silken_net

The project provides a clear upgrade path to newer versions. It uses Semantic Versioning (managed by release-please) and publishes a human-readable CHANGELOG documenting the features and fixes in each release, so users can see exactly what changed when moving between versions. The backend is a continuously-deployed service (users run the latest); the firmware and smart contracts follow the same versioned releases. At this stage (v0.x) upgrades are not breaking-complex, so updating to the newest release is the upgrade path, and any future breaking change will be documented in the CHANGELOG / release notes with the steps needed.
https://github.com/Alexey-Lukin/silken_net/blob/main/CHANGELOG.md
https://github.com/Alexey-Lukin/silken_net/releases

The project uses GitHub Issues as its issue tracker for individual issues.
https://github.com/Alexey-Lukin/silken_net/issues

N/A — no vulnerability reports have been received or resolved in the last 12 months (zero repository security advisories), so there are no reporters to credit. The credit practice for when a report is resolved is now documented in SECURITY.md ("Credit"): reporters are credited in the advisory and/or release notes unless they request anonymity.
https://github.com/Alexey-Lukin/silken_net/blob/main/SECURITY.md

The process for responding to vulnerability reports is documented in SECURITY.md: reports are received privately via GitHub Security Advisories, acknowledged within 72 hours, with a fix or mitigation for high-severity issues targeted within 14 days; the in-scope components, openly-tracked known limitations, and a safe-harbor statement are also included. The open security backlog is tracked in docs/00_07 (SEC.*).
https://github.com/Alexey-Lukin/silken_net/blob/main/SECURITY.md

The project identifies a specific coding style guide for each primary language, documented in CONTRIBUTING.md ("Coding standards"): Ruby — the Rails Omakase RuboCop style (.rubocop.yml inherits rubocop-rails-omakase); Python — Ruff (ruff.toml, a curated pycodestyle/pyflakes/isort/pyupgrade/bugbear ruleset); firmware C — -Wall -Wextra -Wpedantic + cppcheck (MISRA C:2012 advisory); Solidity — forge fmt; C# — .editorconfig (.NET conventions). Contributions are required to comply (CI must be green).
https://github.com/Alexey-Lukin/silken_net/blob/main/CONTRIBUTING.md#requirements-for-acceptable-contributions

The selected styles are enforced automatically in CI by FLOSS tools, and a contribution that violates them fails the build: RuboCop (Ruby) and Ruff (Python) in the ci.yml lint jobs, cppcheck for firmware C (firmware_lint), and forge fmt / Slither for Solidity (solidity_audit.yml). Tool configs live in the repo (.rubocop.yml, ruff.toml, contracts/foundry.toml, tools/cad/.editorconfig). Exceptions are rare and documented in code at their locations — e.g. inline // cppcheck-suppress with a reason, and fingerprinted, annotated entries in config/brakeman.ignore.
https://github.com/Alexey-Lukin/silken_net/blob/main/.github/workflows/ci.yml

The native-binary build (firmware/test/Makefile) honors the compiler/linker variables passed in via the environment or command line: CC ?= gcc (a packager can pass CC=clang) and CFLAGS += <project flags> (env/CLI CFLAGS are honored and the project EXTENDS them with its mandatory -std=c11 -I., never replacing). The recipes are single-invocation gcc compile+link, so linker flags carried in CFLAGS (e.g. -Wl,-z,relro) reach the linker too — making it easy to enable ASAN or distribution-hardening flags. Verified: an env CFLAGS value passes through to the compiler, and CI (firmware_test) is green. (The Rails backend and Solidity contracts produce no native binaries.)
https://github.com/Alexey-Lukin/silken_net/blob/main/firmware/test/Makefile

The build preserves debugging information when requested: passing CFLAGS="-g" is honored by the Makefile (CFLAGS += extends the user's flags) so debug info is generated, and the firmware host tests run in place and are never installed with stripping (there is no "install -s" / strip step). The dedicated coverage (--coverage) and ASAN (-g) lanes also build with debug/instrumentation symbols. (The Rails backend and Solidity contracts have no native-binary build/install that strips symbols.)
https://github.com/Alexey-Lukin/silken_net/blob/main/firmware/test/Makefile

