Skip to content

QEC v177.0 — Acoustic Syndrome Transport and Verified Correction

Choose a tag to compare

@EmergentMonk EmergentMonk released this 14 Oct 14:25
1780772

QEC v177.0 — Acoustic Syndrome Transport and Verified Correction

Cumulative release notes · 14 October 2026

These notes cover the implementation merged into main through PR #607, at merge commit 178077246e615bbada8589d0637ac75157cdc2f0 on 14 October 2026 at 14:00:41 UTC. Its source tree is identical to reviewed and tested PR head 4be798b785ea687baf3fd87f25bb6effdd4dbcb2.

v177.0 is the release designation used here. The checked-in package metadata still reports 173.0.0; the 177.0–177.5 labels in the acoustic contracts identify implementation milestones within this combined release. These notes do not create a tag, publish a GitHub Release or change package metadata.

QEC now has an end-to-end, opt-in acoustic syndrome experiment: encode declared qutrit or packed-ququart measurement symbols as actual PCM, apply deterministic sample-level interference, recover candidate symbols, independently verify the recovered measurement, and invoke the existing exact decoder only when that verification succeeds. The original waveform, noisy waveform, filtered waveform, receiver candidate, rejected evidence and verified correction remain separately inspectable and replayable.

The implemented claim boundary is classical synthetic PCM plus exact finite Pauli/stabilizer software fixtures. It does not establish quantum-state evolution, measured acoustic performance, physical quantum protection or authenticated measurement provenance.

Cumulative scope and reading guide

This is a cumulative release document: the inherited system below remains the foundation of the new acoustic work, and the v177.0 sections describe the additions and corrections in PRs #601–#607. The companion historical archive preserves every published release record found in the repository inventory, rather than compressing earlier release notes into a claim that all historical experiments are active runtime features.

Published tags, package versions and contract milestone labels are separate. In particular, v176.5 is the published combined switching release; the v173.1–v176.5 labels describe its included implementation milestones, not a sequence of separately published tags. Some README, package-description and contract introduction text still carries older frontier or “unreleased” wording. Published release records establish publication status; the exact merged source and scoped contracts establish the capabilities described here. Roadmap proposals do not become implemented features merely because their version numbers appear in this document.

Inherited capabilities included in the cumulative release

Deterministic proof, validation and research boundaries

QEC retains its canonical JSON and SHA-256 evidence architecture: declared inputs become versioned, hash-bound artifacts, and validators reconstruct the relevant content and execution rather than accepting embedded claims. The documented system includes detached/frozen evidence, self-hash exclusion, child-before-aggregate validation, replay checks, explicit failure modes, bounded inputs, source-bound claim handling and adapter-only external boundaries. Hash agreement is byte/contract evidence, not independent proof that a physical assertion is true.

The retained system overview records the earlier receipt families: GameWorld interactions; perturbation and stress; substrate constraints; bounded recursive proof loops; reality-loop composition; global validation/truth/replay; IRC operator replay; invariant discovery; optimization; simulation backends; dataframe and columnar backends; research-automation provenance; inference/tokenization/compression; agent-observation and skill-library boundaries; and quantum-memory, phase-estimation and geometry research signals. These remain evidence and adapter contracts with their stated limits, not a declaration that external tools or generated research are authoritative or that all historic experimental runtimes are enabled. The complete earlier version-by-version record is supplied separately.

Decoder foundations and exact finite-code evidence

The historical decoder and its frozen baseline remain preserved. The opt-in qutrit decoder uses exact GF(3) symplectic stabilizer algebra and bounded syndrome lookup. Its retained code set includes cyclic [[5,1,3]]₃, Shor [[9,1,3]]₃ and ternary Golay [[11,1,5]]₃, with certified bounded sets of 40, 72 and 3,608 non-identity errors respectively. The Golay boundary includes exact weight-one and weight-two enumeration, not a general physical-noise threshold.

