chore(website): i18n coverage audit and stale-content banners#49
Open
agourakis82 wants to merge 106 commits intomainfrom
Open
chore(website): i18n coverage audit and stale-content banners#49agourakis82 wants to merge 106 commits intomainfrom
agourakis82 wants to merge 106 commits intomainfrom
Conversation
…ed native code (+5 tests) BSS-resident shadow buffer propagates ISO/JCGM 100:2008 variance through all four arithmetic operators, across function calls, through ODE loops, and with correlated self-input detection (GUM §5.2). Zero ABI impact on primary value path. Bootstrap fixed point verified (gen2==gen3). Features: variance seeding from Knowledge.value, additive/multiplicative/ division GUM, variance_of() introspection, cross-function propagation via BSS transfer area, compile-time correlation detection for same-slot operands.
…annotations - Restructured gpu_vec_add.sio to use 2 slices (4 ABI args) instead of 3 (6 ABI args) - Removed @known-failure from kernel tests that now pass: - kernel_fn_basic.sio - kernel_fn_gpu_effect.sio - kernel_multi_backend.sio - kernel_ptx_emit.sio - Updated annotations for tests with different failure modes - Added CI workflow for parallel test execution with JUnit output - Created test runner v2 with parallel execution support Test results: 505 pass (+23), 0 fail (-3), 172 known failures (-16) The kernel ABI limitation affects tests with >6 ABI arguments. By using ≤3 slices (≤6 ABI args), kernel tests pass successfully.
…ffect annotation Type-system enforcement of epistemic confidence thresholds. Functions annotated with Epistemic(minconf: N) reject calls where argument confidence < N at compile time. Confidence tracked through measure(), .value extraction, arithmetic (min propagation), and cross-function calls. Bootstrap fixed point verified (gen2==gen3). Error E180 fires when confidence gate is violated — no runtime check needed.
Make ODE solver functions and types public: - Add 'pub' to solve_rk4_exp_decay, solve_rk45_exp_decay, solve_pk3_rk4 - Add 'pub' to SolverODESolution, ODESolutionVec3, State3, etc. - Add effect annotations (Div, Panic) to public functions - Add convenience aliases: exp_decay_analytical, default_pk3_params Fix ODE tests: - Add missing effect declarations to test functions - Remove @known-failure annotations from passing tests Test results: 533 pass (+28), 0 fail, 144 known failures (-28)
…position (+2 tests) Tagged variance decomposition (8-channel BSS shadow) applied to ODE integration and 3-compartment rapamycin PBPK. Single forward pass produces per-parameter ISO/JCGM 100:2008 §8 budget tables — no Monte Carlo required. Key result: Kp_brain (BBB partition) drives 96-99% of predicted concentration uncertainty. GUM linearization validity limit discovered at ~5h for models with division by small uncertain parameters (b^4 amplification).
- Add PTX code generation for GPU kernels - Add CLI_GPU_ENABLED and TARGET_BACKEND flags - Add FN_IS_ASYNC array for async function tracking - Add Task type hash helpers for async runtime - Update GPU-related files (hlir_to_gpu, kernel_ir, lower_to_ptx) - Update stdlib sync primitives (channel, rwlock, thread pool) - Rebuild self-hosted compiler artifact
- Implements octonion SSM forward pass - SoA memory layout for coalesced GPU access - Dual f32/f64 precision paths - Epistemic shadow propagation variants - ~192 FMAs per element, within register budget
Propagate EXPR_CONFIDENCE through if-else (min of branches), while/for loops (5% degradation per iteration), and match expressions (min across arms) in both x86 and A64 backends. Bootstrap fixed-point verified (gen1==gen2, md5=63bdad96).
Add missing `with Mut, Div` to mc_sqrt, lcg_uniform, normal_sample, run_euler_pbpk. Update souc-self-hosted-x86_64 artifact after bootstrap rebuild (gen1==gen2==gen3, md5=880d3180).
Enable --backend gpu, --gpu-target, --gpu-precision flags in bin/souc. Add GPU dispatch in lean_single.sio (TARGET_BACKEND, write_ptx, emit_ptx_header). Add hypercomplex opcodes and helpers to GPU optimizer. Add integration test suite (8/8 passing). Update AGENTS.md.