The build systems do not use recursive make over cross-dependent subdirectories. The only native build (firmware/test/Makefile) is a single, non-recursive Makefile: each target lists its full header/source dependencies and the shared firmware/common headers compile directly into each test via -I (no per-subdirectory independent builds). Its only $(MAKE) calls recurse into the same directory (the asan/coverage re-instrumented rebuilds), never into subdirectories, so the dependency information is complete and accurate. The other toolchains (Bundler, Foundry/forge, npm, conda) are not recursive-make either.
https://github.com/Alexey-Lukin/silken_net/blob/main/firmware/test/Makefile

Builds and code-generation are reproducible bit-for-bit, and CI enforces it. Generated firmware artifacts are checked against their source on every run: tools/ml/scripts/check_firmware_tables.py regenerates the log-mel tables and fails if they differ from the committed headers, and tools/firmware/check_bytecode.py verifies the committed lorenz_bytecode.h matches bio_contract.rb; QEMU jobs additionally assert host↔Cortex-M4 bit-parity. Solidity bytecode is deterministic — solc is pinned (solc_version = 0.8.35) with fixed optimizer settings (foundry.toml), so forge build reproduces the same bytecode. (The Rails backend is interpreted Ruby — no compiled artifact.)
https://github.com/Alexey-Lukin/silken_net/blob/main/.github/workflows/ci.yml

The software is installed via commonly-used conventions. The backend is packaged as a container in a standard format (Dockerfile, multi-stage build on ruby:4.0.5-slim) and deployed with Kamal (config/deploy.yml: image, registry, servers) — kamal setup / kamal deploy installs it, kamal remove uninstalls it. For development it installs at the language level via Bundler (bundle install) per the README Quick Start, and uninstalling is removing the directory/containers. The smart contracts deploy on-chain via Foundry scripts.
https://github.com/Alexey-Lukin/silken_net/blob/main/Dockerfile
https://github.com/Alexey-Lukin/silken_net#readme

N/A — the project has no installation system that writes built artifacts to a
user-selectable location, so the DESTDIR/PREFIX convention does not apply. The
end-user software (the D-MRV Rails platform) is delivered as a self-contained
container image (Docker, deployed by Kamal — config/deploy.yml) or run in place
from its working directory after a Bundler install; neither has a DESTDIR/PREFIX
step. There is no make install / PREFIX / DESTDIR target anywhere in the repo.
The one first-party Python package (tools/ml, "silken-ml") is internal ML tooling
run in place (PYTHONPATH=src) or via an editable pip install -e inside a conda
env — an editable install links the source tree rather than writing built
artifacts to a chosen location — so it is not an end-user installation system
either. (CMSIS-DSP and mruby ship their own packaging but are vendored
third-party submodules, not this project's installation system.)

bin/setup (the standard idempotent Rails dev-bootstrap script — runs
bundle install, bin/rails db:prepare, clears logs, starts the dev server).
The README "Quick Start" documents the system-package prerequisites, the same
steps, and how to run the test suite (bundle exec rspec). The repo is polyglot,
so each domain installs its support + test environment via its standard language
package manager, all listed in CONTRIBUTING.md: Bundler for Ruby/Rails (tests:
rspec); npm ci + Foundry for the Solidity contracts (tests: forge test); a
conda environment file for the Python in-silico/ML tooling
(tools/*/environment.yml); and make -C firmware/test for the host-based
firmware tests (no ARM toolchain needed). A Dockerfile is also provided as the
generated-container path.
https://github.com/Alexey-Lukin/silken_net/blob/main/bin/setup
https://github.com/Alexey-Lukin/silken_net/blob/main/CONTRIBUTING.md
https://github.com/Alexey-Lukin/silken_net#readme

• Ruby/Rails: Gemfile + Gemfile.lock (Bundler)
• Solidity contracts: contracts/package.json + package-lock.json (npm: OpenZeppelin, forge-std)
• Python in-silico/ML: tools/ml/pyproject.toml, tools/*/environment.yml + tools/in_silico/conda-lock.yml (setuptools + conda, pinned lock)
• .NET CAD tooling: *.csproj PackageReference entries (tools/cad/src/**)
• Vendored C/firmware + CAD libraries: .gitmodules (CMSIS, mruby, Monocypher, STM32 HAL, LEAP71 — pinned to exact commits)
• CI: GitHub Actions pinned by commit SHA in .github/workflows/*.yml
  https://github.com/Alexey-Lukin/silken_net/blob/main/Gemfile
  https://github.com/Alexey-Lukin/silken_net/blob/main/.gitmodules
  https://github.com/Alexey-Lukin/silken_net/blob/main/contracts/package.json

External dependencies are monitored continuously and on every CI run across
the polyglot stack:

• Dependabot (.github/dependabot.yml) — weekly, 5 ecosystems: bundler (gems),
  docker (base-image tag+digest), github-actions, npm (contracts/), terraform.
• bundler-audit (the OpenSSF-listed Ruby SCA tool) runs in CI on every push
  (.github/workflows/ci.yml) against the ruby-advisory-db; brakeman runs SAST.
• Slither runs SCA / static analysis on the Solidity contracts
  (.github/workflows/solidity_audit.yml).
• OpenSSF Scorecard runs weekly + on push (.github/workflows/scorecard.yml) and
  publishes results; its Vulnerabilities check queries osv.dev and its
  Dependency-Update-Tool check confirms Dependabot is active.
  The conda-managed Python research tooling and the pinned git-submodule vendored
  C libraries are not on Dependabot, but they are development/in-silico only and
  sit outside the deployed runtime trust boundary (the production service is the
  Rails app + its gems + the Docker base image, all of which are monitored).
  https://github.com/Alexey-Lukin/silken_net/blob/main/.github/dependabot.yml
  https://github.com/Alexey-Lukin/silken_net/blob/main/.github/workflows/scorecard.yml

Met (via both routes — standard package managers and easy-to-update pins; no
forked convenience copies). Every reused component is identified in a
computer-processable manifest and updated by a standard mechanism:

• Gems (Gemfile/lock, bundle update), npm for contracts, conda for the Python
  tooling, .NET PackageReference, and JS via Rails importmap (bin/importmap pin).
• Vendored C/firmware + CAD libraries are git submodules pinned to the CANONICAL
  UPSTREAM repositories (.gitmodules: ARM-software/CMSIS, mruby/mruby,
  LoupVaillant/Monocypher, STMicroelectronics HAL, leap71) — they are not forks,
  so a security fix is applied by bumping the submodule pin to a new upstream tag.
• Front-end JS is supplied by the Rails framework gems (turbo/stimulus/activestorage)
  plus one version-explicit CDN pin (leaflet@1.9.4), updateable via bin/importmap.
• OpenSSL and PostgreSQL are standard system components.
  Dependabot automates update PRs for the package-managed ecosystems weekly, so a
  vulnerable reused component is straightforward to update.
  https://github.com/Alexey-Lukin/silken_net/blob/main/.gitmodules
  https://github.com/Alexey-Lukin/silken_net/blob/main/.github/dependabot.yml

Met. The project runs a current technology stack (Ruby 4.0.5, Rails 8.1,
PostgreSQL 17, Solidity 0.8.35 / EVM cancun, .NET 9) — no EOL or obsolete
frameworks — and actively flags deprecated/obsolete APIs in CI so the FLOSS-current
alternative is used:

• Python: Ruff's pyupgrade ("UP") rule-set is enabled (ruff.toml select), and the
  ML test suite escalates DeprecationWarning to a hard error
  (tools/ml/pyproject.toml: filterwarnings = ["error::DeprecationWarning"]).
• Ruby: RuboCop (rails-omakase) flags deprecated Rails/Ruby idioms.
• C firmware: -Wall -Wextra -Wpedantic + cppcheck warn on deprecated functions.
• Solidity: a fixed current solc pragma + forge fmt.
  Weekly Dependabot keeps library APIs from going stale, so the current alternative
  is always within one PR.
  https://github.com/Alexey-Lukin/silken_net/blob/main/ruff.toml
  https://github.com/Alexey-Lukin/silken_net/blob/main/tools/ml/pyproject.toml

Met. A GitHub Actions automated test suite runs on every check-in — triggered on
push to main and on every pull_request (.github/workflows/ci.yml) — and
produces a pass/fail report via the Actions checks UI and the ci-ok aggregate
required check. The suite spans the polyglot stack: RSpec for the Rails backend
(bundle exec rspec), the firmware host-test suite (make -C firmware/test, plus
an ASan/UBSan lane), Foundry forge test for the Solidity contracts
(solidity_audit.yml), and language smoke workflows (ml_smoke, in_silico_smoke,
cad_smoke, coap_smoke).
https://github.com/Alexey-Lukin/silken_net/blob/main/.github/workflows/ci.yml
https://github.com/Alexey-Lukin/silken_net/actions