The earlier qutrit harmonic layer distinguishes informative H1/H2 observations from state-dark H3 behavior, rejects ambiguous or discordant observations, and provides collective-mode analysis and typed ETQ-303 observation addresses. Its offline Qutrit Harmonic Lab is a demonstration/inspection surface. The new v177 PCM path is additive; it does not replace that algebra or reinterpret the symbolic/harmonic evidence as physical sound-based quantum protection.

The v170 line adds the exact packed [[5,1,3]]₄ ququart model, exact finite FER enumeration, deterministic Monte Carlo and harmonic fault batteries, separately reported receiver/decoder telemetry, report-claim validation, replication receipts and NEXUS execution/replication boundaries. Packed ququart arithmetic retains its two binary lanes. Finite FER evidence, browser demonstrations and replication receipts remain distinct from hardware thresholds, physical fault tolerance or quantum advantage. See the retained foundation and lineage summary.

Decoder governance and symbolic sonification

The completed v166.0–v166.8 governance chain covers the canonical decoder baseline, adapter-only candidates, replay and fast-path equivalence, optimization and implementation boundaries, benchmark ladders, rollback readiness and promotion eligibility receipts. Promotion eligibility does not silently replace decoder source or activate a new runtime.

v167.0 supplies the deterministic symbolic-sonification event skeleton. The former v167.1–v167.9 mapping packs, graph/music mappings, MIDI-export, prompt, telemetry, CLI and benchmark assignments remain unimplemented historical plans. The v177 acoustic experiment has its own PCM, receiver and evidence contracts; it does not retrospectively complete that deferred symbolic programme. These distinctions are retained in the current foundation declaration.

Six native switching architectures

The combined v176.5 release is the frozen switching input/transport baseline for v177. It brings the following architecture families forward:

Architecture Included capability and preserved boundary
Strowger, v170.3.0 Mixed-radix pulse routing, first-free trunk hunting, route/check/state-dark tone checks, canonical operator events and replayable pre-route state. Its lack of native transported payload bytes remains an explicit capability gap.
Panel, v171.x Complete request register and sender programme before actuation; separate control/payload identities; deterministic translation; motor/bank fault and capacity evidence; offline demonstration lab.
Crossbar, v172.0–v172.5 Immutable matrix coordinates, marker/common control, bounded multi-stage selection, owned reservation/release/quarantine, independent complete-path continuity and the three-era equivalence battery.
ESS, v173.0 and additive v173.1–v173.5 Preserved independent-planning skeleton plus source-bound programme/translation manifests, actual native call reservations and releases, exact lifecycle states, logical scheduling, bounded forwarding/retry/fan-out and scoped two-fabric equivalence.
Digital, v174.0–v174.5 Fixed-width frames, deterministic owned time-slot allocation and release, synchronization and slip rejection, time-slot interchange, actual payload reassembly, fault receipts and separate offline SVG/WAV demonstrations.
Packet, v175.0–v175.6 Canonical payload envelopes, exact binary-prefix route selection, bounded priority queues, hop-by-hop evidence, declared faults and alternate-route failover, verified reassembly, offline non-authoritative softswitch observations and a separate evidence lab.

The precise contracts are ESS call processing, Digital switching, Packet/softswitch and the switching lineage. These are deterministic software models. Their historical design inspiration does not imply faithful emulation of particular telephone hardware or a live deployed network.

Cross-era migration, equivalence and selection

The v176 combined work executes the actual six native models against a common corpus: 36 shared availability cases plus 15 architecture-specific fault controls, giving 306 native executions. Migration binds complete native inputs, source/programme identities, topology and policy. Replay compares declared destinations, admitted/rejected outcome classes and payload identity only where natively supported, while retaining architecture-specific failure reasons and lifecycle differences.

Architecture selection uses declared requirements and capabilities; it does not assume that a newer design is better. Capacity cases describe independent initial-contention snapshots, not universal concurrent throughput. Reports compare event counts, receipt sizes, complexity and modeled capabilities without turning them into hardware-speed or latency measurements. See the cross-era contract.

