS, review pack, readiness reports, Makefile, CI, and tests

Current planning summary:

• Best portfolio candidate: low_rank_adapter_demo
• Worst portfolio candidate: transformer_block_manifest_only
• Design-space sweep rows: 108
• Pareto candidates: 4
• Transfer-matrix sandbox: syntheticFixtureOnly=true and publicMeasuredEvidence=false

Claim boundary:
This release remains simulation-only. It does not claim hardware validation, foundry calibration, measured transfer matrices, production inference readiness, real hardware latency, real hardware energy efficiency, quantum advantage, hardware-native intelligence, or power-free computation.

Stage 5, Stage 6, and Stage 7 remain blocked.

v0.1.0 — Stable simulation-only HRM photonic AI planning framework

This release packages hrm-weight-to-phase-validation as a stable simulation-only planning framework for HRM photonic AI mapping.

Included capabilities:

• HRM neural weight-to-phase mapping specification
• SVD mapping demo
• abstract mesh-constrained mapping
• rectangular matrix support
• complex/unitary factor simulation
• matrix-family benchmark suite
• multi-layer toy inference
• model-weight manifest import and eligibility analysis
• scaling and larger-layer benchmark
• model suitability profiler
• parametric hardware scenario estimator
• hardware requirements envelope
• simulation-only error budget
• calibration and transfer-matrix assimilation planning
• model-to-HRM decision report
• release-readiness audit
• v0.2 roadmap

Current decision:

• Model suitability score: 87.647
• Decision: blocked_by_missing_hardware_evidence
• Missing evidence:
  • foundry-calibrated device model
  • measured HRM transfer matrix
  • end-to-end hardware inference benchmark

Claim boundary:
This release is simulation-only.
It does not claim hardware validation, foundry calibration, measured transfer matrices, production inference readiness, real hardware latency, real hardware energy efficiency, quantum advantage, hardware-native intelligence, or power-free computation.

Stage 5, Stage 6, and Stage 7 remain blocked.

v0.1.0-rc.1 — Simulation-only HRM photonic AI planning framework

This release candidate packages hrm-weight-to-phase-validation as a simulation-only planning framework for HRM photonic AI mapping.

Included capabilities:

• HRM neural weight-to-phase mapping specification
• SVD mapping demo
• abstract mesh-constrained mapping
• rectangular matrix support
• complex/unitary factor simulation
• matrix-family benchmark suite
• multi-layer toy inference
• model-weight manifest import and eligibility analysis
• scaling and larger-layer benchmark
• model suitability profiler
• parametric hardware scenario estimator
• hardware requirements envelope
• simulation-only error budget
• calibration and transfer-matrix assimilation planning
• model-to-HRM decision report
• release-readiness audit
• v0.2 roadmap

Current decision:

• Model suitability score: 87.647
• Decision: blocked_by_missing_hardware_evidence
• Missing evidence:
  • foundry-calibrated device model
  • measured HRM transfer matrix
  • end-to-end hardware inference benchmark

Claim boundary:
This release candidate is simulation-only.
It does not claim hardware validation, foundry calibration, measured transfer matrices, production inference readiness, real hardware latency, real hardware energy efficiency, quantum advantage, hardware-native intelligence, or power-free computation.

Stage 5, Stage 6, and Stage 7 remain blocked.

v0.1.0-alpha.8 — Multi-layer toy inference snapshot

This prerelease adds a simulation-only multi-layer toy inference pipeline to the HRM weight-to-phase validation ladder.

New:

• Added layer-stack-inference-demo.json
• Added layer-stack-error-analysis.json
• Added deterministic tiny_mlp_4_6_3 toy model
• Added deterministic projection_chain_4_8_4 toy model
• Mapped linear layers through the abstract HRM simulation path
• Kept bias additions classical outside the optical mesh
• Kept ReLU classical outside the optical mesh
• Added layer-wise and cumulative error reporting
• Added activation-boundary notes
• Integrated reports into Summary, Ledger, ARTIFACTS, Makefile, CI, and tests

Current toy inference summary:

• Primary model: tiny_mlp_4_6_3
• Secondary model: projection_chain_4_8_4
• Primary output relative error: 0.061296868009442
• Worst model by output error: tiny_mlp_4_6_3
• Largest layer-boundary relative error: relu_1
• ReLU is classical and outside the optical mesh
• opticalNonlinearityImplemented=false
• errorsCompoundAcrossLayers=false for these deterministic toy cases

Claim boundary:
This release remains simulation-only. It does not claim hardware validation, foundry calibration, measured transfer matrices, physical accuracy, optical nonlinearities, full neural-network acceleration, quantum advantage, hardware-native intelligence, power-free computation, production inference readiness, physical phase synthesis, foundry layout synthesis, or a completed hardware benchmark.

Stage 5, Stage 6, and Stage 7 remain blocked.

v0.1.0-alpha.12 — v0.1.0 release-readiness hardening snapshot

This prerelease prepares the simulation-only HRM weight-to-phase validation ladder for v0.1.0 release-candidate review.

