Repository navigation
QEC v176.5 — Telecommunications Switching and Cross-Era Replay
Telecommunications Switching: ESS, Digital TDM, Packet Fabrics and Cross-Era Replay
This combined update completes the remaining software implementations in QEC's telecommunications switching roadmap: stored-program call processing, digital time-division switching, an offline packet fabric and softswitch boundary, and six-era migration, replay and architecture selection.
The new work builds on the existing Strowger, Panel, Crossbar and ESS skeleton implementations. It also restores executable CI gates and adds reproducible, source-bound candidate evidence.
Included work: #596, #597, #598, #599 and #600, merged on 14 October 2026.
Versioning: v173.1–v176.5 below are roadmap and contract milestones covered by this combined update, not separately published release tags. These changes leave package metadata at 173.0.0; the combined release tag is a separate maintainer choice.
Highlights
- Actual ESS call lifecycles with owned reservations, continuity verification and release.
- Deterministic Digital TDM framing, slot allocation, synchronization, interchange and verified payload reassembly.
- Canonical packet envelopes, exact routing, bounded queues, hop receipts, declared failover and destination reassembly.
- A shared corpus running all six native switching architectures, with explicit capability gaps and architecture-specific faults.
- Immutable, hash-bound inputs and complete replay validation, including trusted identity pins and negative controls.
- Offline Digital and Packet demonstration labs, clearly separated from canonical evidence.
- Full regression and container gates, policy-compatible HTML checks, and clean-source evidence generation.
ESS: source-bound stored-program call processing — v173.1–v173.5
- Added immutable programme, translation-table, feature and source manifests. Programme identity binds the actual installed bytes of the relevant ESS, Crossbar and canonical-serialization modules, rather than relying on a version label alone.
- Implemented the call-processing sequence
intake → analyse → select → connect → verify → release, with explicit rejection transitions and separately identified retry attempts. - Connected ESS to the native Crossbar ownership ledger. Calls perform atomic whole-path software reservations, observe existing contention, verify continuity and release resources using owner/request-bound commands. Later calls see the effects of earlier reservations and releases.
- Added deterministic logical scheduling. Equal-tick events use fixed priorities: release, retry, then arrival, followed by canonical call/branch/attempt tie-breaks.
- Added bounded one-hop forwarding, rejection-triggered retries and ordered fan-out. Fan-out branches remain independent reservations and may produce partial success; they are not an atomic multicast operation.
- Added fully replayed semantic comparison across two supported multistage Crossbar adapters. Logical commands and outcomes can agree while native coordinates, paths and resource identities remain visible.
- Added strict CLI parsing, trusted-input/source pins, exact UTF-8 byte budgeting and bounded reads. Invalid or oversized evidence fails closed; valid native rejections remain explicit receipts.
The published v173.0 ESS skeleton API, schemas and fixtures remain unchanged. The new lifecycle uses an additive qec.ess-*.v2 family with contract version 173.5.
Digital time-division switching — v174.0–v174.5
- Added canonical fixed-width frames, explicit idle symbols, numbered slots and logical frame ticks.
- Added deterministic owned slot allocation and release, including explicit occupied and quarantined initial state. Synthetic initial occupants cannot collide with a caller's valid request identity.
- Added synchronization and continuity checks for frame loss, duplication, reordering and alignment slips.
- Added replayable input-to-output time-slot interchange and payload reconstruction with end-to-end byte-hash verification.
- Added a native fault battery covering bit corruption, slot collisions, frame loss, duplication, reordering and alignment faults.
- Added strict parsers, receipt validators, trusted hash pins, reports and CLI commands.
- Added an offline browser lab with deterministic interaction tests, plus separately labelled SVG/WAV demonstration exports.
The transport model is a bounded single-request transaction with a separate multi-command allocation API. It does not claim to be a concurrent physical switch server. Canonical Digital artifacts use the additive qec.digital-*.v1 schema family.
Canonical packet fabric and offline softswitch boundary — v175.0–v175.6
- Added immutable packet requests and envelopes binding original payload bytes, fragments, sequence identities and caller-declared provenance hashes.
- Added exact binary longest-prefix routing with canonical route tables, stable metric/route-ID tie-breaks and declared next-hop alternatives.