Existing interfaces, laboratories and dependency boundaries

The retained package entry points include capability inspection, experiment commands, qutrit/ququart demos and benchmark batteries, ququart report validation, NEXUS bridge/replication validation, and Strowger, Panel, Crossbar and ESS command interfaces. Python 3.10 or later is declared; NumPy and SciPy are core dependencies. Scientific and external-backend extras remain optional and separately declared.

Offline laboratories and visual/sonified exports help inspect evidence. Their rendered interactions and audible behavior must be distinguished from Python replay and Node stub/DOM test results. Optional external backends and live observations remain adapters, not authorities. The QEC OS runtime and benchmark-reset programme remains deferred to v193.x; v177 does not activate it.

New in v177.0

Highlights

  • A source-bound, integer-only multitone phase alphabet with canonical PCM16 and WAV output.
  • Twelve deterministic noise and distortion families, explicit reference cancellation, matched-filter and robust receiver alternatives, and classical repetition FEC.
  • An independently pinned measured-syndrome gate before the unchanged qutrit and packed-ququart exact decoders.
  • Complete native Digital and Packet replay of selected acoustic artifact bytes, with separate identities for all six evidence stages.
  • A 38-trial deterministic comparison with explicit resource costs, erasures, raw false accepts and exact-model outcomes.
  • Ordinary WAV/JSON exports and a dependency-free offline viewer using the actual imported samples.
  • A separately reviewed SciPy eigensolver reproducibility repair and expanded cross-platform evidence checks.
  • Merged exporter corrections for generation-time source provenance, bounded metadata, exclusive output ownership and identity-guarded filesystem access.

Source-bound acoustic channel

Channel contract

The new qec.experiments.acoustic.channel module defines immutable channel configuration, PCM and source objects. Clean signal generation uses a frozen 48-entry sine lookup table and exact integer arithmetic. It does not depend on runtime trigonometry, floating-point reductions or random-number state.

The channel supports mono signed PCM16 at 24 or 48 kHz, corresponding to declared 500 or 1,000 Hz carriers. Configurations select harmonic orders, one to eight complete periods per block and one to three repeated blocks per symbol. Inputs are bounded to 32 symbols; invalid clocks, malformed PCM, unsupported values and ambiguous integer types fail explicitly.

Qutrit harmonics distinguish informative orders from state-dark multiples of three. Packed-ququart harmonics distinguish informative odd orders, parity-only orders and state-dark multiples of four. The four acoustic labels remain separate from the decoder's packed GF(2) × GF(2) algebra.

Frozen qutrit and ququart JSON/WAV fixtures bind exact bytes. Independent WAV parsing and Fourier-projection checks exercise the clean alphabet. Strict source-pinned replay reconstructs every serialized field, so changing an artifact and recomputing its outer hash is insufficient to validate it.

Deterministic interference and reference cancellation

Noise and cancellation contract

Noise plans apply actual sample transforms in a declared order, with bounded parameters, a versioned 32-bit generator and explicit int16 saturation. The twelve families are broadband, colored and correlated noise; synthetic speech-like overlap; impulses; clipping; timing slips; missing samples; phase drift; nonlinear distortion; echo; and aliasing.

Each stage retains input and output identities, sample lengths, changed-sample counts, saturation counts and generator state. Operation order is meaningful, and irrecoverable clipping or deletion remains visible.

Reference cancellation accepts only the observed waveform, a separately supplied reference waveform and a declared configuration. It has no access to clean truth, hidden errors or desired symbols. Delay, rational gain, 0°/180° polarity, reference identity, role and permission are explicit. Arbitrary phase rotation, automatic target fitting and general speech separation are outside this implementation.

A known additive generator can provide a privileged synthetic reference without reading the clean target. That extra information and its unmeasured acquisition cost are declared in comparisons. Wrong-phase and unrelated-reference controls retain cases where subtraction worsens the signal.

PCM receiver and classical FEC

Receiver contract

The receiver consumes received PCM, public configuration, a symbol count and a frozen policy. It does not receive ideal symbols, measurement truth, noise plans, hidden errors or a desired answer.