…ep control Introduces tests/run-pass/rapamycin_epistemic_adaptive.sio: Bogacki-Shampine 3(2) adaptive RK with dual step sizing — numerical error control combined with metrological variance-growth control, the first such ODE integrator. Per-step mechanism (lookbehind): err_norm = |e_num|/(atol+rtol|y|) [numerical] u_ratio = var_growth/(u_atol+u_rtol*var_pre) [metrological, GUM] dt_next = min(dt*safety*(1/err_norm)^(1/3), dt*safety*(1/u_ratio)^(1/2)) Run on 3-compartment rapamycin PBPK with CL, kp_brain, fu_plasma uncertain at 10% CV. Integration completes 0→2h in 70 steps with correct concentrations (c_blood=0.190, matches Euler reference 0.175 within 3rd-order discretization) and fires 4 epistemic shrink signals — the variance-aware step sizing is active. Uses lookbehind rather than trial-and-reject because trial-commit patterns (let trial; var = trial) have fragile variance-copy semantics in deep expression chains. In-place mutation (var = var + expr) is reliable and gives the same adaptive behavior via post-step, next-step control.
…sociative multiplication
- β-0: GUM Jacobian through octonion multiplication (structure tensor derivation, associator Jacobian as commutator ∂[a,b,c]/∂b = [L(a), R(c)], numerical verification via hand computation) - β-5: Unified type theory draft for POPL/PLDI. Formal grammar, 10 inference rules (Sub-Refine, Conf-App, K-Value, GUM-Compose, Algebra-Rewrite, etc.), soundness sketches, e-graph integration, 7-generation convergence data. - δ: Epistemic gradual compilation paper for existing Sounio work. Self- referential validation story, guard-marker mechanism, rapamycin PBPK case study, 40 citations. - ζ: Compiler variance-tracking bug fix plan. Root cause = NEXT_SLOT not reset at while-loop boundary, causing slot index overflow past 1024-slot buffer and reading uninitialized BSS as 2^63. 3 locations, 2 lines each.
First fix attempt (save/restore NEXT_SLOT at while-loop boundary) broke bootstrap — gen2 compiled but gen3 failed with "error: no main". Reverted cleanly; bootstrap fixed-point restored (md5=24cfaccb). Likely cause: resetting NEXT_SLOT without also resetting VAR_COUNT leaves stale symbol table entries pointing to slots that subsequent code will re-allocate. Document the failure so the next attempt knows not to repeat. Revised recommendation: try buffer-size expansion (1024 → 16384 slots) before retrying the scope-reset approach. Buffer expansion is a one-liner with zero scoping risk; the scope fix requires understanding Sounio's full variable lifecycle.
- commutator_associator_identity.md: rigorous derivation of ∂[a,b,c]/∂b = [R(c), L(a)] as a structural fact. Alternativity, flexibility, and Moufang identities recast as operator commutator vanishing. Cayley-Dickson tower rank census (C/H commute, O rank-4 generic, S zero-divisor locus). Moreno 1998 zero-divisor theorem in operator form. Fano-selective e-graph recast as syntactic approximation of [R(c), L(a)] = 0. - categorical_knowledge_monad.md: Knowledge<T> as a restricted Kleisli arrow via truncation functor of the Giry monad. Parzygnat quantum Markov category extension to JBW-algebras (Albert algebra Herm_3(O) as octonion observable home). Hanche-Olsen 1985 JBW tensor-product obstruction flagged as genuine categorical limit. Sedenion escape: pushforward discontinuous at the zero-divisor variety. - zeta_variance_deep_investigation.md: refined root cause. VAR_ arrays ([i64; 1024]) overflow when NEXT_SLOT exceeds 1024 — not the variance scratch. Frame-size calc at line 14884 uses final NEXT_SLOT, which is why my first-attempt scope reset broke gen2's frames and gen3 lost its main symbol. Recommended fix: expand VAR_ arrays to 8192. Zero bootstrap risk.
