allaun/Research-Stack
1
0
Fork 0
mirror of https://github.com/allaunthefox/Research-Stack.git synced 2026-08-13 15:40:35 +00:00
Code Issues Projects Releases Packages Wiki Activity Actions
624 commits 18 branches 0 tags 425 MiB
Commit graph

624 commits

This branch
This branch
All branches
Author SHA1 Message Date
Devin AI
ca422d43b7 fix(lean): reduce E4_sq_eq_E8_coeff to single modular-form q-expansion gap 
Fully prove the E4^2=E8 Fourier coefficient identity (E4_sq_eq_E8_coeff) modulo one isolated, honestly-named Mathlib gap.

- Add E4_sq_eq_E8_qExpansion: the lone irreducible step (E4^2=E8 as q-expansions). Blocked on dim M8(SL2Z)=1 / valence formula, absent in Mathlib v4.30 (LevelOne.lean proves Module.rank only for weight <= 0).
- Machine-check the entire coefficient extraction for E4_sq_eq_E8_coeff: E4 coeff = 240*sigma3, E8 coeff = 480*sigma7, constant term 1, antidiagonal coeff_mul split into 480*sigma3 boundary + 240^2*convolutionLHS middle, then exact_mod_cast C->N. Previously a single hand-waved sorry.
- Wrap in section ModularFormReduction with local opens (ModularForm EisensteinSeries ModularFormClass) to avoid name clashes.
- Update AGENTS.md + .cursorrules: E8Sidon sorry inventory 6 -> 4.

E8Sidon sorries now: E4_sq_eq_E8_qExpansion (Mathlib-blocked), collision_excess_decrease, greedy_sidon_extraction, e8_singer_improvement.

Build: 3572 jobs, 0 errors (lake build)
Co-Authored-By: Allaun Silverfox <bigdataiscoming+9i37y6j2@protonmail.com>
 2026-06-16 01:19:06 +00:00
Devin AI
5cacb86ddb feat(lean): close sidon_iff_zero_collision (both dirs) and erdos30_e8_conditional 
Close 3 sorry tokens (forward, backward, erdos30):
- sidon_iff_zero_collision forward: double-inclusion proves filter = same∪swap,
  inclusion-exclusion gives card = (2k-1)k exactly.
- sidon_iff_zero_collision backward: cardinality squeeze shows filter = same∪swap
  when excess=0, then classify each element via same/swap membership.
- erdos30_e8_conditional: difference injection into Icc 1 N + trichotomy
  partition gives k(k-1) ≤ 2N, contraposition via nlinarith closes the bound.

E8Sidon sorry count: 9 → 6 (across 6 theorems).
All remaining sorries documented with TODO(lean-port).

Build: 3572 jobs, 0 errors (lake build)
Co-Authored-By: Allaun Silverfox <bigdataiscoming+9i37y6j2@protonmail.com>
 2026-06-15 23:10:53 +00:00
Devin AI
10feb8d717 proof(E8Sidon): close sidon_energy_bound via RRC dimensional classification 
Add finset_pair_eq_iff lemma (the 'dimensional annotation' step):
  {a,b} = {c,d} as Finsets → (a=c ∧ b=d) ∨ (a=d ∧ b=c)
Proved via Set.pair_eq_pair_iff + Finset.coe_inj coercion.

Use it to close sidon_energy_bound (E(S) ≤ 2k²):
  1. Route every filtered quadruple to 'same' or 'swap' shape
  2. Each shape is an image of S×S (card ≤ k²)
  3. Union ≤ same.card + swap.card ≤ 2k²

Inspired by Trail of Bits dimensional analysis technique:
annotate once (finset_pair_eq_iff), validate mechanically.

Sorry count: 9 tokens across 8 theorems (down from 10/9).

Build: 3210 jobs, 0 errors
Co-Authored-By: Allaun Silverfox <bigdataiscoming+9i37y6j2@protonmail.com>
 2026-06-15 22:23:23 +00:00
Devin AI
15fb4bf30c proof(E8Sidon): prove e8_levelset_density via Finset.sup' 
Close sorry for e8_levelset_density by:
- Adding 0 < n precondition (original statement was false at n=0)
- Using Finset.sup' over Icc 1 N to get finite C
- Showing r8 n ≤ C * 1 ≤ C * n^3

Also improves sidon_energy_bound proof sketch with concrete approach.
Sorry count: 10 tokens across 9 theorems (down from 11/10).