Integer harmonic projections are checked against public state templates using explicit distance and margin gates. Informative orders must agree, partial orders constrain the candidate, state-dark orders are checked against their invariants, and a full-waveform residual guard rejects excessive unexplained energy. Missing, weak, ambiguous, discordant and uninformative observations remain explicit rejections.

The standard matched-filter estimator is accompanied by a bounded median-period alternative. Repetition policies include first-block-only, unanimous and strict-majority decisions. Every compared method receives the same transmitted repetition word; ignoring repetitions does not earn an artificial energy or duration discount.

Receiver acceptance is retained separately from verified recovery. Coherent phase shifts can map one valid waveform into another valid symbol and pass harmonic checks. Those raw false accepts remain in the evidence even when the independent reference later vetoes correction.

Independently verified correction

Syndrome bridge contract

The new bridge replays the receiver observation from actual PCM and checks recovery against an independently retained measured-syndrome reference and digest pin. Malformed evidence, substituted inputs and mismatched measured symbols fail before a decoder call. A correctly recovered syndrome outside the bounded decoder table remains an attempted but rejected correction.

Verified qutrit measurements reach the existing cyclic [[5,1,3]]_3 decoder. Packed-ququart labels are unpacked into the existing eight binary checks of the two-lane [[5,1,3]]_4 model. The exact decoders, correction tables and underlying Pauli/stabilizer algebra are unchanged.

The bounded fixture tests cover 41 qutrit cases and 76 packed-ququart cases: identity plus every single-site Pauli error on five sites. Each case passes through PCM encoding, sample-level perturbation, reception, independent checking and exact correction. Success requires an accepted correction and a residual in the stabilizer group; zero residual syndrome alone is insufficient.

Hidden error information is confined to the separate post-decoding fixture evaluator. Negative controls include no-tone rejection, coherent wrong-symbol acceptance followed by reference veto, forged or rehashed observations, changed reference authority, out-of-table syndromes and zero-syndrome non-stabilizer logical operators. A consistently false caller-approved reference can still defeat the authority assumption; hashes cannot establish physical truth.

Native Digital and Packet transport

Transport contract

The additive transport adapter carries complete canonical UTF-8 JSON bytes through the unchanged native Digital and Packet software models. It preserves six distinct roles: original, observed, filtered, candidate, rejected and verified.

Selected artifacts are limited to 65,536 bytes, split into at most 64 chunks of 1,024 bytes. Each chunk binds its source, role, ordinal, offset, length and actual-byte digest. Oversized artifacts fail explicitly; the adapter does not silently replace a waveform with its hash.

Validation reconstructs the exact partition, performs full native replay for every chunk and checks the complete reassembled bytes. Corrupt or missing chunks suppress the whole adapter output while retaining rejected native evidence. Valid rejection evidence is distinguishable from successful delivery.

The standard laboratory export transports all six actual stage payloads through Packet. A separate small Digital specimen carries a compact binding record only. Successful transport proves the selected bytes survived the declared model; it does not authenticate their origin or promote a candidate into verified acoustic evidence.

Comparative evidence

Comparison and export contract

The default battery contains 19 named controls for each of two alphabets, producing 38 deterministic trials. It compares:

  1. Unfiltered matched-filter recovery without FEC.
  2. Reference cancellation without FEC.
  3. Unfiltered matched-filter recovery with majority repetition FEC.
  4. Reference cancellation with majority repetition FEC.
  5. Median-period recovery with majority repetition FEC.
  6. A privileged direct measured-syndrome reference.
  7. A privileged exact-clean-PCM reference.

The five channel methods share the same transmitted PCM, disturbance, seed, repetition count, nominal harmonic budget, energy and whole-frame duration. Additional cancellation waveforms and measured-reference symbols are counted explicitly. Clean and direct references are privileged controls. These comparisons do not establish equal-total-resource superiority.

Metrics separate symbol errors, erasures, error-or-erasure rates, natural-binary-label BER, raw accepted-wrong frames, reference verification, decoder attempts and exact-model outcomes. Undefined rates retain 0/0 with defined=false. Missing-length frames are erased, and waveform error is undefined when lengths differ. Direct-syndrome controls mark receiver metrics inapplicable.

