NGSPICE

1. Overview

Analog and mixed-signal circuit simulator that runs from human-readable netlists without network dependencies. Results depend on device models, solver settings, and version, so comparisons need stated tolerances.

Category: Engineering & CAD.

2. Version and scope assessed

Version assessed: Pinned source 032b1c32c4dbad45ff132bcfac1dbecadbd8abb0.

4. Principle profile

5. Fundamental requirements

6. Evidence and verification

• E1: ngspice reads text circuit decks and documents saving simulation vectors to ASCII or binary raw files. (source), confidence high
• E2: Text netlists, raw files, and printed outputs provide practical migration and independent analysis paths. (source), confidence medium
• E3: The pinned raw writer emits variable names/types, point counts and real/complex flags before ASCII Values or native sizeof(double) binary writes. Independent binary decoding needs matching numeric/layout assumptions; this is a source-level format finding, not a cross-platform reader test. (source), confidence medium
• E4: Simulation runs locally from user-controlled circuit files. (source), confidence medium
• E5: The raw writer selects append modes a/ab versus truncating w/wb from its app argument. Existing output can be overwritten without append; the low-level writer itself does not provide a confirmation dialogue. (source), confidence medium
• sovereignty: A positive Sovereignty finding requires established E1, E2 and E4 in the declared scope. The existing access/exit/custody record for NGSPICE does not establish all three minimums. The prior positive finding is withdrawn; recorded storage and portability strengths remain facts, not a verified failure. (source), confidence medium
• longevity: E3 is not established for NGSPICE's existing essential-record scope. The previous Longevity finding is replaced by Not assessed, not a verified failure or a positive placeholder. (source), confidence medium
• locality: The solver operates locally and offline. (source), confidence medium
• transparency: The positive control observations above do not yet establish whole-scope E5. The methodology requires that minimum before a positive Transparency and Control finding; the previous positive finding is therefore withdrawn to Not assessed, not changed to Does not meet. (source), confidence medium
• E3: The raw writer documents variable metadata and ASCII/native-double payload layout, but no real output was independently decoded and checked for scales/units, complex values and platform layout. Complete netlists/models and numerical reproduction remain unestablished for the original scientific profile. (source), confidence medium
• E5: Append versus truncating file modes are positively established, and the low-level writer has no confirmation dialog. This is not proof of absent higher-level authorization or a failure by itself; effective caller/output selection, script/model authority and correction remain insufficiently covered. (source), confidence medium
• E2: The former E2 Pass does not establish an independent simulation/analysis workflow preserving model dependencies, scales/units and numeric semantics. Available netlist/raw formats are a migration lead, not a demonstrated full-scope transition. (source), confidence medium

7. Migration and recovery

Native storage format: Plain-Text SPICE Netlists (.cir / .sp).

Synchronization or runtime model: Standalone POSIX CLI.

Example export or inspection command (not verified against any particular version or data set): ngspice -b -r rawfile.raw circuit.cir

8. Architectural strengths

Established strengths and their scope are recorded in the evidence and assessment findings on this page.

9. Architectural limitations

• Pinned source inspection, not application execution.
• Independent decoding of real outputs, wrdata unit/scale comparison, external model closure and version-specific numerical behavior remain untested.

10. Assessment findings and remaining work

Verification scenario: Replay netlists with device models, solver settings, and sourced libraries; compare within tolerances.

What may invalidate an overconfident claim: Describing a simulator as deterministic is not a substitute for documented numerical assumptions.

Documentation starting point: https://ngspice.sourceforge.io/docs.html

Evidence update: 23 October 2026

• Resolved: The pinned raw writer emits variable names/types, point counts and real/complex flags before ASCII Values or native sizeof(double) binary writes. Independent binary decoding needs matching numeric/layout assumptions; this is a source-level format finding, not a cross-platform reader test. Source.
• Resolved: The raw writer selects append modes a/ab versus truncating w/wb from its app argument. Existing output can be overwritten without append; the low-level writer itself does not provide a confirmation dialogue. Source.
• Open: Independent decoding of real outputs, wrdata unit/scale comparison, external model closure and version-specific numerical behavior remain untested. E5 needs caller-level output/overwrite and model/control-script policy, inspectability and recoverable source/output states.

Assessment follow-up: These bounded findings are incorporated into the evidence above. Eligibility and principle ratings are deliberately unchanged; their reconciliation and any blank-status fixes are deferred to the next pass. Reproduction details and retained inputs are in evidence/third-pass/batch-8/.

Human-readable criteria: Can an independent party preserve the source model, units, dependencies, and editable structure, reproduce the result, and exchange it without losing important design meaning?

Rating reconciliation: 24 October 2026

E3 and E5 remain Not Verified for the existing declared scope. Longevity and Transparency and Control are Not assessed pending their required minimums. Sovereignty is also Not assessed because the required access/exit/custody minimums are not all established. The prior E2 Pass is withdrawn; its essential-record migration comparison remains incomplete. This is a targeted evidence decision, not a finding that the product fails. Existing E1, E4 and Locality findings were not re-certified.

Remaining: Independent decoding of real outputs, wrdata unit/scale comparison, external model closure and version-specific numerical behavior remain untested. E5 needs caller-level output/overwrite and model/control-script policy, inspectability and recoverable source/output states.

Next without installing locally: Inspect a synthetic circuit/model/real-output packet with variable/scale/unit expectations and platform metadata; compare an independent decoder and externally recorded numerical results, then review caller-level write/script safeguards.

The individual decision and provenance are retained in evidence/rating-reconciliation/batch-8/. Overall: Insufficient evidence.

Individually reconciled in rating batch 8 on 24 October 2026. E3/E5 remain Not Verified for explicit missing coverage; dependent unsupported findings corrected. Previously resolved questions are retained, not counted again. E1/E4/Locality were not re-certified by this targeted review. The assessment procedure is described in the methodology.