Met. The standing practice is to land a bug fix together with a regression test in
the same commit/PR. An audit of the bug-fix commits over the last six months shows
a clear majority shipped with a test change in the same commit; the share is higher
for the shipped product code (Rails backend, firmware, smart contracts) — e.g.
"Fix OTA packager 8-bit overflow + implement missing CRC16" (firmware host tests),
"fix: rollback receipt envelope, OTA HMAC trailer, Celo double-pay" (RSpec),
"fix(queen): FW.51 — telemetry not lost on CoAP-flush failure" (firmware host
tests), "Fix double-anchoring risk in EthereumAnchorWorker" (RSpec). Regression
tests live in the automated suites: RSpec (spec/), firmware host tests
(firmware/test/), and Foundry (contracts/test/). A class of in-silico
research-script tuning fixes (simulation NaN/SCF convergence) are not
unit-regression-testable and are outside the shipped-product bug count.
https://github.com/Alexey-Lukin/silken_net/tree/main/spec
https://github.com/Alexey-Lukin/silken_net/blob/main/CONTRIBUTING.md#requirements-for-acceptable-contributions

The Ruby/Rails backend — the bulk of the codebase — is measured by SimpleCov
(FLOSS, with branch coverage enabled), and the full CI suite enforces a hard gate of
minimum_coverage line: 99, branch: 95 (spec/spec_helper.rb): the build fails below
99% statement coverage, well above the 80% bar. A per-group tripwire additionally
floors Services/Workers/Models at ~99% line so no single area can erode while the
global average holds. The other languages are measured with FLOSS coverage tools too:
the firmware C host suite via gcov/lcov (make -C firmware/test coverage) and the
Solidity contracts via forge coverage --report lcov (→ Codecov). The coverage-scope
policy is documented in docs/04_06 §B.
https://github.com/Alexey-Lukin/silken_net/blob/main/spec/spec_helper.rb
https://github.com/Alexey-Lukin/silken_net/blob/main/.github/workflows/ci.yml

The project has a formal written testing policy in CONTRIBUTING.md
("Requirements for acceptable contributions"): "Tests are mandatory for new
functionality (project policy). When you add or change functionality you MUST add
or update automated tests; as major new functionality is added, tests for it MUST
be added to the relevant automated suite (RSpec / Foundry forge / firmware host
tests). A pull request that adds major new functionality without accompanying
tests will not be merged." The policy is backed in practice by the project's
regression-test record (see regression_tests_added50) and the SimpleCov 99%
line-coverage gate enforced in CI.
https://github.com/Alexey-Lukin/silken_net/blob/main/CONTRIBUTING.md#requirements-for-acceptable-contributions

The add-tests policy is documented in CONTRIBUTING.md under "Requirements for acceptable contributions".
https://github.com/Alexey-Lukin/silken_net/blob/main/CONTRIBUTING.md#requirements-for-acceptable-contributions

Met — strict where practical, enforced in CI across every language:

• Firmware C: compiled with -Wall -Wextra -Wpedantic (-std=c11); cppcheck is a hard
  CI gate (firmware_lint: --enable=warning,performance,portability,style,
  --error-exitcode=1, MISRA C:2012 addon, --check-level=exhaustive), plus an
  ASan/UBSan runtime lane.
• Ruby: RuboCop (rails-omakase) gates CI and fails on any offense.
• Python: Ruff's strict rule-set (E/W/F/I/UP/B/C4/SIM/RUF) gates CI, and the test
  suite escalates DeprecationWarning to an error.
• Solidity: forge fmt --check and Slither static analysis gate CI.
• Security SAST: Brakeman gates CI.
• .NET CAD: default .NET analyzers run on the first-party code; strictness is
  relaxed only in the third-party LEAP71 wrapper, where it is not practical.
  https://github.com/Alexey-Lukin/silken_net/blob/main/firmware/test/Makefile
  https://github.com/Alexey-Lukin/silken_net/blob/main/.github/workflows/ci.yml

Met. Secure design principles are implemented across the stack:

• Fail-safe / deny-by-default: minting refuses unless every condition holds —
  BlockchainMintingService raises "Security Breach" unless the telemetry is
  verified_by_iotex?, oracle_status_fulfilled?, and KYC-approved; in production
  WEB3_STRICT_MODE=true turns missing security config (e.g. CHAINLINK_HMAC_SECRET)
  into a hard raise instead of a silent fallback.
• Secure by default (deployment): production enforces config.force_ssl + HSTS
  (preload, 1-year, subdomains), a Content-Security-Policy with a per-request
  nonce, X-Frame-Options: DENY, X-Content-Type-Options: nosniff, a Permissions-
  Policy, and session cookies set httponly + same_site:lax + secure.
• Complete mediation: per-resource Pundit policies (app/policies/*) plus role gates
  in the API base controller (authorize_admin!/super_admin!/forester!); thin
  controllers and AASM state machines make state changes non-bypassable.
• Least privilege / separation of privilege: on-chain roles are split and gated by
  onlyRole(...), admin actions are routed through a Timelock, and mint()/slash()
  run on physically separate keys (MINTER_ROLE vs SLASHER_ROLE).
• Defense in depth: a manual_review double-spend guard freezes funds when a
  transaction state is unknown; anomaly/tamper telemetry is zeroed for minting in
  BOTH firmware and backend; no weak crypto is used (no MD5/SHA-1/DES/RC4) —
  Argon2id passwords, an HKDF/HMAC key ratchet (NIST SP 800-108), AES + Ed25519.
• Economy of mechanism: an explicit YAGNI "lazy-senior" ladder + Ruthless Pruning
  keep mechanisms minimal (CLAUDE.md §4), with input validation at trust boundaries.

https://github.com/Alexey-Lukin/silken_net/blob/main/docs/03_05_Hardware_Symmetric_Crypto_and_Security.md
https://github.com/Alexey-Lukin/silken_net/blob/main/config/environments/production.rb

Verified across the whole stack — no algorithms with serious known weaknesses are used. Hashes: SHA-256 / HMAC-SHA256 (keccak256 in contracts); no SHA-1 or MD5 anywhere. Password hashing: Argon2id. Signatures: Ed25519. RNG: SecureRandom (backend) + hardware TRNG (firmware). The CoAP backhaul is AES-256-CBC with a fresh random IV per message (semantic security; not the SSH-CBC weakness); the LoRa target is authenticated AES-128-CCM. The only weak mode is the transitional LoRa AES-128-ECB, whose risk and mitigations are documented in docs/03_05 and which is migrating to AES-128-CCM (FW.2).
https://github.com/Alexey-Lukin/silken_net/blob/main/docs/03_05_Hardware_Symmetric_Crypto_and_Security.md

Met (SHOULD). The project uses multiple cryptographic algorithm families and is
built to switch primitives rather than hard-wiring one:

• Hashes: both SHA-2 (SHA-256 — backend integrity, HMAC-SHA256, HKDF, audit-log
  hash chain, firmware) and SHA-3 (keccak256 — Solidity roles/parameter keys and
  Eth::Util.keccak256 on the backend) are in active use.
• Signatures: two schemes — Ed25519 (SE050 device attestation, peaq DID, M2M auth)
  and secp256k1/ECDSA (EVM chains, IoTeX verification).
• Symmetric: AES in four selectable modes (ECB, CCM, CBC, GCM). The backend names
  the cipher as an OpenSSL string (OpenSSL::Cipher.new("aes-256-cbc")), so switching
  to any OpenSSL-supported cipher is a one-line change, and the live
  ECB->authenticated-CCM migration is runtime-selectable via the
  TELEMETRY_CCM_ENABLED / FW2_CCM_ENABLED flags — a working demonstration of
  switching a primitive when one is found weak. A documented PQC migration roadmap
  (docs/03_05 §10) plans the post-quantum transition.
  The symmetric cipher is AES (not Twofish/Serpent) because the STM32 nodes are bound
  to the hardware AES engine; agility there is at the mode level plus the documented
  migration roadmap.
  https://github.com/Alexey-Lukin/silken_net/blob/main/docs/03_05_Hardware_Symmetric_Crypto_and_Security.md

Met. Credentials and private keys are stored in dedicated files/stores — separate
from code, in-repo config, the database, and logs — and are replaceable at runtime
without recompiling (Ruby is interpreted; secrets are read at boot):

• App secrets (DB password, API/RPC keys, the Chainlink HMAC secret, etc.) live in
  Rails' encrypted credentials (config/credentials.yml.enc) decrypted by a separate
  config/master.key, and/or in environment variables (config/database.yml and code
  read them via ENV.fetch / Rails.application.credentials). master.key and .env* are
  git-ignored; at deploy RAILS_MASTER_KEY and other secrets are injected via Kamal
  secrets (.kamal/secrets) from the environment / a secrets manager, never committed.
• User passwords are stored only as Argon2id hashes (HasArgon2Password concern,
  OWASP-recommended), never in plaintext.
• Private cryptographic keys are rotatable without code changes: a key ratchet
  (Cryptography::KeyRatchet), an OTA HMAC-key service, and an over-the-air
  key-rotation worker replace device keys at runtime; the SE050 secure element holds
  non-extractable keys.
• Secrets are kept out of logs by an explicit filter_parameters allowlist
  (passwords, tokens, *_key, private_key, mnemonic, wallet_private_key, signature…).
  https://github.com/Alexey-Lukin/silken_net/blob/main/.kamal/secrets
  https://github.com/Alexey-Lukin/silken_net/blob/main/config/initializers/filter_parameter_logging.rb