The retained four-platform exports from acoustic run 38559170610, at submitted commit 3b72ff0b8fab81171e5c17e52f20d25c0ec2ba60, produced these exact-model successes out of 38 selected trials:

Method Exact-model successes Raw accepted-wrong frames before reference veto
Unfiltered 6/38 4
Cancellation only 14/38 4
FEC only 10/38 2
Combined cancellation and FEC 18/38 2
Median-period receiver with FEC 10/38 2
Direct measured-syndrome reference 38/38 Not applicable
Exact-clean-PCM reference 38/38 0

The canonical battery SHA-256 is aa00b4699f225b9f1325d1372c3e69239aab66b33f2cab0ad102afb38a6406b7. All four final-head platform jobs in run 38574574661 report that same battery identity and 38 trials. The earlier retained battery summaries supply the counts above; the final-head published identities show that the corrective exporter work preserved this canonical battery. Host-specific manifests are expected to differ. These are finite selected controls using one single-site Pauli fixture per alphabet, not a population estimate, threshold result or physical noise study.

Offline laboratory and file interchange

Offline lab guide

The standalone docs/labs/acoustic-evidence.html viewer reads local evidence with its accompanying JavaScript. It displays imported original, observed and filtered PCM, method summaries, receiver candidates, rejections and reference-gated decisions. Source-only inputs do not cause it to invent unavailable waveforms or results.

Waveform plots use the actual samples on a shared PCM16 scale. Playback is click-only, uses the actual sample rate and applies a fixed 25% output gain. Stop, reset, imports and selection changes stop obsolete playback. There is no microphone capture, autoplay, network use or persistence.

Bounded JSON and WAV parsing, safe text rendering and stale-read protections cover malformed input and asynchronous replacement. The viewer is an inspection and listening surface; it does not authenticate hashes or replace canonical Python replay. Node DOM/audio-stub tests exercise the interface, but rendered-browser and audible QA were not completed and are not claimed.

Exports contain canonical source and result JSON, individual trials, actual PCM16 WAV files, six stage artifacts, native transport receipts, file hashes and environment observations. Export size is capped at 32 MiB. Runtime and peak Python allocation observations are host-dependent and excluded from canonical scientific identities; they are neither process RSS nor physical latency measurements.

SPECTRAL interoperability is limited to ordinary local WAV/JSON interchange. No external SPECTRAL UI, repository, project format or hardware integration is included.

Reproducibility and compatibility fixes

Separate SciPy repair · Reproducibility contract

The release includes the separately reviewed legacy nonbacktracking eigensolver repair. Sparse eigensolver calls use a fresh local Generator(PCG64(0)) when SciPy's public API supports rng, while older versions omit the unsupported keyword. The strict scientific threshold classification, dense fallback, fixtures and artifact schemas remain unchanged. This is a targeted restart-RNG repair, with no universal cross-version floating-point identity claim.

The merged acoustic exporter also includes Windows path-versus-descriptor ctime compatibility handling. Cross-API file identity comparisons omit only that incompatible Windows timestamp; same-API mutation checks remain strict. The Unix-socket test uses a short bound path for platform portability.

Export provenance and filesystem corrections

PR #607 completes the exporter corrections that were outstanding after #606. The final merged contract documents:

  • Source identity before generation. Commit, working-tree state and tracked code-file identities are captured before either fixture builder runs, then checked again during and after generation. Loaded Python function code, including the exporter functions, is compared with independently compiled tracked source before fixture generation, catching source changes that occurred after import.
  • Bounded environment metadata. Environment fields are validated before expensive fixture generation and checked again before export, so successful generation stays within the validator's metadata contract.
  • Exclusive output ownership. The writer claims the empty destination before writing evidence. A second generator cannot share that claim, and artifact files use exclusive creation.
  • Handle-anchored generation. POSIX descendant creation is anchored to retained no-follow directory descriptors; Windows creation is relative to retained native directory handles. Manifest digests come from the bytes written through those handles, followed by a guarded inventory of the completed output.
  • Identity-guarded replay. Directory guards remain open while bounded snapshots are read. POSIX retains each enumerated regular file's device, inode, size and timestamps and checks subsequent access against that identity. Windows uses enumerated native file IDs with full volume and 128-bit identity verification. The root manifest receives the same guarded treatment as payload files.
  • Explicit failure behavior. Special files, symlinks/reparse points, unlisted entries, detected replacement or mutation, unsupported filesystem primitives and unavailable required native capabilities fail closed. Windows SMB and unsupported metadata/identity operations do not receive an unsafe pathname fallback.