- Added bounded per-node queues, explicit priorities, deterministic logical scheduling and a declared overflow rule.
- Added global and per-packet receipt chains recording admission, queue operations, lookups, attempted alternatives, forwarding, transformations, drops and reassembly.
- Added deterministic failure intervals, alternate-route selection and recovery after declared faults end.
- Added destination reassembly and independent recomputation of the original payload hash.
- Added an offline SIP/RTP/SDP observation boundary. These observations are explicitly external and non-authoritative; they cannot approve native proof claims or activate a live network.
- Added reports, CLI validation and a separate offline lab covering success, failover, overflow, corruption, drops and recovery. All six embedded lab scenarios were checked against native exports.
Canonical Packet artifacts use qec.packet-*.v1 with contract version 175.6. Binary addresses and scheduling rules are QEC software-model choices, not a claim of carrier protocol or hardware fidelity.
Six-era migration, replay and architecture selection — v176.0–v176.5
- Added one canonical compatibility corpus and a migration contract binding destination mappings, topology, policies, initial state, fault plans and complete native inputs.
- Added execution and replay through the actual Strowger, Panel, Crossbar, ESS, Digital and Packet implementations.
- Added a default 51-case corpus producing 306 native executions:
- 36 shared destination/availability cases;
- 15 architecture-specific fault controls that must remain distinguishable.
- Added an independently stated semantic oracle. It checks admission, selected lane, reached destination, exact native rejection reason and failure class. Unexpected verification failures cannot be normalized into ordinary capacity rejection.
- Added pairwise comparison across all 15 architecture pairs, complete native evidence retention, tamper/substitution controls, frozen artifact identities and hash-seed determinism tests.
- Added detached nested artifact snapshots so changing a downstream report cannot mutate its source receipt.
- Added capability-based architecture selection using explicit requirements and preference order, including an explicit “no eligible architecture” result.
- Added an evolution report deriving receipt sizes, event/frame counts, capacity models and fault observations from validated evidence.
The common comparison covers four semantic destinations and two-lane independent initial-state experiments. Migration translates declared inputs; it does not transfer live reservations or assert universal concurrent capacity.
Capability gaps remain explicit: the frozen Strowger API has no native payload transport. Consequently, the implementation does not claim six-way native payload or lifecycle equivalence. Crossbar planning, ESS reservations, Digital frames and Packet queues retain their different meanings. Architecture-specific fault controls are expected differences, not failures hidden by a summary score.
Evidence, CI and repository maintenance
- Replaced workflow dependencies rejected by the repository's existing Actions policy with GitHub-created actions, the runner's installed Docker tooling and repository-owned checks. The security policy was not relaxed or bypassed.
- Made the full repository pytest result an unswallowed regression gate.
- Exercised the container suite on pull requests as well as main, with a non-root matching UID/GID, read-only source checkout and isolated writable HOME/artifact overlays. Historical artifact fixtures are preserved in those overlays; Git credentials and local secrets are excluded from the image.
- Added an offline HTML/link checker for tracked documents, local targets, fragments, duplicate IDs and checkout-root escapes. Generated FER links are explicitly deferred before generation and required afterward.
- Preserved the existing auto-assignment audience and permissions while replacing its disallowed action.
- Added an exact-head telecommunications workflow covering all native eras, historical switching regressions and CI tests. It installs Node for the integrated Digital lab behavior test.
- Added a clean-working-tree evidence gate producing 38 artifacts, their byte sizes and full file hashes, native replay results, source commit and separate environment metadata.
- Updated the roadmap and artifact/acceptance documentation to distinguish published versions, implemented contracts, demonstrations and deferred work.
Principal new evidence includes programme/call/schedule receipts, Digital allocation/synchronization/interchange receipts, Packet hop/failover receipts, the compatibility corpus and migration contract, cross-era replay/equivalence receipts, architecture-selection results and the switching-evolution report.
Validation and provenance
The final reviewed implementation head was 595ce677f8b19a20267e070dcbb29fb633b359a7. Its tree is identical to merged main c514a4ecd2a9bd815558010e1b3f7821f3bce588.
Completed validation on that reviewed tree:
- Full local suite: 20,201 passed, 17 skipped, 4 warnings and 4 passing subtests.