Co-Authored-By: Allaun Silverfox <bigdataiscoming+9i37y6j2@protonmail.com>
 2026-06-15 22:09:30 +00:00
Devin AI
86f9c3b537 feat(lean): fully prove greedy_sidon_sqrt and fiber_partition 
greedy_sidon_sqrt: Eliminated last sorry. Full proof uses:
  - offDiag involution (Prod.swap bijects filter(>) with filter(<))
  - card_nbij' for the cardinality bijection
  - mul_tsub for k²-k = k(k-1) factoring
  - card_le_card_of_injOn for the injection into Finset.Icc 1 sup

fiber_partition: New full proof that fiber has even cardinality.
  Same involution technique: split into gt/lt halves via swap bijection.

Sorry count reduced from 12 to 11 (10 theorems).

Build: 3210 jobs, 0 errors
Co-Authored-By: Allaun Silverfox <bigdataiscoming+9i37y6j2@protonmail.com>
 2026-06-15 22:00:58 +00:00
Devin AI
2da4377798 feat(lean): prove greedy_sidon_sqrt injection argument 
The k(k-1)/2 ≤ sup(S) bound now has the full injection proof:
- Defined pairs = offDiag.filter(fun p => p.1 > p.2)
- Proved diff is injective on pairs (via sidon_diff_injective)
- Proved diff maps pairs into Finset.Icc 1 (S.sup id)
- Applied card_le_card_of_injOn to get |pairs| ≤ S.sup id
- Final calc chain: k(k-1)/2 = |pairs| ≤ |Icc 1 sup| = sup

Only remaining sorry: |offDiag.filter(>)| = k(k-1)/2
(offDiag involution σ(a,b)=(b,a) bijects filter(>) with filter(<))

Build: 3210 jobs, 0 errors
Co-Authored-By: Allaun Silverfox <bigdataiscoming+9i37y6j2@protonmail.com>
 2026-06-15 21:51:47 +00:00
Devin AI
3b14e80133 feat(lean): prove sidon_diff_injective and advance §8 greedy extraction 
Fully proved:
- sidon_diff_injective: for Sidon S, all positive differences a-b
  (with a > b, both in S) are distinct. Uses Finset pair membership
  and the Sidon property to derive injectivity.
- exists_collision_witness: ¬IsSidonSet → ∃ collision element.

Advanced (structured proofs with remaining Finset sorries):
- sidon_iff_zero_collision: split into forward/backward with energy counting
- collision_excess_decrease: documented energy decrease argument
- greedy_sidon_extraction: full well-founded induction structure + √ bound path
- greedy_sidon_sqrt: trivial case proved, k≥2 case reduced to Finset card bookkeeping
  (the hard mathematical content is fully captured in sidon_diff_injective)

Build: 3210 jobs, 0 errors (lake build Semantics.E8Sidon)
Co-Authored-By: Allaun Silverfox <bigdataiscoming+9i37y6j2@protonmail.com>
 2026-06-15 21:14:32 +00:00
Devin AI
9e43f50257 chore(nix): remove rs-surface package from flake (Garnix shutting down) 
Remove rsSurface and rsSurfaceImage Nix package definitions from flake.nix.
The Rust source under 4-Infrastructure/infra/embedded_surface/rs-surface/
is preserved — build with cargo directly when needed.

This eliminates the 3 Garnix CI checks (rs-surface, rs-surface-image,
All Garnix checks) that were failing due to the Garnix shutdown.

Co-Authored-By: Allaun Silverfox <bigdataiscoming+9i37y6j2@protonmail.com>
 2026-06-15 01:43:47 +00:00
Devin AI
42b14d9bfe fix(infra): pin base64ct and postgres-types to edition 2021 versions 
base64ct 1.8.3 and postgres-types 0.2.13 require edition 2024 (Rust 1.85+),
which is not available in nixpkgs-24.11 (Rust 1.83). Downgrade:
- base64ct 1.8.3 → 1.7.3
- postgres-types 0.2.13 → 0.2.9

Verified: cargo +1.83.0 build --release succeeds.
Co-Authored-By: Allaun Silverfox <bigdataiscoming+9i37y6j2@protonmail.com>
 2026-06-15 01:41:16 +00:00
Devin AI
23068ae0c5 fix(infra): downgrade rs-surface deps to avoid edition 2024 requirement 
postgres-protocol 0.6.11 pulled in sha2 0.11.0 and cpufeatures 0.3.0,
which require Rust edition 2024 (Rust 1.85+). nixpkgs-24.11 ships
Rust ~1.83, causing the Garnix CI build to fail.

Fix: downgrade tokio-postgres 0.7.17→0.7.13 and postgres-protocol
0.6.11→0.6.9, which use sha2 0.10 and cpufeatures 0.2 (edition 2021).
Verified: cargo build --release succeeds.

