On this page
Release maintenance view
> Refreshed for each stable release. Automation of this page is future work.
This table is a maintenance snapshot for the current stable release (1.9.9), not a scorecard. Every count is reproducible from the cited files:
- Stable capability families — stable families tracked in
capabilities.json. The capability inventory deliberately enumerates only the curated families that carry qualification evidence; domains it does not enumerate are markednot measuredrather than estimated. - Units — public units in the frozen
public-api-1.9.jsonsnapshot, matching theprovenance-audit-1.9.9.jsonunit list. - Public API rows — declarations in the frozen 2,880-row 1.9 snapshot, summed per domain. The count includes every snapshot classification (recommended, advanced, compatibility, experimental, and implementation support), so it is an upper bound on the application-facing surface, not the curated common path.
- Runnable examples — examples in
examples/README.mdwhose domain column names the domain; the cross-domain migration example (23_api_migration_preview.pas) is counted separately. - Qualification status — provenance audit coverage in 1.9.9. Deeper numerical, performance, and workflow evidence is in the release-specific reports listed below.
- Known limitations — the guide or capability-inventory boundary for each domain, not an exhaustive list.
| Domain | Stable capability families | Units | Public API rows | Runnable examples | Qualification status | Known limitations |
|---|---|---|---|---|---|---|
| AlgebraLib | 12 | 13 | 1064 | 6 | Provenance-audited (13 units) | Dense row-major, no broadcasting; LU Solve is square-only; no nonsymmetric/generalised/polynomial eigensystems; sparse products may create fill; partial eigensystems are largest-magnitude only |
| CombinatoricsLib | not measured | 1 | 43 | 1 | Provenance-audited (1 unit) | Fixed-width integers overflow at documented maxima (Factorial > 20, CatalanNumber > 30, BellNumber > 18); K <= N required; not arbitrary precision |
| EngineeringLib | 1 | 8 | 421 | 6 | Provenance-audited (8 units) | Temperature conversions are affine; time units are fixed-duration conventions; Haar is power-of-two; equiripple/Chebyshev/elliptic IIR design and wavelets beyond Haar are outside 1.8 |
| FinanceLib | not measured | 3 | 71 | 2 | Provenance-audited (3 units) | Rates are decimals; discrete period compounding except Black-Scholes (continuous); IRR needs a positive initial investment and one positive future flow |
| GeometryLib | not measured | 1 | 132 | 1 | Provenance-audited (1 unit) | Normalise raises on zero/non-finite vectors; arithmetic operators propagate NaN/Infinity; degenerate constructions raise |
| InterchangeLib | 1 | 1 | 14 | 1 | Provenance-audited (1 unit) | Only the documented spline/FIR/standardiser/scalar-Kalman adapters persist; decomposition factors and general object graphs are not stable formats |
| MathBase | 5 | 9 | 239 | 7 | Provenance-audited (9 units) | StudentT covers X >= 0 only; triangle/circle helpers do not validate degenerate input; budgets are corpus-tested, not worst-case proofs |
| MLLib | 1 | 2 | 129 | 3 | Provenance-audited (2 units) | Dense, serial, in-memory baselines; no boosting, distributed training, or multiclass LDA; forest importance is not causal |
| NumericsLib | 3 | 4 | 272 | 4 | Provenance-audited (4 units) | Explicit non-stiff ODE path only (stiff systems are a post-2.0 gate); LagrangeInterp ill-conditioned beyond ~10 knots; GaussLegendre5 uses five evaluations |
| OptimizationLib | 2 | 2 | 126 | 3 | Provenance-audited (2 units) | Local solvers are local only; SimplexLP is standard-form min c'x with Ax <= b, x >= 0; no MILP or interior-point LP |
| ProbabilityLib | not measured | 1 | 76 | 2 | Provenance-audited (1 unit) | PDF/PMF return 0 and CDF clamps out-of-domain X; parameter order is (X, params...); StudentTTwoTail for two-sided tests |
| StatsLib | 2 | 3 | 177 | 6 | Provenance-audited (3 units) | Sample Variance (n−1) versus population StdDev (n); tests are two-sided; covariance-style standard errors need full rank; no survival/factor/robust-covariance families |
| TimeSeriesLib | 1 | 2 | 116 | 4 | Provenance-audited (2 units) | ARIMAForecast needs the original series to undifference; Kalman filters are linear-Gaussian baselines; Undifference needs D initial values |
Qualification evidence for this release
Qualification and evidence are published per release; the 1.9.9 set includes the qualification report, workflow qualification, numerical evidence, performance evidence, portability evidence, and the provenance audit. The capability inventory is the authority for maturity and limitations.
What this page does not measure
Line counts and test-file counts are not the principal health measure and are not used to rank domains. As a secondary context note, the current src/ contains about 49,400 lines of Pascal across 50 units, dominated by AlgebraLib (~17,200 lines) followed by EngineeringLib (~4,800) and MathBase (~4,600). Size reflects history and algorithm density, not health.