First concrete demonstration of the epistemic singularity at a non-associative
algebra's zero-divisor locus. Uses Moreno's canonical sedenion pair
a = e_3 + e_{10}, b = e_6 - e_{15}
with |a|=|b|=√2 but a·b = 0 exactly. Stdlib math::sedenion::sed_mul confirms
|a·b|² = 0 at the algebraic level. With σ=0.01 on each coefficient, the
hand-derived output components give E[r_5] = E[r_12] = 0 with
variance = 4σ² = 4·10⁻⁴, matching first-order GUM prediction exactly.
Signal-to-noise ratio is formally undefined — mean=0, variance>0. This is the
"confidence-collapse event" predicted by the commutator-associator identity:
at the zero-divisor locus ker L(a), the associator's middle-Jacobian
[R(c), L(a)] has a specific rank-drop pattern that makes uncertainty
dominate signal.
- Add 'pub' to impl methods in bitset, deque, heap, hashmap, hashset - Fix effect declarations in oct_minimal.sio (now passing) - Partial fixes for sobol_advanced.sio effects Test: oct_minimal.sio now passes
…name bug workaround) The native compiler misreads the field named "central" (7-char, second field in this specific layout) from PBPKState, returning the value at gut_unc's offset (32) instead of central's offset (8). Renaming to "cen" routes around the hash-collision bug in the field-lookup path. All 7 epistemic_pbpk tests now pass: T1–T7.
- darwin_pbpk: pub on PBPKState14, PBPKParams14, Tsit5StepResult14, EpPrior14, and their constructor/step fns - log/lib: array-of-structs → struct-of-arrays (native codegen compat) - signal/fractal: inline higuchi/dfa helpers to avoid nested BSS aliasing - nn, ode, optimize, special, theorem, interop, io: pub on exported types
Inlines all round-function helpers into sha256_compress to eliminate fn-call aliasing with the K-constants array in the native backend.
Replaces `var fail = 0; fail = fail + 1; return fail` pattern with
`if !cond { return 1 }` to avoid BSS aliasing of the counter variable
in the native backend. Affects heliobiology, interop, medlang, nn,
ode, optimize, pbpk, signal, snn, stats, stochastic, theorem, util.
- build_ontology_validation_souc: run_diagnostics_check() extracts error/warning/item/fn_sig counts from debug-check binary; confidence scoring (1.0 unanimous, 0.7 fallback_override, 0.5 disagreement) - run_ontology_validation: extended verdict + provenance fields - run_sio_test_suite_v2: agent_witness JSON field, SOUNIO_TEST_RESULTS_DIR, SOUNIO_TEST_PRESERVE_JSON flag - prepare_abide_campaign_snapshot: minor cleanup - docs/ABIDE_RUNBOOK: trim stale scaffold sections
G0:26% → G6:90% (gates live) → G9:97% (0 guarded) → Current:100% Current: 88,056 expr, 9,851 direct calls, 0 guarded, 22,475 lines
Year updated from 2025 to 2026. License corrected from MIT to Apache-2.0 to match the repository LICENSE file. Both footer_copyright and footer_license keys updated across all six locale files (en/pt/el/es/zh/ja).
The mobile-menu div was a sibling of <nav> inside <header>, causing all navigation links to exist outside the single navigation landmark. Moved the mobile-menu div inside the <nav> element and wrapped the top bar row in an inner div, so there is exactly one <nav aria-label="Main navigation"> containing both desktop and mobile link sets.
The footer contained two sections both titled "Language": one for the Sounio language navigation links (Value Proposition, Compiler + Runtime, Scientific Domains) and one for the locale switcher (English, Português, etc.). Renamed the locale-switcher heading to "Translations" to disambiguate. The navigation section heading remains "Language".
…and compile-time confidence gate Replace audience-split, metaphor-heavy layout with a single linear page organised around three hard facts: self-hosting fixed point (stage2==stage3), the 168=|PSL(2,7)| theorem verified in Sounio and submitted to AACA, and compile-time rejection of under-confident vancomycin dosing. Adds Pappou+Yiayia diptych with explicit caption. Honest-status section mirrors README WORKS/SCAFFOLDING/MISSING table verbatim. All "pressure surface" metaphor occurrences removed. Archive of prior homepage at _archive/index.v0.astro. Branch forked from website-restructure-01-objective-fixes (PROMPT 1 not yet merged) to include footer and header deduplication fixes.