Co-Authored-By: Allaun Silverfox <bigdataiscoming+9i37y6j2@protonmail.com>
 2026-06-15 01:36:08 +00:00
Devin AI
07c3504b53 fix(infra): make math-first scripts executable for pre-commit 
Co-Authored-By: Allaun Silverfox <bigdataiscoming+9i37y6j2@protonmail.com>
 2026-06-15 01:26:10 +00:00
Devin AI
0613305be6 feat(infra): add math-first CI scripts and fix wolfram-verification 
Create the five missing scripts referenced by .pre-commit-config.yaml
and .github/workflows/math-check.yml:

- validate_deepseek_receipts.py — validates *.receipt.json against schema
- validate_claims_registry.py — validates claims.yaml against schema + path checks
- require_math_evidence.py — enforces evidence alongside math-track edits
- test_validate_deepseek_receipts.py — 4 self-tests for receipt validator
- test_require_math_evidence.py — 3 self-tests for evidence checker

Fix wolfram-verification workflow:
- Change permissions from issues:write to pull-requests:write (fixes 403)
- Add TODO(wolfram-verify) annotations to E8Sidon.lean false positives
  ("normalized" in docstrings matching the normalize pattern)

Co-Authored-By: Allaun Silverfox <bigdataiscoming+9i37y6j2@protonmail.com>
 2026-06-15 01:24:48 +00:00
Devin AI
a591baeb8c fix(lean): correct sorry inventory count in §12 (10 → 12) 
Co-Authored-By: Allaun Silverfox <bigdataiscoming+9i37y6j2@protonmail.com>
 2026-06-15 01:17:58 +00:00
Devin AI
38b61fc790 feat(lean): add E8Sidon module — Eisenstein coefficient identity and Sidon framework 
New module Semantics.E8Sidon formalizes the connection between E₈ lattice
theta series, Eisenstein series identities, and Sidon set theory.

Definitions (all proven/computable):
- sigma3, sigma7: divisor sum functions via Mathlib ArithmeticFunction.sigma
- convolutionLHS: Cauchy product convolution of σ₃
- IsSidonSet, sidon8, sidonSlack, additiveEnergy, r8
- bernoulli_four, bernoulli_eight: B₄ = B₈ = -1/30 (native_decide)
- E4_normalization, E8_normalization: -(2k/B_k) = 240, 480

Computational witnesses (#eval):
- σ₃(1..4) = 1, 9, 28, 73
- σ₇(1..3) = 1, 129, 2188
- E₄²=E₈ identity verified for n=2,3,4: (61920,61920), (1050240,1050240), (7926240,7926240)
- sidon8.card = 8, sidonSlack = 128

Sorry inventory (12 total, all with TODO(lean-port)):
- E4_sq_eq_E8_coeff: blocked on Mathlib missing valence formula / dim M₈=1
- Sidon theory: energy bound, collision theory, greedy extraction (4)
- E₈ lattice: r8_via_sigma3, r8_one, levelset density (3)
- Conditional: Singer improvement, Erdős bound, fiber partition (3)

Axiom (1): e8_additive_completeness (open problem)

Build: 3572 jobs, 0 errors (lake build)
Co-Authored-By: Allaun Silverfox <bigdataiscoming+9i37y6j2@protonmail.com>
 2026-06-15 01:06:22 +00:00
devin-ai-integration[bot]
0f6ebe36dd
test(coverage): add 138 unit tests for least-covered modules (#78) 
Four test files covering the modules with zero or near-zero test coverage:

- scripts/qc-flag/test_lean_qc_flagger.py (55 tests)
  Covers all pure helpers, code-line detection, QCIssue/FileResult classes,
  and all 5 check functions (structural, naming, Q16, proof quality, deps).

- 4-Infrastructure/shim/test_braid_diat_codec.py (22 tests)
  Covers Q0_2 encode/decode, ChiralityDIAT roundtrips and verify_b,
  MountainPacked dict↔bytes, BraidResidualPacked bracket↔bytes.

- 5-Applications/tools-scripts/llm/test_emitter_utils.py (33 tests)
  Covers sha256_bytes, canonical_json_bytes, repo_relative, safe_slug,
  timestamps, context bundles, message building, extract_answer,
  auth guard, and verify_receipt schema rules.

- 4-Infrastructure/shim/test_validate_rrc_predictions.py (28 tests)
  Covers require_path, parse_equation, build_proof_metrics,
  build_receipt with all domain mappings, and MATH_* constant sets.

Build: all 138 tests pass (pytest), py_compile clean

Co-authored-by: Devin AI <158243242+devin-ai-integration[bot]@users.noreply.github.com>
Co-authored-by: Allaun Silverfox <bigdataiscoming+9i37y6j2@protonmail.com>
 2026-06-14 19:31:16 -05:00
devin-ai-integration[bot]
09a173aa71
fix(scripts): replace bare except clauses with specific exception types (#77) 
Replace 30+ bare `except:` and `except: pass` patterns across 17 files
with specific exception types (OSError, ValueError, etc.) and add
logging so errors are no longer silently swallowed.

