Repository navigation
Releases
geotechCLI v0.4.146
Engine correctness, phase 2 — pile capacity, seepage, USCS classification
An audit of the six engines that had no published-value coverage at all found three more defects. As with 0.4.142, every one was invisible to the existing suites because those assert direction and ordering rather than absolute values, and all three erred on the unconservative side.
pile— a measured undrained shear strength was being discarded. The method is auto-selected from the dominant soil, so a mostly-sand profile resolved tospt-meyerhofand then readN/2for every clay layer, ignoring the laboratory strength completely. Su could be varied from 25 to 200 kPa — an eightfold change in the governing parameter — without moving the answer by a single kilonewton. A layer with Su = 200 kPa was contributing fs = 5 kPa where the α-method gives 80 kPa, a factor of 16. For soft clay the effect was unconservative: a soft-clay profile returned 1563 kN where the measured strength supports 916 kN, 71% too high. A measured Su now governs its own layer, and the pile base does the same. An explicit--methodis still honoured exactly as given; only the automatic choice defers to the better datum.seepage— the reported exit gradient was the average gradient.exitGradientwas returning H/L. The gradient at the downstream face is always the larger of the two, which is precisely why piping initiates there, so the factor of safety against piping was systematically optimistic in a life-safety check. The engine already solves Dupuit-Forchheimer, so the exit gradient is now the one that solution actually produces — differentiating h(x)² = h₁² − (h₁² − h₂²)x/L and evaluating at the downstream face gives i_exit = (h₁² − h₂²) / (2·h₂·L). For a 10 m / 2 m head across 8 m the exit gradient goes from 1.00 to 3.00 and the piping factor of safety from 0.97 to 0.32 — the previous figure was three times optimistic. With zero tailwater the Dupuit exit gradient is unbounded; that is now reported as a critical exit condition instead of a finite number that reads as safe.classify— organic soils were unreachable. The result type declared anorganicgroup but no code path could ever return it and there was no organic input, so peat and organic clays were being reported as MH or CH — the most compressible and weakest soils classified as ordinary ones. ASTM D2487 separates them on the ratio of liquid limit after oven drying to liquid limit not dried, so--ll-rationow classifies OL below LL 50 and OH at or above it when the ratio is under 0.75, and--organicclassifies Pt at 75% organic content or more (ASTM D4427). Inorganic soils are unaffected.
Provenance corrected rather than implied. Two citations in pile did not match the code. The base factor was labelled Berezantsev (1961) but evaluates Prandtl-Reissner, giving 18.4 / 33.3 / 64.2 at φ = 30 / 35 / 40 where Berezantsev's deep-pile factor is substantially larger and also depends on L/D — so base resistance is under-predicted for deep piles. The α correlation was labelled API RP 2GEO but is the Tomlinson-style total-stress form on Su alone, where API normalises on ψ = Su/σ′v0. Both are now documented for what they are, along with the engine's other standing limitations: working stress only, compression only, no group effects, and a fixed 0.3 m H-section flange assumption. Restoring the true Berezantsev and API formulations needs the source tables.
Pinned with 17 further golden tests, taking engine-golden-values.test.ts to 61. Eight of the new tests fail against the previous code — verified before the fixes were written.
Install
npm install -g geotechcliPublished with build provenance. Full history is in CHANGELOG.md.
Full Changelog: v0.4.145...v0.4.146
geotechCLI v0.4.145
Dependency security — 13 high-severity advisories down to 1
npm audit reported 20 vulnerabilities (13 high) and CI ran the check with || true, so nothing ever failed a build on it and the backlog grew unnoticed. Total is now 8 (1 high).
- Next.js: 8 advisories fixed by moving 15.5.18 → 15.5.22. These were the ones that mattered most, because Next is the runtime of the deployed Worker rather than build-only tooling: SSRF in Server Actions on custom servers, SSRF in rewrites via attacker-controlled destinations, cache confusion of response bodies, unauthenticated disclosure of internal Server Function endpoints, unbounded Server Action payloads on the Edge runtime, and DoS in both Server Actions and the image optimization API.
- postcss → 8.5.23, clearing an arbitrary file read, a source-map path traversal and a stringify XSS, and with it
@tailwindcss/postcss. Next pins postcss exactly, so this is carried by the rootoverridesblock that already existed for that purpose. - sharp → 0.35.3, clearing inherited libvips CVEs. This one ships to npm users as a declared dependency of
@geotechcli/core. - wrangler → 4.114.0, clearing
miniflare,undiciandws. - js-yaml → 4.3.0, tmp → 0.2.7 (reaches the published CLI through exceljs), form-data → 4.0.6, vite → 8.1.5.
brace-expansionis deliberately left at 5.0.5. Version 5.0.8 fixes the advisory but removes the package's default export, andminimatchinside the OpenNext build chain doesimport expand from 'brace-expansion'— forcing the upgrade breaksbuild:cfoutright, which was confirmed by trying it. The nested copies in the 1.x and 2.x lines are already at their patched releases (1.1.16 and 2.1.2); the advisory's<=5.0.7range sweeps those older majors regardless.
The lockfile was regenerated to apply these cleanly, and two releases were burned learning the same lesson: npm overrides only rewrite transitive resolutions. When a workspace also declares the package directly, the override wins in the lockfile while the manifest still demands its own version, the two disagree, and npm ci fails at install with Missing: <pkg>@<version> from lock file — on the runner but not locally, because npm versions differ in how strictly they validate this. It happened first with sharp (which additionally leaves stale @img/sharp-* platform entries behind) and again with postcss.