Adds stdlib/onn/lib.sio providing:
- Octonion struct {w,x,y,z,e4,e5,e6,e7: f32} (e4-e7 avoid unit-system conflicts)
- OctLinearLayer and OctConv2dLayer layer descriptor types
- Cayley-Dickson multiplication (verified norm-multiplicative)
- Component-wise activations: relu, leaky_relu, sigmoid, tanh, gelu
- oct_linear_fwd/bwd for dense layer forward/backward passes
- oct_make constructor (workaround for 8-field SRET import limitation)
Test kept with local struct definition pending native binary rebuild
(8-field struct SRET from imported functions not yet working in current binary).
Add OS-thread async runtime (Phase 1) to lean_single.sio:
- `async { expr }` — soft keyword, eager eval in current thread, identity semantics
- `spawn { expr }` — fork(2) + MAP_SHARED mmap slot; child executes block,
stores i64 result to shared slot, then exit(0); parent gets TaskHandle
- `expr.await` — if EXPR_TY==100 (TaskHandle): wait4(pid) + read result slot;
otherwise identity (for async fn calls that return values directly)
TaskHandle is a 2-slot anonymous struct (EXPR_TY=100 sentinel):
[ptr+0] = pid (fork return value in parent)
[ptr+8] = slot_ptr (mmap'd MAP_SHARED page)
Soft keyword detection via src_match inside the k==3 handler avoids
breaking `fn spawn(...)` method declarations in stdlib/async/executor.sio.
Fixed-point: gen2 == gen3 (md5: 677304075bc399d6d7f7d186044f557e).
Tests: async_basic PASS, async_spawn PASS, async_spawn_syscall_pid PASS.
Suite: 543 pass / 1 fail (gpu_vec_add, known) / 233 skip.
Add channel primitive to complete Phase 1 async runtime: - `channel::<T>()` — pipe2(2) into a stack-resident int[2] buffer; sign-extends read_fd/write_fd to i64; returns a 2-slot tuple pointer (EXPR_TY=6, EXPR_TY_HASH=1002) compatible with `let (tx, rx) = ...` destructuring: [ptr+0] = write_fd (tx), [ptr+8] = read_fd (rx) - `.send(v)` — write(fd, &v, 8); intercepts on EXPR_TY==1 by method name - `.recv()` — read(fd, &buf, 8); returns 8-byte value from buf slot All 4 async tests now pass: async_basic, async_spawn, async_spawn_syscall_pid, async_channels Fixed-point: gen2 == gen3 (md5: 6ed4e05ab219e611ae68f54791f75095). Suite: 544 pass / 1 fail (gpu_vec_add, known) / 232 skip.
Compiler: - Allow `expr as Unit2` when Unit1 and Unit2 share the same dimension (e.g., mg→kg, s→ms); cross-dimension casts (mg→s) still rejected. Fix applied to both x86-64 and ARM64 postfix cast handlers. Tests unblocked: - unit_cast_compatible, unit_cast_time: same-dimension cast now passes - turbofish, pattern_match_basic, covid_2020_kernel: promoted check-only→run-pass - stdlib/ode/test_rk4_fn_ref_e2e: removed stale segfault-era ignore Fixed-point: gen2==gen3 (md5: 8404ba5d9c3c4fab406b397e64f088ff). Suite: 547 pass / 1 fail (gpu only) / 229 skip.