Met (SHOULD). All IP/web network communications use TLS by default and no insecure
protocol is enabled:

• The web app and API enforce HTTPS in production (config.force_ssl = true) with
  HSTS (1 year, includeSubdomains, preload) and assume_ssl behind the TLS-terminating
  Kamal proxy (Let's Encrypt, TLS 1.2/1.3). HTTP is redirected to HTTPS.
• All outbound calls use HTTPS — chain RPC (Alchemy/Infura), Chainlink Functions,
  IoTeX and peaq endpoints — with default certificate verification (no VERIFY_NONE).
• A scan of the code finds no FTP, telnet, SSLv3/earlier, SSHv1, or plain-HTTP
  endpoints.
  For the constrained IoT radio link (Soldier->Queen LoRa, Queen->Rails CoAP) TLS/DTLS
  is not feasible on a tens-of-bytes LoRa frame, so confidentiality and integrity are
  provided at the application layer with AES: AES-128-CCM (authenticated; the
  LoRaWAN/Zigbee/Thread/BLE golden standard for constrained IoT) on the LoRa link and
  AES-256-CBC with a per-message random IV on the CoAP backhaul. This design — and why
  heavier schemes such as Kyber do not fit the radio MTU — is documented in docs/03_05.
  The transitional AES-128-ECB LoRa mode is documented and is migrating to AES-128-CCM.
  https://github.com/Alexey-Lukin/silken_net/blob/main/config/environments/production.rb
  https://github.com/Alexey-Lukin/silken_net/blob/main/docs/03_05_Hardware_Symmetric_Crypto_and_Security.md

Met (SHOULD). The project uses TLS and supports TLS 1.2+ everywhere; nothing is
configured to allow TLS 1.1 or earlier:

• Inbound HTTPS is terminated by the Kamal proxy with automatic Let's Encrypt
  certificates (config/deploy.yml, proxy ssl: true); the proxy (Go crypto/tls)
  negotiates TLS 1.2/1.3 and does not offer TLS 1.0/1.1. The Rails app enforces it
  with config.force_ssl = true and HSTS (1 year, includeSubdomains, preload) in
  production.rb, so HTTP is redirected to HTTPS.
• Outbound HTTPS (chain RPC, Chainlink, IoTeX, peaq) is made by Ruby 4.0.5 on
  OpenSSL 3.6.2, which negotiates TLS 1.2/1.3 by default and has TLS 1.0/1.1
  disabled; no client sets a lower ssl_version/min_version.
  No SSLv2/SSLv3 or TLS 1.1-or-earlier is configured or supported anywhere in the stack.
  https://github.com/Alexey-Lukin/silken_net/blob/main/config/environments/production.rb
  https://github.com/Alexey-Lukin/silken_net/blob/main/config/deploy.yml

Met. TLS certificate verification is left at the secure default everywhere and is
never disabled. A scan of the entire repository (app, lib, config, scripts, firmware,
tools, contracts) finds no VERIFY_NONE, verify: false, ssl_verify false,
verify_peer: false, "insecure", curl -k, or --no-check-certificate — nothing
overrides certificate verification.

• Outbound HTTPS uses standard clients at their defaults: the eth gem (Eth::Client,
  over Net::HTTP) for chain RPC and HTTPX for other services, both of which verify the
  server certificate by default (OpenSSL VERIFY_PEER) for https URLs. No custom
  SSLContext or verify_mode is set anywhere.
• Subresources in the web UI (importmap JS, Leaflet) are loaded over HTTPS and a
  Content-Security-Policy restricts their origins, so the browser performs normal
  certificate verification on them as well.
  https://github.com/Alexey-Lukin/silken_net/blob/main/config/initializers/content_security_policy.rb

Met. Private information is only sent over a connection whose certificate has been
verified:

• Server side: session cookies are flagged secure in production
  (config/initializers/session_store.rb: secure: Rails.env.production?, plus httponly
  and same_site: :lax), and config.force_ssl = true with HSTS (1 year,
  includeSubdomains, preload) guarantees the browser transmits the cookie only over
  HTTPS after verifying the server certificate, and never over plain HTTP (HSTS
  preload blocks downgrade).
• Client side: outbound requests carrying private headers (e.g. Authorization: Bearer
  API keys to dClimate/Filecoin, the Chainlink HMAC signature, M2M tokens) are sent
  over HTTPS via Net::HTTP/HTTPX with default certificate verification (OpenSSL
  VERIFY_PEER), which is never disabled anywhere in the codebase — the TLS handshake,
  including certificate validation, completes before the request headers are sent.
  https://github.com/Alexey-Lukin/silken_net/blob/main/config/initializers/session_store.rb
  https://github.com/Alexey-Lukin/silken_net/blob/main/config/environments/production.rb

Met. The project's public generated deliverable — the production container image
mirrored to GHCR (ghcr.io/alexey-lukin/silken_net) for widespread use (e.g. Akash
providers) — is cryptographically signed with a Sigstore-keyless SLSA
build-provenance attestation produced by actions/attest-build-provenance in
.github/workflows/mirror-ghcr.yml (GitHub Actions OIDC → Fulcio, logged in the
public Rekor transparency log) and pushed to the registry alongside the image. The
signing key is ephemeral (Fulcio-issued per build, never stored) — it is not on the
distribution site. The documented user verification process is in SECURITY.md
("Verifying release artifacts"): gh attestation verify
oci://ghcr.io/alexey-lukin/silken_net:<tag> --owner Alexey-Lukin. Source-tree
commit/tag signing is tracked separately (OPS.10).
https://github.com/Alexey-Lukin/silken_net/blob/main/.github/workflows/mirror-ghcr.yml
https://github.com/Alexey-Lukin/silken_net/blob/main/SECURITY.md

Met. The project's public generated deliverable — the production container image
mirrored to GHCR (ghcr.io/alexey-lukin/silken_net) for widespread use (e.g. Akash
providers) — is cryptographically signed with a Sigstore-keyless SLSA
build-provenance attestation produced by actions/attest-build-provenance in
.github/workflows/mirror-ghcr.yml (GitHub Actions OIDC → Fulcio, logged in the
public Rekor transparency log) and pushed to the registry alongside the image. The
signing key is ephemeral (Fulcio-issued per build, never stored) — it is not on the
distribution site. The documented user verification process is in SECURITY.md
("Verifying release artifacts"): gh attestation verify
oci://ghcr.io/alexey-lukin/silken_net:<tag> --owner Alexey-Lukin. Source-tree
commit/tag signing is tracked separately (OPS.10).
https://github.com/Alexey-Lukin/silken_net/blob/main/.github/workflows/mirror-ghcr.yml
https://github.com/Alexey-Lukin/silken_net/blob/main/SECURITY.md

Met — input is validated with an allowlist approach at every untrusted boundary:

• HTTP/API: Rails strong parameters (params.require(...).permit(...)) accept only an
  explicit allowlist of attributes; everything else is dropped. Models add
  allowlist-style validations enforced before persistence — numbers are range-checked
  (actuator_command duration numericality: { greater_than: 0, less_than_or_equal_to:
  3600 }, ai_insight scores { in: 0.0..100.0 } / { in: 0.0..1.0 }), values are
  constrained to allowlists (inclusion: { in: %w[evm solana celo] }, inclusion: { in:
  HARDWARE_TYPES }), and strings are checked for syntax/length (format: { with: ... }
  for hex addresses/bytecode, length: { maximum: ... }). Invalid records are rejected.
• Untrusted IoT telemetry (binary LoRa/CoAP): TelemetryUnpackerService validates the
  decoded packet (valid_sensor_data?) before processing and rejects malformed or
  out-of-range frames; it also enforces a SEC.10 panic-replay counter and CCM
  key-size checks.
• On-chain (Solidity): every external function guards its inputs with require(...) —
  non-zero addresses, array-length equality, batch-size bounds (<= MAX_BATCH_SIZE),
  positive amounts — reverting on invalid input.
• Oracle callbacks (Chainlink) are authenticated by HMAC before their payload is
  accepted.

https://github.com/Alexey-Lukin/silken_net/blob/main/app/models/blockchain_transaction.rb
https://github.com/Alexey-Lukin/silken_net/blob/main/app/services/telemetry_unpacker_service.rb

Met (SHOULD). Hardening mechanisms are applied on both the web and firmware sides.

