Skip to content

CHANGELOG.md

Latest commit

 

History

History
159 lines (139 loc) · 8.78 KB

CHANGELOG.md

File metadata and controls

159 lines (139 loc) · 8.78 KB
Raw

Changelog

0.5.0 — Normalized observation and accountable-memory alignment

  • Upgraded the Sonic Matter Framework boundary to MASA 0.2.0 and masa-processing-request 0.2.0 while retaining GERM's canonical sound_id and optional, non-fatal sidecar behavior.
  • Added preferred cosmo/modulation/v0.2 support with explicit v0.1 frame compatibility. V0.2 frames must carry a local cosmo/signal-catalog/v0.2 reference and complete signal metadata, finite normalization ranges, a [0, 1] output, clamp behavior, and a stated basis before GERM routes them.
  • Added the bounded /api/signals bridge and public GET /cosmoaudition/signals endpoint, preserving source filtering while keeping all provider acquisition and credentials inside Cosmoaudition.
  • Added a distinct Hydrosphere Field to the backend manifest, dashboard, provenance, and smoke coverage so marine signals are not folded into atmospheric or geological observations.
  • Preserved catalog normalization through mapping receipts and dashboard provenance. The Mapping Loom now declares its second-stage normalized projection instead of leaving source-range metadata attached to a [0, 1] value.
  • Raised the Akousma dependency to Earworm 0.7.0 / schema 1.6 and documented Oída 0.10.0, gateway v0.6, Akousmata 0.7.0, and AKOÚŌ 0.9.2. GERM-created sounds remain cultivation records: accountable auditum data is preserved when supplied, never fabricated as a listening report.

0.4.1 — Sonic Matter Framework contract corrections

  • Fixed the Cosmoaudition frame bridge to consume the published outputValue field, expose an explicit GERM value, and retain the control's status, mapping parameters, confidence, and epistemic note. Real applied, held, and uncertainty routes are no longer silently withheld.
  • Kept canonical sound_id, Sonic Lineage, and Earworm identifiers stable when filenames change; refreshed written MASA sidecars after path changes and refused renames that would break an immutable Akousmata file locator. Rename companion writes are transactional, and matching deferred-latent locators follow their audio without changing identity.
  • Restricted bulk file deletion's primary path to supported audio formats, required the actual metadata companion, preserved Akousmata locators, and removed only verified written MASA sidecars under explicit companion-delete semantics.
  • Preserved Cosmoaudition frame-level signals and sources beside resolved routes, and withheld malformed or out-of-range executable controls with an explicit bridge reason.
  • Accepted finite JavaScript wire numbers whose integer spelling exceeds a signed 64-bit persistence value, so high-magnitude observations such as the Bitcoin hashrate cannot invalidate an otherwise valid modulation frame.
  • Aligned active project language to Sonic Matter Framework: MASA describes, Cosmoaudition observes and modulates, and GERM cultivates. The projects remain distinct.

0.4.0 — Sonic Matter Stack alignment

  • Fixed a MASA protocol break. Every sidecar cited smo.sonicfield.org, a pre-release identifier root. $schema is a protocol constant rather than a hint, so the published MASA 0.1.0 validator rejected every record GERM had ever written. Records now cite masa.sonicfield.org and validate against the released reference implementation.
  • Fixed a silent modulation bug. A zero mapping amount was scaled through instead of read as the operator turning a route off. On a reversed output range such as (760, 180) that emits outputRange[0] — the strongest value the mapping can produce — reported as applied. Silence requested, maximum delivered. A zero amount is now skipped / route-disabled, matching Cosmoaudition's own engine.
  • Added the MASA processing layer to the Micro modules. Each declares a granular or spectral operation in MASA's engine-neutral terms and can emit a portable masa-processing-request via POST /micro/processing-request; GET /micro/processing-operations reports the mapping. GERM states the intention and binds no DSP library.
  • Added the Cosmoaudition modulation framework. The bridge allowlist now covers /api/modulation, /api/frame, and /api/snapshot/masa, and GET /cosmoaudition/frame resolves one verified cosmo/modulation/v0.1 frame into GERM routes. It reads the frame's controls, never the bare values map, so a value never travels without the decision that produced it; withheld routes and emitted absences are reported rather than dropped. /api/stream stays unbridged because Server-Sent Events cannot be read by a bounded request/response client.
  • Added uncertaintyOutput to Cosmoaudition mappings, so a map-uncertainty policy emits its declared value under an uncertainty status instead of degrading silently to skip. Declaring the policy without the value is now refused at the schema boundary.
  • Named GERM's place in the Sonic Matter Stack alongside the Listening Stack, and retired the earlier "Observatory" naming across the server, dashboard, documentation, and module labels.