Changes by category:

Infrastructure (surface/main.py):
- GPU/process probes now catch FileNotFoundError | SubprocessError

Application scripts:
- API fetchers (finalize_database, final_push, bulk_10x) now catch
  URLError | JSONDecodeError | KeyError | OSError and print to stderr
- Hardware probes (unified_hardware_surface, swarm_transport_layer)
  catch OSError | ValueError for /proc/meminfo reads
- Backend availability checks (prover_backend_interface, hot_swap_manager)
  catch requests.RequestException and log at debug/warning
- Code inspector (swarm_system_inspector) catches OSError | UnicodeDecodeError
  for file reads
- Model fallback (map_all_equations) now logs the intermediate error
- Minor: gpu_pist_compress → RuntimeError | ZeroDivisionError,
  mathlib_to_parquet → ValueError, moe_utils → OSError,
  quandela_remote → OSError | IndexError, topology inline scripts
  → ImportError, consolidate_language_databases → UnicodeDecodeError

All touched files pass py_compile.

Co-authored-by: Devin AI <158243242+devin-ai-integration[bot]@users.noreply.github.com>
Co-authored-by: Allaun Silverfox <bigdataiscoming+9i37y6j2@protonmail.com>
 2026-06-14 19:30:28 -05:00
Allaun Silverfox
a5f77f8d95 feat(lean): Prove all TransportTheory theorems 
- Prove flexure_reduces_resistance: Flexure joints reduce local resistance
- Prove randers_descent_cheap: Downhill motion cheaper than uphill (using discrete Randers metric)
- Prove lower_transport_higher_density: Lower τ → higher T when C constant
- Prove moores_law_improves: Higher C → higher T when τ constant
- Prove tau_scaling_improves: Lower τ → higher T
- Prove lyapunov_descent_is_geodesic_flow: PIST strict_descent = geodesic flow
- Prove transport_cost_is_pist_potential: Transport cost = PIST potential

All theorems proven. No sorry in committed code.

Build: 3562 jobs, 0 errors (lake build)
 2026-06-12 23:24:16 -05:00
Allaun Silverfox
f369864d4c feat(lean): Add TransportTheory module with Finsler-Randers geometry formalization 
- Define TranslationResistance, TransportMetric, and TransportSpace
- Formalize Randers metric F = alpha + beta with Riemannian base and Lyapunov drift
- Connect to PIST via pistMass as transport cost functional
- Define IntelligenceDensity T = C/tau as system capability metric
- Provide executable witnesses for all definitions
- Follow AGENTS.md naming conventions and Q16_16 arithmetic

Build: 3561 jobs, 0 errors (lake build)

Per AGENTS.md: Lean is the source of truth. All computational gates have #eval witnesses.
 2026-06-12 23:14:13 -05:00
Allaun Silverfox
55236fc627 feat(lean): Add TransportTheory module with Finsler-Randers geometry formalization 
- Define TranslationResistance, TransportMetric, and TransportSpace
- Formalize Randers metric F = alpha + beta with Riemannian base and Lyapunov drift
- Connect to PIST via pistMass as transport cost functional
- Define IntelligenceDensity T = C/tau as system capability metric
- Provide executable witnesses for all definitions
- Follow AGENTS.md naming conventions and Q16_16 arithmetic

Build: 3561 jobs, 0 errors (lake build)

Per AGENTS.md: Lean is the source of truth. All computational gates have #eval witnesses.
 2026-06-12 23:12:47 -05:00
Brandon Schneider
3a43f7632f fix(adversarial-review): resolve 35 critical coding bugs across 8 subsystems 
Security & correctness fixes from full adversarial review:

Lean (7 fixes):
- FixedPoint.lean: guard false theorem with n > 0 precondition
- QFactor.lean: remove double-scaling error in energy decrease
- AVMIsa/Step.lean: implement addSatQ16/subSatQ16 primitives
- BraidEigensolid.lean: fix crossStep second output argument swap
- SSMS.lean: complete ACI preservation proof (with rounding caveat)
- HouseholderQR.lean: add n > 0 precondition to spectral theorem

Verilog (7 fixes):
- q16_lut_core.v: fix multiply shift (16 → 32 bits)
- q16_lut_top.v: fix valid bit (0 → 1)
- cff_accelerator.v: fix SHA-256 padding (len < 448 check)
- research_stack_top.v: fix trigger aliasing (unique counters)
- Blitter6502OISC_small.v: fix address width (15 → 16 bits)
- spatial_hash_bram.v: add OOB write guard
- tmr_oepi_safety_fsm.v: fix double-increment race

WGSL (6 fixes):
- shaders.wgsl: atomicAdd for concurrent writes
- frustration_qubo.wgsl: double-buffer + CAS loop
- braid_fft.wgsl: workgroupBarrier synchronization
- burgers_scar_filter.wgsl: atomic E_bins array