Web (Rails — the network-facing attack surface):

• A strict Content-Security-Policy (config/initializers/content_security_policy.rb):
  default_src/base_uri/form_action 'self', frame_ancestors 'none', frame_src 'none',
  object_src 'none', and a per-request nonce for inline scripts (no 'unsafe-inline'
  on script-src).
• A full security-header set (config/initializers/security_headers.rb): X-Frame-Options
  DENY, X-Content-Type-Options nosniff, Referrer-Policy strict-origin-when-cross-origin,
  Cross-Origin-Opener-Policy / Cross-Origin-Resource-Policy same-origin, a restrictive
  Permissions-Policy, and X-XSS-Protection 0 (disables the legacy exploitable filter).
• HSTS with preload (1 year, includeSubdomains) + config.force_ssl; session cookies are
  httponly + secure + same_site:lax.

Firmware C (a memory-unsafe language → compiler hardening):

• The entire host test suite is compiled and executed under AddressSanitizer +
  UndefinedBehaviorSanitizer (-fsanitize=address,undefined -fno-sanitize-recover=all) on
  every CI run, so undefined behavior, buffer overflows and use-after-free abort the build
  before release — the criterion's "compiler flags to eliminate undefined behavior"
  example. The host build also honors -fstack-protector-strong / -D_FORTIFY_SOURCE / RELRO.

https://github.com/Alexey-Lukin/silken_net/blob/main/config/initializers/content_security_policy.rb
https://github.com/Alexey-Lukin/silken_net/blob/main/firmware/test/Makefile

Met. The project provides a structured security assurance case at
docs/SECURITY_ASSURANCE_CASE.md. It states the top-level security claims; a threat
model (assets, threat actors, and attack surfaces across the Soldier->Queen->Rails->chain
pipeline); an explicit identification of every trust boundary and the guard that
enforces it; an argument that Saltzer-Schroeder secure-design principles are applied,
with code anchors; and an OWASP Top 10 (2021) table showing how each common
implementation weakness is countered - followed by an honest residual-risk and
assumptions section. Each claim is backed by the test / SAST / SCA evidence listed in
the document.
https://github.com/Alexey-Lukin/silken_net/blob/main/docs/SECURITY_ASSURANCE_CASE.md

The static analyzers used are specifically designed to find common vulnerabilities: Brakeman targets Rails vulnerability classes (SQL injection, XSS, mass assignment, CSRF), CodeQL runs security queries mapped to CWE, Slither has detectors for common Solidity vulnerabilities (reentrancy, access control, arithmetic), and cppcheck flags memory/defect issues in C. CodeQL runs security queries mapped to CWE (GitHub default setup, across all six
languages — Ruby, C/C++, C#, JS/TS, Python, Actions).
https://github.com/Alexey-Lukin/silken_net/blob/main/.github/workflows/solidity_audit.yml

The project ships memory-unsafe C firmware (STM32 Soldier/Queen + the One-Home libraries in firmware/common/), so N/A does not apply. Every host-based unit test (firmware/test/, ~22 binaries covering the untrusted-input parsers — AT-command/CoAP PDU tokenizer, circular-DMA UART RX ring, flash KV/ring/OTA journals with power-cut fault injection — plus the crypto, DSP and TinyML paths) is compiled and executed under AddressSanitizer + UndefinedBehaviorSanitizer on every CI run, via make -C firmware/test asan in the firmware_test job of .github/workflows/ci.yml (gating through the ci-ok aggregate required check). Flags: -fsanitize=address,undefined -fno-sanitize-recover=all -fno-omit-frame-pointer -g -O1. AddressSanitizer detects heap/stack/global buffer overflows (the explicit "buffer overwrite" concern), use-after-free/return and double-free; UndefinedBehaviorSanitizer detects signed-integer overflow, out-of-bounds shifts, misaligned/null pointer dereferences and invalid enum/bool loads; halt_on_error=1 makes the first finding fail the build. LeakSanitizer is intentionally disabled (detect_leaks=0) because the only allocations outliving main() are OpenSSL's one-time global init in two tests, which the short-lived test processes deliberately do not tear down — a "still reachable" false positive, not a memory-safety defect. This dynamic lane complements the existing static analysis (cppcheck firmware_lint, CodeQL c-cpp, and -Wall -Wextra -Wpedantic).
https://github.com/Alexey-Lukin/silken_net/blob/main/firmware/test/Makefile