0.3.3 — Additive cultivation lineage

  • Raised the Earworm/Akousma floor to 0.6.1 so remembered generations cannot silently replace an earlier listening account.
  • Clarified that a generated child is a new cultivated object with causal lineage, not evidence that a claim about its parent was true.
  • Aligned public stack versions and expanded environment-file ignore coverage.

0.3.2 — Cosmoaudition status privacy

  • Keep Cosmoaudition health-check failures in local logs while returning only a stable public error at the loopback API boundary.
  • Add a regression test proving low-level backend details cannot enter the status payload.

0.3.1 — Security and stack alignment

  • Overrode Stable Audio 3's upstream Torch 2.7.1 constraint with the locally validated Torch and Torchaudio 2.10 pair, removing every fixable advisory from the optional Python-provider dependency graph.
  • Added an all-extras dependency audit and a dated security exception for the two remaining upstream PyTorch findings in APIs GERM does not call directly.
  • Updated Setuptools to 83.0.0 to close its Unicode-normalization sdist issue.
  • Removed unsafe DOM-to-HTML and DOM-to-download flows, bounded note parsing, kept backend exception details out of API payloads, and confined upload writes to managed roots.
  • Updated the embedded Earworm/Akousma package from 0.4.0 to 0.6.0 and aligned the documented Listening Stack versions with the canonical public releases.

0.3.0 — Cosmoaudition, Matter Analysis, and audio reliability

  • Added the loopback-only Cosmoaudition bridge, explicit observation mappings, archives, module palette, Matter Analysis, and optional MASA 0.1 sidecars.
  • Kept unavailable, errored, stale, and zero-valued observations distinct; missing Mapping Loom inputs now execute their declared policy without a fabricated neutral value.
  • Streamed bounded spectral aggregation with cached FFT windows, validated Matter lineage shapes, and exposed the final MASA companion state in API and dashboard results.
  • Flushed queued AudioWorklet PCM before recorder shutdown, preserved explicit zero-gain wavetable previews, removed confirmed dead dashboard helpers, and expanded backend, audio, and browser regression coverage.

0.2.5 — Backend integrity and integration hardening

  • Hardened request, persisted JSON, path, upload, WAV, and provider artifact validation across generation, editing, control, listener, session, Micro, wavetable, library, and file workflows.
  • Made provider execution, cancellation, job retention, caches, metadata, lineage, Akousmata companion writes, and multi-step render transactions bounded and failure-safe.
  • Updated Stability and Gemini integrations to their current contracts, strengthened Python and MLX provider output checks, and aligned the test client dependency with Starlette's httpx2 backend.
  • Improved dashboard data escaping, native macOS daemon lifecycle handling, launch scripts, and regression coverage for integration boundaries.

0.2.0 — Listening-informed cultivation

  • Added OÍDA re-listening, prompt derivation, immutable evidence summaries, covenant-aware retention, and Akousmata lineage updates.
  • Aligned generated records with Earworm 0.4 / akousma spec v1.3 and akouo/v0.7.
  • Added the native macOS shell, daemon supervision, server-owned sessions, improved provider diagnostics, and a streamlined dashboard toolbar.
  • Expanded Stable Audio API and MLX provider behavior, Micro/Matter state, file operations, tests, and release validation.

0.1.0 — Initial modular lab

  • Published the local FastAPI server, browser dashboard, generation/edit routes, providers, Micro/Matter modules, control layer, Wavetable Forge, Strains, library, and Earworm export.