- Independent full-history checkout: the same result, with a clean working tree.
- Hosted switching/evidence gate: 1,098 tests, followed by clean-source generation and validation of all 38 artifacts.
- Hosted container suite: 20,200 passed, 18 skipped; its extra skip is the optional Digital Node test because that image has no Node. The Node-enabled hosted switching job exercises it.
- Exact-head Codex reviews and independent source/native audits completed for all five PRs. Final native PRs #597–#600 had successful applicable premerge workflows.
- Maximum-size cross-era corpus/replay checks passed, including a roughly 17.66 MB evolution report and trusted-hash CLI validation.
Final premerge full/container workflow · Final premerge canonical evidence workflow
Postmerge status is recorded separately below; premerge evidence is not presented as a completed postmerge run.
Compatibility and claim boundaries
- All pre-existing files under
src/andtests/are unchanged by this work. The released decoder and v170–v173.0 switching contracts and historical fixtures are preserved, including the existing compatibility regressions. - New core execution uses declared inputs, topology, programme/policy, initial state, logical event order and fault plans. Canonical model identity does not depend on wall-clock timing.
- Trusted pins bind caller expectations. Without them, replay establishes consistency with embedded inputs, not authenticated external provenance or decoder correctness.
- Demonstrations are not canonical proof artifacts. Reports do not establish hardware performance, physical telephony, quantum advantage, thresholds or universal architecture superiority.
- No live telecom adapter, v193 operating-system work, or deferred older sonification programme is activated.
Known issues and remaining limits
Legacy spectral reproducibility issue remains unresolved
The CI restoration exposed a pre-existing intermittent failure in tests/test_spectral_threshold_search.py::test_artifact_determinism. With the observed SciPy 1.18.1 eigensolver, restart randomness and floating-point values near the bulk boundary can produce differing legacy search artifact bytes despite identical declared search inputs.
The proposed compatibility-gated local-RNG repair is not included in these merges. No scientific boundary rule, historical fixture or legacy numerical output was changed. Later passing runs do not establish that this known intermittent defect is fixed. The original failed FER workflow on the CI prerequisite remains part of the validation record.
Additional postmerge review findings
A Sourcery review posted after merge identified five CI/tooling follow-ups. Read-only triage established these boundaries; no fixes are included in this revision:
- Proof HTML temporarily persists its repository-scoped, read-only checkout token until cleanup. No disclosure was observed; avoiding persisted credentials remains a follow-up.
- Head-only PR tests do not cover a future target-branch divergence. Each of the five actual merge trees in this update was independently checked and matched its reviewed/tested head.
- Leading-slash local links are interpreted from the checkout root rather than the deployed Pages root. No current tracked HTML uses such a local link.
- The standalone container's default test command requires mounted Git history. This prerequisite is documented; the correctly mounted CI suite passes.
- The HTML checker can falsely report a duplicate when one legacy anchor has the same
idandnamevalue.
These unresolved follow-ups are separate from the completed Codex reviews and native switching validation.
Demonstration and HTML coverage
- Digital and Packet labs are offline HTML under
docs/labs/, linked from their contracts. The existing Pages workflow uploadsviz/; these new labs are not automatically published there. - Native, static and Node-based checks were performed. Rendered-browser, acoustic and real-keyboard accessibility QA remain unperformed.
- The Packet DOM harness was run manually; it is not wired into the hosted pytest gate in this merged revision.
- The repository-owned HTML checker does not claim full HTML-standard conformance, rendered-browser behavior or external HTTP-link liveness.
Included pull requests
- #596 — Restore policy-compatible CI and enforce full regression gates
- #597 — Complete source-bound ESS call processing and fabric equivalence
- #598 — Complete deterministic Digital TDM switching and evidence lab
- #599 — Complete canonical Packet fabric, failover and softswitch boundary
- #600 — Complete bounded six-era migration, replay and selection
Postmerge verification snapshot
As of 14 October 2026, 08:22 UTC, main is c514a4ecd2a9bd815558010e1b3f7821f3bce588 and matches the reviewed tree. Final-main HTML and switching-evidence checks passed; full/container and Strowger push workflows were still running. This snapshot does not claim those postmerge gates are complete.
Final-main Build and Test · Final-main switching evidence · Final-main Strowger regression