New:

• Added release-readiness-v0.1.0.json
• Added release-readiness-v0.1.0.md
• Added scripts/check_claim_boundary.py
• Added v0.1.0 readiness audit covering documentation, report indexing, artifact hashes, local-path hygiene, claim-boundary enforcement, and blocked hardware gates
• Enforced the claim-boundary guard in Makefile and CI
• Updated pyproject.toml to 0.1.0-alpha.12
• Verified Stage 5, Stage 6, and Stage 7 remain blocked

Verification:

• make check
• unit tests
• JSON validation
• SHA-256 artifact hash validation
• artifact hash coverage
• local-path hygiene
• claim-boundary guard
• whitespace check

Claim boundary:
This release does not claim hardware validation, foundry calibration, measured transfer matrices, physical accuracy, real hardware latency, real hardware energy efficiency, quantum advantage, hardware-native intelligence, power-free computation, production inference readiness, or a completed hardware benchmark.

Stage 5, Stage 6, and Stage 7 remain blocked. The project remains simulation-only and future-work only.

v0.1.0-alpha.7 — Matrix-family benchmark snapshot

This prerelease adds a deterministic matrix-family benchmark suite to the simulation-only HRM weight-to-phase validation ladder.

New:

• Added matrix-family-benchmark.json
• Added matrix-family-analysis.json
• Added deterministic benchmark families:
  • identity_4x4
  • diagonal_dynamic_range_4x4
  • low_rank_6x4
  • rank_deficient_5x3
  • ill_conditioned_4x4
  • sparse_like_6x6
  • dense_seeded_4x4
  • rectangular_tall_8x4
  • rectangular_wide_4x8
  • complex_phase_dominant_4x4
  • unitary_like_4x4
• Added best/worst case analysis
• Added family ranking by reconstruction error
• Added condition-sensitivity proxy ranking where applicable
• Integrated reports into Summary, Ledger, ARTIFACTS, Makefile, CI, and tests

Current benchmark summary:

• caseCount: 11
• easiest/control case: identity_4x4
• hardest case: rectangular_tall_8x4
• max mesh-constrained error / max delta: 0.099951757092972
• average mesh-constrained error: 0.045353461797016

Claim boundary:
This release remains simulation-only. It does not claim hardware validation, foundry calibration, measured transfer matrices, physical accuracy, quantum advantage, hardware-native intelligence, power-free computation, production inference readiness, physical phase synthesis, foundry layout synthesis, or a completed hardware benchmark.

Stage 5, Stage 6, and Stage 7 remain blocked.

v0.1.0-alpha.6 — Complex/unitary factor simulation snapshot

This prerelease adds supplemental complex/unitary simulation support to the HRM weight-to-phase validation ladder.

New:

• Added complex-unitary-mesh-support.json
• Added deterministic complex cases:
  • unitary_like_2x2
  • complex_4x4
  • phase_dominant_4x4
• Added complex QR unitary-factor handling
• Added relative Frobenius, phase-aware, amplitude-aware, and unitary-factor deviation metrics
• Integrated report into Summary, Ledger, ARTIFACTS, Makefile, CI, tests, and README/docs

Claim boundary:
This remains abstract simulation only. It does not claim hardware validation, foundry calibration, measured transfer matrices, physical phase synthesis, production inference readiness, quantum advantage, or a completed hardware benchmark.

Stage 5, Stage 6, and Stage 7 remain blocked.

v0.1.0-alpha.5 — Rectangular matrix support snapshot

This prerelease adds supplemental rectangular matrix support to the simulation-only HRM weight-to-phase validation ladder.

New:

• Added rectangular-matrix-support.json
• Added tall 6x4, wide 4x6, and rank-deficient 5x3 deterministic cases
• Added compact SVD, full orthogonal completion, and rectangular Sigma transfer-core representation
• Integrated rectangular support into the validation summary, evidence ledger, artifact hashes, tests, Makefile, and CI
• Added artifact hash coverage guard to ensure every generated JSON report is listed in ARTIFACTS.sha256
• Tightened README wording around rectangular support and remaining limitations

Current simulation scope:

• Stage 0: future-work specification
• Stage 1: numerical SVD mapping demo
• Stage 2: abstract mesh-constrained approximation with supplemental rectangular matrix support
• Stage 3: uncalibrated perturbation model, deterministic sweeps, and sweep analysis
• Stage 4: synthetic/oracle calibration simulation, calibration sweeps, and analysis

Blocked hardware evidence gates:

• Stage 5: no foundry-calibrated device model
• Stage 6: no measured HRM transfer matrix
• Stage 7: no end-to-end hardware inference benchmark

Claim boundary:
This release does not claim hardware validation, foundry calibration, measured transfer matrices, physical accuracy, quantum advantage, hardware-native intelligence, power-free computation, production inference readiness, physical phase synthesis, foundry layout synthesis, or a completed hardware benchmark.

The project remains future-work only and does not improve hardware readiness.