# Mermaid Semiconductor Source Evidence

Source discovery dated September 30, 2026. This discovery snapshot supports the eleven implementation batches approved on September 30, 2026. The semiconductor is named mermaid and the interface destination is mermaid.tide.casa. No framework build or tests were rerun in this phase. Statements about source-present implementations and stored fixture results retain that limitation.

## Source locations and instructions

| Source | Location and status |
| --- | --- |
| Primary Bangel framework | `<user-profile>/Documents/Codex/2026-09-24/ba/work/bangel-language`, clean at `229548f13a4856455b3d30d408cf845fb5b557fc`. |
| Studio framework copy | `<user-profile>/Documents/Codex/2026-09-24/ba/work/studio-bangel-source`, same clean revision. |
| Structured journal | `<user-profile>/Documents/Codex/2026-09-30/let/work/journal-records.json`: 8,914 records with recovered bodies, guides, metadata, source links, body digests, and chapter connections. |
| Structured book | `<user-profile>/Documents/Codex/2026-09-30/let/work/book-blocks.json`: 1,233 blocks, 49 chapters. |
| Foundation and courses | `<user-profile>/Documents/Codex/2026-09-30/let/outputs/MJ Assistant/MJ Foundation Guide.txt` and `Knowledge`. Canonical course source: `<user-profile>/Downloads/BarTide-Independent-Hosting/subdomains/learn.tide.casa/source`. |
| Encyclopedia register | `<user-profile>/Documents/Codex/2026-09-30/let/outputs/Quantum Programming with Bangel Encyclopedia/Paper and Chapter Location Register.csv`. Page numbers below refer to the complete encyclopedia, not an original colored source PDF. |
| Live project catalog | Read-only HTTP 200 retrieval of [the public catalog](https://bangel.tide.casa/library/data/product-catalog.json) and brand metadata on September 30. The selected thermoelectric project confirms the source mapping and simulation boundary. |

Applicable ancestor instructions were checked. No AGENTS.md was found in the relevant framework or completed-book paths. The hosting collection’s AGENTS.md requires respecting independent application identities and preserving source junctions; it was read and no application was edited. The public blue.tide.casa source/hosting identity was not located, and hostname resolution failed from this environment. That is a discovery gap, not a deployment-state conclusion.

## Complete available software component inventory

The primary checkout contains one `components/` package, `thermoelectric-reference`, and five `plugins/` packages: `mj_bangel_neural`, `mj_neural_net`, `mj_memory_recall`, `mj_predictive`, and `mj_decision`. The general language, managed implementation, paper adapters, SDK/API, and studies are separate repository surfaces.

All relative paths in the table resolve inside the primary checkout. Published immutable references can be constructed from [the pinned repository revision](https://github.com/Mjos23/bangel-language/tree/229548f13a4856455b3d30d408cf845fb5b557fc).

| Component | Scope | Source references |
| --- | --- | --- |
| General Bangel 1.0 | Typed source, finite real/vector math, units, UNKNOWN, bounded controls, independent admission, receipts. Elsa/1 → JP-IR/1 → Joanna/1. No physical actuation or complete resource/lease host. | `language/python/README.md:39,45,52`; `docs/CONTRACT-RECONCILIATION.md:23,25` |
| GIN | Interval/order/staleness/tolerance and admitted clock transforms; known, bounded, unknown, inapplicable, disputed, incomparable temporal states. | `docs/GIN-MODELS.md:25`; `docs/GIN-INTERVALS.md`; `docs/GIN-CLOCK-TRANSFORMS.md` |
| Scheduler and host tools | Deterministic bounded dependency scheduling, replay, SDK/API. Wall-clock timeout and isolation are host duties. Logical counts are not joules. | `language/python/docs/B04-SCHEDULER.md:45,92`; `language/python/src/bangel` |
| Original neural baseline 0.1.0 | Associative recall, three nine-activation layers, bounded sweeps, relationship filtering, first-order nine-command prediction. | `plugins/mj_bangel_neural/README.md:3`; `plugins/mj_bangel_neural/source/neural.bangel` |
| MJ Neural Net 0.1.0 | Preserved baseline plus five reference diagram overlays, bounded graph diffusion, optional spatial/composition layer. Biology and broad learning superiority are unvalidated. | `plugins/mj_neural_net/README.md:3`; package Bangel source and Python transport |
| Memory Recall 0.1.0 | Bangel recall/context scoring, projection, command prediction, lifecycle/update rules. Python catalog and SQLite I/O; assessment does not install updates. | `plugins/mj_memory_recall/README.md:3`; `plugins/mj_memory_recall/docs/KERNEL.md:144` |
| MJos Predictive 0.6.0.dev1 | Preserved PMOS Bangel math, comparability, dual independently admitted workers, guardian/transport in Python. Advisory, uncalibrated forecasts. | `plugins/mj_predictive/README.md:3`; `plugins/mj_predictive/src/mj_predictive/_pmos/bangel/forecast.bangel`; `docs/MJ-L5-API.md:32` |
| MJ Decision 0.1.0.dev1 | Bangel ranking, cost/uncertainty and gate precedence; host supervision and independent evidence. Advisory selection; execution permission false. | `plugins/mj_decision/README.md:3`; `plugins/mj_decision/src/mj_decision/integration.py:42` |
| Thermoelectric reference 0.2.0rc1 | Separate Bangel 0.1 MJPS declarations and frozen JP-IR/MJPS-1, with Python steady physical kernel and bounded optimizer. Version is distinct from grammar. | `components/thermoelectric-reference/README.md:3,31,35` |
| Paper adapters | 136 candidate computational profile IDs and 1,200 documentary profiles; published preview retains its original 37 scope. Profiles are not processor counts. | `papers/python/README.md:5,25`; `papers/python/src/bangel_papers/data/pink/profiles.json` |
| .NET implementation | Syntax, semantics, compiler, admission, numerical, runtime, SDK and conformance source exists. Fresh platform/parity evidence is pending. | `language/dotnet` |