The published regression coverage includes builder-time source mutation, post-import source drift, metadata bounds, directory substitution, competing writers, enumerated-file replacement, native directory records and platform-specific identity behavior. These are fixes completed on the merged branch and documented in its review and CI; preparing these notes did not rerun those tests or perform a new filesystem audit.

The contract is identity-guarded snapshotting, not a globally atomic filesystem transaction. A sufficiently privileged concurrent process can rename opened directories or mutate an export after return. Use a private, access-controlled working directory where adversarial same-user processes are in scope, and retain source and manifest pins independently. Hashes and filesystem guards do not authenticate physical measurement provenance.

Validation and review evidence

The following are published GitHub results for final submitted head 4be798b785ea687baf3fd87f25bb6effdd4dbcb2. GitHub's head-to-merge comparison reports no changed files, connecting that tested source tree to the merged release tree. No new local test execution is claimed here.

Final-head gates

Acoustic Classical Channel Evidence, run 38574574661 completed successfully:

Gate Published result
Acoustic exact-head full repository regression 20,781 passed, 18 skipped, 4 subtests passed; 573.65 seconds
Ubuntu, Python 3.10 acoustic contracts 572 passed, 1 skipped; export/replay and artifact upload passed
Ubuntu, Python 3.12 acoustic contracts 572 passed, 1 skipped; export/replay and artifact upload passed
macOS 14, Python 3.12 acoustic contracts 572 passed, 1 skipped; export/replay and artifact upload passed
Windows, Python 3.12 acoustic contracts 556 passed, 17 skipped; native export/replay and artifact upload passed

These are distinct job counts, not totals to be added together. Platform-specific skip counts remain visible. Windows evidence includes execution of the real native export/replay path, beyond mocked platform tests.

All four final-head platform logs report the same canonical clean-channel identity, 1398ddd31b4464d55fed2f53ca4c8cd9dcedebbc4fa2e22f16baf097a1e68487, and the same 38-trial battery identity, aa00b4699f225b9f1325d1372c3e69239aab66b33f2cab0ad102afb38a6406b7. Their environment records and whole-export manifests are host-dependent. The final-head log comparison is not a new independent byte-for-byte comparison of every uploaded file.

Proof HTML run 38574575033 passed its repository-local-link and fragment checks. That gate does not establish rendered-browser or audible QA.

The Build and Test Pipeline run 38574574603 also completed successfully. Its full host regression recorded 20,781 passed, 18 skipped and 4 subtests passed in 576.27 seconds. Its separate container gate recorded 20,779 passed, 20 skipped and 4 subtests passed in 494.09 seconds. These environment-specific results are retained separately. The final documentation check at 14:06 UTC found all three final-head workflows returned by GitHub completed successfully.

The final exact-head Codex review, posted at 14:00:08 UTC, reported no major issues for 4be798b785. All six earlier PR #607 review threads were marked resolved. Four late findings received corrective code and regressions; two Windows API/layout allegations were adjudicated as non-defects in the published discussion. Review outcomes are not a universal correctness or security guarantee.

Earlier evidence, retained with its original identity

Before #607, PR #606 supplied head 3b72ff0b8fab81171e5c17e52f20d25c0ec2ba60, merged as 7b39fd8450a29ca39e54b6a97b10448651732ac2. Its acoustic run 38559170610 recorded 20,717 passed, 17 skipped and 4 subtests in 494.26 seconds. These are historical counts, superseded by the final-head results above.