Both are now declared directly where they belong — sharp as ^0.35.3 in @geotechcli/core, postcss as 8.5.23 in @geotechcli/web — and verify:consistency gained an assertion that fails the build when an override disagrees with a direct declaration in any workspace, so this cannot reach CI again.
Verified with the full pipeline against the regenerated tree: consistency, all builds, the OpenNext Cloudflare bundle, web typecheck, web route smoke, FEM draft-run, the agent-task benchmark, and 874 + 152 tests.
CLI banner points at the production site
The startup banner advertised beta.geotechcli.com. It now shows www.geotechcli.com. The hosted-beta proxy endpoint is unchanged and still resolves to beta.geotechcli.com/api/proxy, which is where the service actually runs.
Install
npm install -g geotechcliPublished with build provenance. Full history is in CHANGELOG.md.
Full Changelog: v0.4.142...v0.4.145
geotechCLI v0.4.141
Surface verification waits for propagation
The surface gate added in 0.4.140 failed on its first real run, and it was wrong to fail: npm had beta at 0.4.140 and the deploy had succeeded, but beta.geotechcli.com still answered 0.4.139 for a few seconds afterwards. Cloudflare serves the site from many edge locations and npm serves dist-tags through a CDN, so both keep reporting the previous version briefly after a successful deploy — the deploy job's own smoke check already polls for exactly this reason, and the new gate read each surface once.
verify-release-surfacesnow polls each remote surface until it agrees or the attempts run out (20 attempts, 15s apart by default, tunable withSURFACE_CHECK_ATTEMPTS/SURFACE_CHECK_DELAY_MS), matching the retry behaviour of every other remote check in the pipeline.- Surfaces that already agree still pass immediately, so the happy path adds no waiting.
- A surface that never catches up still fails the run, and the error now reports how many attempts it took before giving up.
Install
npm install -g geotechcliPublished with build provenance. Full history is in CHANGELOG.md.
Full Changelog: v0.4.140...v0.4.141
geotechCLI v0.4.139
Tagging had fallen behind: the last GitHub Release was v0.4.124 (5 June), while 14 versions shipped to npm and to the site in the meantime. This release catches the tag up to the published state. Everything below is already live.
Highlights since v0.4.124
Open source. geotechCLI is now Apache-2.0 with the full source public (0.4.133).
Agentic surface. A queryable GroundModel the agent can interrogate directly (0.4.129), chat auto-context, force-agent and a wider data window (0.4.130), and an Agent Task Evaluation Benchmark that scores the agentic harness deterministically instead of by impression (0.4.131).
Interchange and data quality. DIGGS and AGSi export adapters (0.4.127), lithology normalization (0.4.126), and borehole continuity plus coordinate validation on ingest (0.4.125).
Models. GLM-5.2 is the default text model (0.4.128).
Docs and Changelog rebuilt around the information-architecture patterns from Sphinx and the Furo theme (0.4.137). The docs were 17 sections on a single scroll behind a flat anchor grid; the changelog rendered 127 releases fully expanded with no way to search. Both now have a persistent grouped sidebar, full-text filtering, and progressive disclosure.
Sponsorship settled on Patreon. GitHub Sponsors was trialled and reverted (0.4.134 → 0.4.135); Patreon is the only sponsorship flow.
Citation metadata. CITATION.cff carries the author ORCID, and its version is now asserted against the shared release metadata so it cannot silently drift (0.4.138).
npm publishing fix. Trusted publishing (OIDC) resolves per package and fails silently, which left one package unpublished behind a bare 404. The publish job now supplies an automation token as well, requests provenance explicitly so the token path keeps its attestation, and fails fast with a named cause when neither credential is usable (0.4.139).
Versions in this range
| Version | |
|---|---|
| 0.4.139 | Fixed npm publishing for the CLI package |
| 0.4.138 | Author ORCID on the citation metadata |
| 0.4.137 | Docs and Changelog rebuilt around Sphinx/Furo navigation patterns |
| 0.4.135 | Patreon is the only sponsorship flow |
| 0.4.134 | Sponsorship rework (superseded by 0.4.135) |
| 0.4.133 | geotechCLI is now open source (Apache-2.0) |
| 0.4.132 | CI secret-scan false-positive hotfix |
| 0.4.131 | Agent Task Evaluation Benchmark |
| 0.4.130 | Agentic surface sharpening |
| 0.4.129 | Queryable GroundModel for the agent |
| 0.4.128 | Default text model GLM-5.2 |
| 0.4.127 | DIGGS / AGSi export adapters |
| 0.4.126 | Lithology normalization |
| 0.4.125 | Borehole continuity and coordinate validation on ingest |
0.4.136 was version-tagged for the docs work but never published to any channel; 0.4.137 is the released version of it.
Install
npm install -g geotechcliPublished with build provenance. Full per-release detail is in CHANGELOG.md.
Full Changelog: v0.4.124...v0.4.139
geotechCLI v0.4.124
Full Changelog: v0.4.123...v0.4.124
geotechCLI v0.4.123
Full Changelog: v0.4.122...v0.4.123
geotechCLI v0.4.122
Full Changelog: v0.4.121...v0.4.122
geotechCLI v0.4.121
Full Changelog: v0.4.120...v0.4.121
geotechCLI v0.4.120
Full Changelog: v0.4.119...v0.4.120
geotechCLI v0.4.119
Full Changelog: v0.4.118...v0.4.119