Rust (9 fixes):
- thermodynamic.rs: Arc::from_raw → Arc::clone (double-free)
- thermodynamic.rs: Box::into_raw → Box (leak)
- tools/src/lib.rs: shell injection → shlex.quote
- ene-node/src/lib.rs: LRU caps, constant-time HMAC, peer caps

Python (6 fixes):
- similarity/__init__.py: pickle.load → RestrictedUnpickler
- AI-Feynman: torch.load → weights_only=True (14 calls)
- fetch_arxiv.py, fetch_s2.py: eval → ast.literal_eval
- topology.py: os.system → shutil.copy2
- SSH pipe: os.system → base64 pipe

Build: lake build 3572 jobs, 0 errors
 2026-05-31 23:38:03 -05:00
Brandon Schneider
5a691a0d2b docs: add Reality Audit section to k3s-flake README 
Document discrepancies between the aspirational architecture (in the
flake/docs) and what's actually deployed. Covers control plane location,
edge Caddy divergence, Traefik port, cross-namespace services, and
hostname mismatches. Load the research-stack-infrastructure-bootstrap
skill for the actual deployment procedure.

Build: no build needed
 2026-05-31 23:19:42 -05:00
Brandon Schneider
6438dc7214 fix(ingress): correct Authentik forward-auth service namespace 
- Change middleware address from authentik-server.services.svc.cluster.local
  to authentik-server.authentik.svc.cluster.local (Authentik lives in
  authentik namespace, not services)
- Create ExternalName service authentik-server in services namespace so
  the rs-auth Ingress can route to the cross-namespace service

Build: no build needed
 2026-05-31 23:15:46 -05:00
Brandon Schneider
f40e0630bd fix(infra): standardize k3s cluster — Traefik, ingress, architecture docs, Ray GPU workers 
Architecture alignment:
- Rewrite k3s-server.nix from aspirational role=server to actual role=agent
  (cupfox is the real control-plane at 100.110.163.82:6443)
- Update flake.nix: correct serverAddr for all nodes, annotate dead nodes,
  fix hostname from nixos-laptop to nixos
- Update join-agent.sh default SERVER to cupfox
- Document actual vs intended architecture in comments

Ingress:
- Add rs-apps-books Ingress for audiobookshelf (/apps/books → media ns)
- Add strip-apps-books middleware for prefix stripping
- Create ray-ingress.yaml for Ray dashboard at /server/ray

Ray:
- Fix raycluster.yaml: enable dashboard, mount /dev/dri on gpu-workers
- Point gpu-workers to qfox-1 (was neon-64gb)
- Remove nvidia.com/gpu resource dependency (use /dev/dri via Mesa)

Security:
- Move OPENID_CLIENT_SECRET from plaintext to K8s Secret ref
- Update manifest to use valueFrom.secretKeyRef

Cleanup:
- Remove 53 committed test PNG screenshots from git tracking
- Remove auth-state.json from git tracking
- Add *.png, *.json to tests/.gitignore

Build: no build needed
 2026-05-31 23:13:04 -05:00
Brandon Schneider
56615a353b feat(homarr): add Playwright test specs for login, onboard, OIDC, board creation, and integrations 
17 new .spec.ts files and 4 debug .png screenshots capturing:
- OIDC login flows (oidc-setup, oidc-login)
- Login variants (admin, v2, quick, cookie capture)
- Board creation and management
- Integration setup flows
- Onboarding test sequence
- Check/init/do-init lifecycle tests

Build: 0 jobs, 0 errors (no Lean files touched)
 2026-05-31 20:26:16 -05:00
Brandon Schneider
db1781c075 chore(homarr): onboard started, OIDC configured, DB created 
- Homarr OIDC env vars set (sso-homarr)
- Database created at /data/db.sqlite
- Config schema updated to v2
- Onboard wizard needs final UI completion

Build: 3313 jobs, 0 errors
 2026-05-31 15:12:44 -05:00
Brandon Schneider
40655a2805 feat(jellyfin): add media libraries (Movies, TV Shows, Music) 
- Libraries created via API (204 on creation + path update)
- Paths: /media/movies, /media/tv, /media/music
- SSO-Auth plugin Active, SSO button on login page
- Paths may need manual setting through Dashboard UI

Build: 3313 jobs, 0 errors
 2026-05-31 14:55:59 -05:00
Brandon Schneider
0c2245f376 feat(jellyfin): SSO-Auth plugin active, branding configured 
- SSO-Auth plugin Active with Authentik OIDC provider
- SSO redirect endpoint verified (302 -> auth.researchstack.info)
- Login disclaimer and custom CSS for SSO button
- All 22+ services operational with HTTPS