- Fix <TPARAMS> strip in mono_pass token copy: replace fixed +3 skip with depth-tracking loop that correctly handles <T>, <T,U>, and nested angle brackets like <Knowledge<T>> - Add GEN_FN_TP2_S/E arrays to record second type parameter from fn<T,U> - Add MONO_TY2_S/E arrays to store second instantiation type per mono entry - Extend Pass 0c to detect and record second type param (TK==28 comma check) - Extend Pass 0d turbofish branch to extract first/second types separately; fix MONO_TY_S/E to cover only first type (was spanning entire arg list) - Fix compile_primary turbofish lookup: use TE[(EP+2)] not TE[(tf_p-1)] so mono_find_inst correctly matches single-type key in all cases - Add second-param substitution branch in copy loop - Add tests/run-pass/generics_multi_param.sio: first<T,U>/second<T,U> - Update KNOWN_LIMITATIONS.md: monomorphization is working, not pending - Rebuild bin/souc-native (336/338 run-pass, 0 new failures)
…commands Five sections with anchor IDs matching homepage deep-links: #theorem-168, #fixed-point, #compile-fail-vancomycin, #test-gate, #lean-proofs. Compiler output in §3 captured live (2026-04-20): error: Type mismatch — expected confidence level not satisfied: Knowledge ε boundary violation at line 27 typecheck: failed Test gate numbers in §4 captured live (548 pass / 1 fail / 229 skip); 1 failure is gpu_vec_add.sio (no CUDA on CI runner, not a hidden failure).
Features implemented and working in native binary: - async channels (.send()/.recv() via fork+pipe) - borrow checker at call boundaries (no false positives) - MCMC/stochastic-calc borrow paths - effect handler discharge syntax - affine struct drop semantics - vancomycin epistemic propagation (runs, not just checks) All 7 tests pass at exit 0; annotations were left over from pre-native-rebuild state. Suite: 336/338 run-pass.
- New /language page: four sections (Epistemic types · Effects & linearity · Compiler · Runtime) with sticky left TOC (active-section highlighting) - /platform → 308 redirect to /language in astro.config.mjs (all 6 locales) - platform.astro and [lang]/platform.astro replaced with client-side redirect stubs that add the #compiler fragment (HTTP Location cannot carry fragments) - Header nav: removed duplicate "Platform" entry; "Language" is the single top-level link - Footer: rewrote '/platform' → '/language' and '/language#compiler' - Archived originals to _archive/language.v0.astro and platform.v0.astro - TODO(i18n) marker in language.astro frontmatter for locale translations
Three fixes in compile_borrow_primary (x86 + ARM64): 1. [..n] half-open range: when .. appears first in [..n], emit start=0 directly instead of calling compile_primary() which consumed the .. token, causing both range-check and base-check warnings to fire. 2. .as_bytes() builtin: string.as_bytes() is now a no-op pass-through builtin that keeps the string pointer and type, enabling subsequent slice borrows. Previously fell through to field-access dispatch, producing EXPR_TY=0 and a null-pointer segfault. 3. String base for slice borrows: &string_var[a..b] now accepted. Runtime strlen is computed and stored; esiz=1; result is &[i8]. Test: tests/stdlib/runtime_regression/runtime_dynamic_slice.sio PASS. Suite: 548 PASS, 1 FAIL (gpu_vec_add needs CUDA), 229 SKIP.
- Update banner in BaseLayout: correct text, English link, remove isDocsPath suppression so /learn/* critical pages show the notice - Add locale_banner_notice + locale_banner_link to all 6 JSON files; el banner intentionally left in English (no machine translation) - Remove stale TODO(i18n) markers from index, language, proof - Add website/i18n/AUDIT-2026-04-20.md: full locale x route matrix - Add website/i18n/pt/TRANSLATION_TODO.md: Portuguese checklist
|
The latest updates on your projects. Learn more about Vercel for GitHub.
|
github-actions
bot
added
ci
documentation
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
1 participant
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Summary
isDocsPathsuppression so critical pages like/learn/getting-startednow show the bannerlocale_banner_noticeandlocale_banner_linkkeys to all 6 locale JSON filesTODO(i18n)markers from index.astro, language.astro, proof.astrowebsite/i18n/AUDIT-2026-04-20.md: full locale x route coverage matrixwebsite/i18n/pt/TRANSLATION_TODO.md: Portuguese page-by-page checklist for DemetriosGreek locale note
el.jsonis 13 KB vs 7.8–9.4 KB for other locales, indicating a careful prior translation. No Greek text was added or modified. The banner for el is in English pending a native-speaker review.Test plan
/pt/language: shows Portuguese banner with link to/language//el/language: shows English banner (deliberate)/pt/learn/getting-started: shows banner (previously suppressed)