The earlier retained-artifact comparison checked four artifact ZIP digests and found 104 canonical evidence files byte-identical across those platforms. Its Windows source-hash differences were attributable to exact LF-to-CRLF checkout conversion. That historical artifact comparison is evidence for the #606 head; it is not relabeled as a freshly executed #607 artifact audit. The unchanged channel and battery identities in the final-head logs provide the narrower final-head continuity evidence described above.

The earlier telecommunications run 38559170393 and HTML run 38559170750 also succeeded at the #606 head. Their status is not substituted for a different commit's checks.

Preservation and scientific limits

This work is additive and opt-in. It preserves the earlier exact decoder contracts, symbolic runtime, native Digital/Packet schemas and historical fixtures. The acoustic label encoding does not replace the packed-ququart algebra. A passing transport receipt, audible waveform, coherent harmonic bank or self-consistent hash cannot establish a physical protection result.

The v178–v182 research gates remain open: full quantum-state and measurement models; device-specific phononic interaction and control studies; bosonic/hybrid simulations; qualified hardware observation and authorized feedback; and independently supported logical quantum-protection results. No later-stage result is implied by v177.0.

Included changes

  • #601: acoustic-to-quantum research roadmap and claim gates.
  • #602: compatible SciPy eigensolver restart-RNG repair.
  • #603: source-bound acoustic contract and clean PCM alphabet.
  • #604: deterministic interference, reference cancellation, receiver and classical FEC.
  • #605: independently verified PCM-to-syndrome bridge and exact-model correction.
  • #606: native transport replay, comparative evidence, bounded export and offline laboratory.
  • #607: completed export-provenance, metadata-boundary, exclusive-writer and identity-guarded filesystem corrections, with final-head regression coverage.

Getting started

For reproducing the published checks in a configured checkout:

PYTHONPATH=src python -m pytest -q tests/acoustic
PYTHONPATH=src python -m pytest -q

To generate ordinary local evidence, choose a private, access-controlled empty destination:

PYTHONPATH=src python scripts/generate_acoustic_channel_evidence.py --output /tmp/qec-acoustic-channel
PYTHONPATH=src python scripts/generate_acoustic_trial_evidence.py --output /tmp/qec-acoustic-evidence

Open docs/labs/acoustic-evidence.html with its JavaScript file beside it, then import the generated demonstration.json. Retain source, measured-reference and manifest pins independently when validating exchanged evidence. Keep listening volume low, and use the Python validators for canonical acceptance.

Complete historical release archive

The companion archive covers all 953 published GitHub releases returned by the repository inventory on 14 October 2026, spanning the first published demo record, v2 on 22 October 2025, through the combined v176.5 release on 14 October 2026. It preserves every returned release-note body and its original publication identity. The inventory also found 958 tags, including five tag-only references without published GitHub release notes.

The supplied files are:

  • QEC-all-published-release-notes.md: complete original bodies for all 953 published records, with source links and publication metadata. This is the full historical appendix to these cumulative notes.
  • QEC-release-index.md: compact, newest-publication-first index of all 953 titles, dates and release links, followed by the five tag-only references.
  • QEC-all-published-releases.json: structured archival data, including original bodies, body hashes and tag-reference information.

The repository's older 898-entry release-history export matched the corresponding 898 live release bodies; the archive adds the 55 later published records. The preserved Markdown bodies were checked against the returned API strings. This checks archival completeness and fidelity, not the continuing validity of every historical scientific or implementation claim.

Tag-only references are 164.0, v137.5.5, v168.0, v169.0 and v86.1. The index records their verified commit identities and commit dates. No release-note text has been invented for them. Versions are not assumed to form a consecutive numerical series, and contract milestone numbers are not counted as releases unless publication evidence exists.

The historical archive retains source-authored wording, including superseded claims and older status language. For current scope, use the exact-source capabilities and limitations in these cumulative notes. A historical release title or body does not by itself show that an old experimental feature remains active, prove a physical result or complete an unimplemented roadmap assignment. At the inventory snapshot, no v177 release or tag was present; the new v177.0 content here describes merged source prepared for publication.