Build: 3313 jobs, 0 errors (lake build Compiler)
 2026-05-31 14:48:21 -05:00
Brandon Schneider
0413c9f7b0 chore(jellyfin): SSO-Auth config file + Playwright setup script 
- SSO-Auth plugin config written to Jellyfin config directory
- Playwright automation script for SSO-Auth setup created
- Plugin not auto-loading - needs UI catalog install

Build: 3313 jobs, 0 errors
 2026-05-31 14:24:44 -05:00
Brandon Schneider
ad68bd49f9 feat(infra): Jellyfin wizard complete, SSO-Auth plugin files deployed 
- Jellyfin setup wizard completed, admin account created (admin/jellyfin-admin)
- SSO-Auth plugin v4.0.0.3 downloaded and placed in plugins directory
- Authentik OIDC provider for Jellyfin updated with correct redirect URI
- All services verified running

Build: 3313 jobs, 0 errors (lake build Compiler)
 2026-05-31 14:14:20 -05:00
Brandon Schneider
71e6b4fc11 feat(infra): complete OIDC config, homarr manifest, vw api verified 
- Homarr OIDC env vars in manifest + applied
- Navidrome, Kavita, Audiobookshelf OIDC env vars verified
- Vaultwarden API running with allaun@researchstack.info
- Jellyfin DB reset - needs web UI first-run setup

Build: 3313 jobs, 0 errors (lake build Compiler)
 2026-05-31 13:48:05 -05:00
Brandon Schneider
2517b04743 fix(infra): configure vaultwarden-api with user credentials 
- Vaultwarden signups enabled then disabled after user created
- vaultwarden-api configured with allaun@researchstack.info
- API verified responding at vaultapi.researchstack.info

Build: 3313 jobs, 0 errors (lake build Compiler)
 2026-05-31 13:36:51 -05:00
Brandon Schneider
54dbac5893 feat(infra): media stack, Vaultwarden, Cloudflare tunnel, OIDC providers, Caddy TLS 
- Media stack deployed: qBittorrent, Sonarr, Radarr, Lidarr, Prowlarr, Bazarr,
  Navidrome, Audiobookshelf, Jellyfin, Homarr, Kavita, MeTube, SuggestArr, BoxArr
- Romm, ebook2audiobook (GPU CUDA), TrailBase, FlareSolverr deployed
- Vaultwarden + Vaultwarden-API deployed
- Caddy reverse proxy on racknerd with Let's Encrypt TLS at *.researchstack.info
- Authentik 7 OIDC providers + 12 proxy providers for SSO
- NVIDIA GPU support enabled on k3s (device plugin + containerd runtime)
- Cloudflare tunnel created (DNS on Porkbun)

Build: 3313 jobs, 0 errors (lake build Compiler)
 2026-05-31 11:01:39 -05:00
Brandon Schneider
92e7accb2a feat(infra): Garage 6-node RF3 cluster, Cloudflare Workers deploy, k3s node additions 
- Garage scale-out complete: 6 nodes across 6 zones with replication_factor=3
- neon-64gb: added to Garage cluster (zone netcup-arm, 93 GiB)
- steamdeck: installed Garage v2.3.0, added to cluster (zone gpu, 373 GiB)
- nixos-laptop: converted from k3s server to agent in cupfox cluster
- qfox-1: k3s agent rejoined cupfox cluster
- Cloudflare Workers: WASM trinary VM deployed at wasm-compute-edge.researchstack.workers.dev
- node-registry.json: updated with all 6 Garage nodes
- Documentation: AGENTS.md, WIKI.md, ROADMAP.md, LLM-Context.md updated
- Lake build: 3313 jobs, 0 errors

Build: 3313 jobs, 0 errors (lake build Compiler)
 2026-05-31 03:39:11 -05:00
Brandon Schneider
79859bc225 feat(infra): migrate external service registry to aiven mysql with ssl 
- Migrated service_registry.py defaults from InfinityFree to Aiven MySQL.
- Enforced REQUIRED SSL mode on connections (ssl_mode for mysql-connector, ssl dict for pymysql).
- Added helper _get_dict_cursor to resolve pymysql compatibility issues.
- Configured registry password and encryption key in decrypted/encrypted .env via SOPS.

Build: 3313 jobs, 0 errors (lake build)
 2026-05-31 02:10:10 -05:00
Brandon Schneider
c5724734a6 chore(infra): bootstrap racknerd and configure multi-node garage cluster 
- Bootstrapped microvm-racknerd node on its live Tailscale IP 100.80.39.40.
- Connected racknerd to the central QFox cluster.
- Assigned racknerd to its own vps zone with 1G capacity.
- Made garage-cluster-init.sh robust and idempotent.
- Updated comments and topology mappings.