The 18 courses are educational tracks. Foundation Guide lines 19–24 define framework roles and state that the course release did not establish a complete bootable MJos OS. Its line 48 records historical release checks. Course README lines 52–54 bind current labs to general Bangel classical calculations and the real Elsa → JP → Joanna chain, rather than physical hardware jobs.

## Profile boundaries and quantum extension

`language/python/docs/COMPATIBILITY.md:3` separates strict general 1.0, explicit alpha compatibility, historical Physics 0.1 experiment compatibility, and archive Bangel 0.1. The thermoelectric profile is its own package, not interchangeable with archive compatibility.

Within `language/python/src/bangel`, `compiler.py:64,228` selects the legacy parser for Physics 0.1 compatibility. `parser.py:30–147` implements an experiment envelope and real calculation statements, not qubit/apply/sample gates. `evaluator.py:240` contains real Quantity and tuple/vector operations. No complex-amplitude/Born/multi-qubit/noise simulator was found in this source closure.

The historical reference explicitly marks its migration as documentary (`specs/roles/BANGEL-PHYSICS-01-HISTORICAL-REFERENCE.md:5–7`). Its later gate definitions are requirements/reference material. The book’s example-verification register likewise distinguishes executed general calculations from the documentary Bell listing.

MJ.23.R2.ACT proposes independent MJ/Elsa/Joanna/JP cells with typed CAPO exchanges and leases. Its lines 5–7 and 53–61 mark the historical architecture nonexecuting and unbuilt. It differs from the current compiler/admission/runtime pipeline. Its eight Angel nodes, Sister reserve/memory, Shekina placement, Bellatrix defense, Mermaid implication, Samson monitoring, Lecanto visibility, Moana interaction, and Aviator workflow are proposed roles, not extra installed semiconductor cores. DRO/Pearl have no standalone package in this checkout.

## Semiconductor records and computational gaps

| Record | Encyclopedia page | Evidence boundary |
| --- | --- | --- |
| BP-PROD-037 JP Semiconductor DRO and Pearl Computational Integration | 2423 | `PLANNED_ONLY_NO_PHYSICAL_ARTICLE`. |
| MJPE-COMPLETE-20260925-SEMICONDUCTOR-DRO-PEARL-PINK | 2071 | Separate measurement, simulation, power roles and unresolved hardware evidence. |
| Same project YELLOW | 5223 | SemiconductorProfile, DROEnergyLedger, PearlNode, WakeReserve. Computational retention is not physical energy storage. |
| BP-OBJ-052 SEMICONDUCTOR_DRO_ENERGY_LEDGER | 2380 | Typed charge/energy/wake/restart accounting. |
| BP-OBJ-053 SEMICONDUCTOR_PEARL_NODE | 2381 | Computational memory/evidence limits. |
| MJPE-COMPLETE-20260925-DIGITAL-TWIN-PINK and YELLOW | 2017 and 5173 | Distinguish actual received state, predicted continuation, comparison, and receipt; drawing is not calibrated geometry. |

The registry `papers/python/src/bangel_papers/data/pink/profiles.json` maps F-152 to a scalar thermal step (line 1213), F-230 to a thermal inequality (1751), F-232 to byte-budget addition (1766), and F-233 to a reserve inequality (1783). Stored fixtures expect 302.5, true, 24, and true respectively. Their encyclopedia records are TR-PINK-F-152/230/232/233 on pages 4046/4127/4129/4130. Those fixtures do not establish measured units, thermal calibration, memory allocation, or physical permission.

F-159 workload power, F-168 semiconductor geometry, F-231 sampled charge integral, and F-234 profile isolation are absent from the computational registry. `papers/python/src/bangel_papers/host.py:95–101` raises `PAPER_PROFILE_UNIMPLEMENTED` for unavailable IDs. The documentary records remain on pages 4053/4063/4128/4132. New contracts must address units, times/sampling, uncertainty, frame/bounds, and isolation rather than replacing a missing operation with a caller-supplied Boolean.