Build: 3313 jobs, 0 errors (lake build)
 2026-05-31 02:04:32 -05:00
Brandon Schneider
36dc2f096a fix(infra): service registry is backup/fallback, not primary 
Primary: Tailscale mesh + internal PostgreSQL/SQLite
Backup:  InfinityFree MySQL (this file)

Use cases for backup path:
  - Edge nodes that cant reach mesh (ESP32, Cloudflare Workers)
  - Mesh-down fallback (Tailscale outage)
  - Cross-mesh discovery (different tailnets)
  - Low-impact config distribution
 2026-05-30 21:19:24 -05:00
Brandon Schneider
886428d4a0 feat(infra): external service registry via InfinityFree MySQL 
service_registry.py — mesh-independent node discovery and credential store.

Tables:
  nodes       — registered devices with capabilities, tier, IPs
  credentials — encrypted blobs (ChaCha20) with TTL auto-expiry
  config      — distributed key-value configuration

Features:
  - auto_register() — uses device_capability_probe to register
  - discover_nodes() — find nodes by tier, with max-age filter
  - store/get_credential() — encrypted at rest, short TTL
  - heartbeat() — keepalive for node registry
  - CLI: init, register, discover, store, get, cleanup, config-set/get

Any node with internet can reach it (no Tailscale required).
Credentials encrypted with ChaCha20, key from REGISTRY_ENCRYPT_KEY env.
 2026-05-30 21:14:51 -05:00
Brandon Schneider
da02dc4051 fix(docs): full soft color override — entire page #f5f0eb 
No white anywhere. Everything is warm soft gray:
  Body/content: #f5f0eb (warm parchment)
  Header: #3d3d3d (muted dark, no green)
  Headings: #4a4540
  Code inline: #ebe6e0 bg
  Code blocks: #3d3d3d bg
  Tables: #e5e0da headers
  Links: #5a7a8a muted blue
  All with !important to override Cayman defaults
 2026-05-30 21:07:18 -05:00
Brandon Schneider
46d9828915 fix(docs): correct header selector for Cayman theme 
.page-header not .header — Cayman uses page-header class.
Also override heading colors to match dark theme.
 2026-05-30 21:00:32 -05:00
Brandon Schneider
4f5da735ac docs: soft theme for GitHub Pages — Cayman + muted palette 
- Cayman remote theme (soft blue gradient header)
- Body: #f0f2f5 soft gray (not glaring white)
- Content: #ffffff with subtle shadow
- Code blocks: dark (#2d3436) with soft text
- Tables: muted headers (#dfe6e9)
- Links: #0984e3 blue, hover #74b9ff
 2026-05-30 20:57:50 -05:00
Brandon Schneider
108b825ef3 docs: hyperlink 50x claim to VCN Pipeline compression table 2026-05-30 20:55:06 -05:00
Brandon Schneider
a8ed01dcfa docs: update wiki with VAAPI, FLAC DSP, tier limitations, Cloudflare/GitHub 2026-05-30 20:52:58 -05:00
Brandon Schneider
a2c76c7ae0 fix(infra): tier limits leave headroom for device function 
Each tier now reserves resources for its primary function:
  GPU_CUDA:     8GB VRAM (not 12 — keep 4GB for display/compositor)
  GPU_VAAPI:    256MB (keep VRAM headroom for desktop)
  GPU_APU:      128MB (shared DDR, OS needs bandwidth)
  CPU_FFMPEG:   64MB, half cores (leave for OS/k3s)
  BATCH:        1500 min/month (reserve 500 for actual CI/CD)
  ETHERNET:     500ms timeout (leave bandwidth for SSH/mgmt)
  FRAMEBUFFER:  768KB (half — keep display visible, compute in top rows)
  WASM:         512B payload, 8ms CPU (leave 2ms for JSON overhead)
  DSP:          2048 samples (half FFT, leave for overlap buffer)
  ESP32:        512B (WiFi/BLE stack needs ~80KB of 520KB SRAM)
 2026-05-30 20:45:50 -05:00
Brandon Schneider
78d41bc474 feat(infra): per-tier device limitations for Ray scheduling 
DeviceLimitations dataclass with hard constraints per tier:
  GPU_CUDA:     1GB payload, 16 concurrent, 60s, NVENC, 12GB VRAM
  GPU_VAAPI:    512MB payload, 8 concurrent, 60s, VAAPI HW
  GPU_APU:      256MB payload, 4 concurrent, 30s, shared DDR
  CPU_FFMPEG:   128MB payload, 2 concurrent, 120s, software
  BATCH:        64MB payload, 1 concurrent, 6h, 2000 min/month
  ETHERNET:     1400B payload, 1 concurrent, 1s, virtio-net
  FRAMEBUFFER:  1.5MB payload, 1 concurrent, 100ms, DMA only
  WASM:         1KB payload, 1 concurrent, 10ms, 100K req/day
  DSP:          16KB payload, 1 concurrent, 5s, FFT only
  ESP32:        2KB payload, 1 concurrent, 100ms, Q0_16 scalar

get_limitations(caps) returns actual hardware-aware limits
(vram override, framebuffer capacity, memory override)
 2026-05-30 20:44:31 -05:00
Brandon Schneider
e273d0268c feat(infra): add AMD VAAPI + FLAC DSP to FrameDispatcher 
FrameDispatcher now routes 6 tags:
  TAG_STRAND(0x01)  → BraidBackend (VCN compute)
  TAG_CROSSING(0x02) → BraidBackend (VCN compute)
  TAG_PIST(0x03)    → BraidBackend (VCN compute)
  TAG_LUPINE(0x04)  → CUDABackend (NVIDIA CUDA)
  TAG_VAAPI(0x05)   → VAAPIBackend (AMD/Intel VA-API)  ← NEW
  TAG_FLAC(0x06)    → FLACBackend (PipeWire/FLAC DSP)  ← NEW

New backends:
  - VAAPIBackend/LocalVAAPIBackend: AMD/Intel hardware encode/decode
  - FLACBackend/LocalFLACBackend: FFT spectral analysis, centroid, RMS
  - RayVAAPIBackend: Ray actor for VA-API operations
  - SyncVAAPIWrapper/SyncFLACWrapper: sync bridges for FrameDispatcher

Capability probe: DSP tier(1) added between FRAMEBUFFER(2) and ESP32(0)
 2026-05-30 20:35:42 -05:00
Brandon Schneider
55dc2dbeaf docs: add wiki to GitHub Pages source (/docs) 2026-05-30 20:17:01 -05:00
Brandon Schneider
dd5b7382b6 docs: comprehensive wiki — full architecture reference (May 2026) 
512-line wiki documenting the entire Research Stack:
  1. Architecture Overview (5-layer stack)
  2. Compute Tiers (GPU_CUDA through OFFLINE)
  3. Infrastructure (k3s, Tailscale, KubeRay)
  4. VCN Pipeline (50x compression)
  5. Ray Integration (FrameDispatcher over Ray)
  6. Device Capability Probe (multi-GPU, framebuffer fallback)
  7. Compute Surfaces (GPU, Ethernet, framebuffer, ESP32)
  8. Mesh Networking Plan (Ray over Tailscale)
  9. Formal Verification (23 Lean files, 12 sorries)
  10. Key Files + Quick Reference
 2026-05-30 20:11:58 -05:00
Brandon Schneider
734d9ef125 feat(infra): integrate edge WASM and GitHub batch compute tiers 
- Update device_capability_probe.py to add BATCH and WASM tiers and fix a NameError bug on has_virtio_net.
- Build Cloudflare Workers WASM compilation and JS fetch handler in 4-Infrastructure/cloudflare/ executing trinary VM steps.
- Create GitHub Actions batch_compute.yml workflow to harvest runner minutes.
- Keep 4-Infrastructure/AGENTS.md updated with the WASM core library anchor.

Build: 3313 jobs, 0 errors (lake build)
 2026-05-30 20:08:31 -05:00
Brandon Schneider
c4f7fa7f13 docs(infra): plan — mesh networking layers over Ray 
Architectural plan for Tailscale mesh + Ray + VCN + compute surfaces.

5-layer stack:
  1. Tailscale connects everything (WireGuard, DERP relay)
  2. Ray schedules work across the mesh
  3. VCN compresses data (50x bandwidth reduction)
  4. FrameDispatcher routes by tag
  5. Compute surfaces execute (GPU/CPU/Ethernet/framebuffer/MCU)

Key insight: every device in the Tailscale mesh is a potential
compute node. Framebuffer and Ethernet surfaces turn devices
that "cant run Ray" into compute participants.

4 phases:
  1. Ray over Tailscale (mostly done)
  2. Multi-tier scheduling (probe done, placement pending)
  3. Framebuffer + Ethernet integration (host-side pending)
  4. Edge devices (ESP32, 1-Wire sensors)
 2026-05-30 20:03:57 -05:00
Brandon Schneider
745fb9757c feat(infra): add ETHERNET compute tier for virtio-net PistPacket DMA 
New tier between CPU_FFMPEG and FRAMEBUFFER:
  GPU_CUDA(7) > GPU_VAAPI(6) > GPU_APU(5) > CPU_FFMPEG(4) >
  ETHERNET(3) > FRAMEBUFFER(2) > ESP32(1) > RELAY(0)

- _detect_virtio_net(): probes /sys/class/net for virtio driver (0x1af4)
- PistPacket computation via TX/RX descriptor rings
- Host vhost-user backend does matrix transforms
- CRC32 hardware offload = witness verification
- Works in any VM with network (even without framebuffer)
 2026-05-30 20:02:06 -05:00