## Cooling work and name mapping

The supplied [thermoelectric project](https://bangel.tide.casa/#project=bangel-thermoelectric) maps in the live catalog to **MJPS Thermodynamics**, evidence level `implementation_source`, with no physical controller, board backend, or actuation. Exact “Mermaid effect” was not recovered from the structured corpus. Mermaid Reconciliation and the historical Mermaid implication gate are different documented roles. The plan retains the user’s cooling label provisionally and reuses the recovered thermal work as requested.

Relevant records are `mjpe-complete-20260925-bangel-thermoelectric-pink`, `-teal`, and `-indigo`, on encyclopedia pages 2005, 2974, and 7984. They preserve steady thermoelectric simulation, expected/anticipated/delivered temperatures, calibration states, and later bench/HIL/field evidence gates.

`components/thermoelectric-reference/source/01_hybrid_bus_pass.bangel:7–11` defines two modules: `engine_teg` for Seebeck harvesting and `os_tec` for Peltier cooling/heating. The fixture uses alpha 0.05 V/K, resistance 2 ohm, conductance 0.5 W/K, nominal hot/cold temperatures 320/300 K, temperature uncertainty 0.2 K, 12 W cooling demand, and 12 W external limit. It is an illustrative bounded mission, not an identified chip.

The remaining source paths in this cooling section are relative to `components/thermoelectric-reference`. The Python kernel implements:

```text
deltaT = Th - Tc
V   = alpha*deltaT + I*R
Pin = V*I
Qc  = alpha*Tc*I - I*I*R/2 - K*deltaT
Qh  = alpha*Th*I + I*I*R/2 - K*deltaT
residual = Qh - Qc - Pin
```

Sources: `src/mjps/physics.py:52–80`; `profiles/JOANNA-MJPS-RUNTIME-PROFILE-1.json:78–109`. Four temperature corners bound guaranteed cooling/harvest and worst electrical inputs. The finite optimizer has a 200,000-combination limit; its bus credits guaranteed harvest after DC losses and subtracts worst pump input after driver losses plus auxiliaries (`profile:128–157`).

Stored physics tests expect nominal 16 W cooling, 26 W hot-side rejected heat, 10 W electrical input, COP 1.6; uncertainty lowers guaranteed cooling to 15.78 W and raises worst input to 10.04 W. Stored hybrid runtime tests expect 0.12005 W harvest and approximately 11.291508 W external draw, with zero unmet cooling. These are model/fixture results, not measurements (`tests/test_physics.py:38–71`; `tests/test_runtime.py:39–51`).

The profile excludes temperature-dependent properties, Thomson effects, transient junction/contact-resistance identification, aging, and physical validation. It names a separate `mjps.network.ThermalNetwork` at lines 160–175, but its implementation/tests are absent from the packaged source inspected. The source closure therefore establishes a steady kernel; transient recovery or implementation is a distinct batch task. Results retain `physical_effect = NONE`, `authority_effect = NONE`, and simulation dispositions such as PASS_PROPOSAL/HOLD.

## Book guidance and external engineering references

Relevant chapters: 17–20 language/compiler/admission/runtime; 25 time and reconciliation; 26 experiment/profile selection; 31 reproducible quantum experiments; 36 forecast evaluation; 37 neural logic; 38 recall identity; 39 supervised decisions; 40 digital twins; 46 checked roadmap.

The book supplies independent small-circuit checks, including Bell bit-flip mismatch `2*p*(1-p)` for independent errors. It requires availability cutoffs, held-out forecasts, same-horizon baselines, MAE/bias, coverage denominators, and a distinction between analog spread and calibrated uncertainty. Python transport/persistence and Bangel scoring are explicitly separated. The neurological umbrella record, page 2055, remains `source_recovery_pending`; the related software modules were recovered.

External sources were checked as engineering references, not evidence about a Bangel device:

- [Philips and colleagues, silicon six-qubit processor](https://arxiv.org/abs/2202.09252): supports silicon spin qubits as a candidate semiconductor technology with control, initialization, and readout requirements. It does not transfer their measured results to this proposal.
- [Ferrotec thermoelectric modeling](https://thermal.ferrotec.com/technology/thermoelectric-reference-guide/thermalRef11/): gives cooler input, heat-pumped, heat-rejected, and COP relations and discusses temperature-dependent coefficients. Coefficients need a valid envelope for the chosen device.
- [Fellous-Asiani and colleagues, full-stack resource efficiency](https://arxiv.org/abs/2209.05469): motivates including electronics, wiring, cryogenics, noise, and task performance in energy comparisons. Its illustrative technology/parameters are not Bangel inputs.
- [IBM QPU information](https://quantum.cloud.ibm.com/docs/en/guides/qpu-information): distinguishes throughput and quality metrics and calibration data. Simulator wall time is not a physical QPU performance metric.

Raw focused discovery notes, live-source downloads, and extracts remain under this workspace’s `work/` directory. User-facing deliverables are under `outputs/`. The original book, courses, and framework files were only read.