Skip to content

Releases

v0.1.17

Choose a tag to compare

@github-actions github-actions released this 28 Oct 12:55
v0.1.17
e8b37a7

KRY 0.1.17 includes the runtime and verifier-tooling fixes merged after 0.1.16.

Before upgrading, read the migration guide. kry_action.record() now hashes all raw content, including 64-character hex strings. Callers passing existing commitments must use precommitted=True with valid lowercase SHA-256 digests for every supplied content field.

  • Concurrent settlements sharing one ReceiverLedger now preserve both receiver credits. Separate instances remain independent; receiver storage is still in-memory and settlement is not atomic across ledgers.
  • Nitro and SNP adapters require supplied measurement IDs to match the verified payload.
  • Release proofs bind to the tested commit, tree and tag, and join the published checksums and build provenance. Package-input validation rejects unexpected untracked or ignored build inputs.
  • OpenRouter exports stop on unresolved rate limits, and the JavaScript batch verifier rejects invalid explicit multiplier files.

See the complete changelog for all fixes. Existing receipt and registry formats are unchanged. Usage arithmetic and declared evidence tiers do not establish provider payment, settlement or proven cash savings.

Install the distribution with python -m pip install --upgrade kry-attest==0.1.17; the import remains kry. Stop shared writers and back up their complete records before upgrading them together. Publishing this release does not upgrade running installations.

The attached release-verify.json is the full hosted gate proof. Its digest is included in checksums.txt and it is covered by build provenance alongside the wheel, source distribution and source archive.

v0.1.16

Choose a tag to compare

@github-actions github-actions released this 25 Oct 00:25
v0.1.16
f151b20

What's Changed

Other Changes

Full Changelog: v0.1.15...v0.1.16

v0.1.14

Choose a tag to compare

@github-actions github-actions released this 24 Oct 04:16
v0.1.14
844e70f

KRY v0.1.14 fixes ledger recovery and settlement retry failures, and includes the fixes from the unpublished v0.1.13 tag.

Accounting and recovery

  • Unknown debit outcomes keep the settlement grant consumed. Partial-debit adjustment preserves that protection throughout its update.
  • Interrupted public operations and cache refreshes cannot leave replayable changes behind. Capacity checks read durable state without flushing pending changes.
  • Recovery acknowledgement refuses negative restored balances, preserves debit history and reports success only when accounting is available.
  • Included v0.1.13 changes protect concurrent accounting, uncertain writes, missing or corrupt ledger history, precision boundaries and retained holdout history.

Other included fixes

  • Explicit model and dated-snapshot pricing replaces substring matching. Unknown or unsupported models earn no new credit; historical receipt verdicts remain unchanged.
  • Optional SNP minting requires supported AMD identity and explicit image/TCB policy. Custom-root inspection cannot authorize minting. Optional SNP/PQC inputs are bounded, and dry-run results distinguish inspection from hardware policy.
  • Request-ID reconciliation and TEE/TLSN identifiers preserve their evidence and duplicate-credit boundaries.

Upgrade guidance

Back up data and upgrade all writers sharing a directory together. Legacy or interrupted accounting may require explicit recovery using independently established debit records. Do not reconstruct missing spending as zero. Read the recovery guide before resuming a blocked ledger.

The signed v0.1.13 tag remains in history; no v0.1.13 GitHub assets or PyPI distributions were published.

Verification

The signed-tag release gate passed all 12 checks. Its initial suite and each of ten reproduction rounds recorded 2,901 passed / 18 optional skips; all ten lab, zero-loss concurrency and lease demonstrations passed.

  • PR and main checks passed across Python 3.11–3.14; Windows recorded 2,877 passed / 42 platform or optional skips.
  • Dedicated native cryptography jobs on Python 3.11 and 3.13 each recorded 238 passed with zero skips. LiteLLM integration recorded 8 passed with one existing upstream Pydantic warning.
  • The JavaScript corpus passed all 402 cases. Vector regeneration preserved every corpus file byte for byte.
  • This follow-up adds 100 regression cases without changing existing tests or historical vector verdicts. Package content and provenance are checked separately from the test results.

The three distribution/source archives carry GitHub build-provenance attestations. The separate release-verifier report records the workflow gate; it is not itself an attested archive. These tests establish the exercised implementation behavior, not provider payment settlement, savings veracity or independent adoption.

Full changelog

v0.1.12

Choose a tag to compare

@github-actions github-actions released this 23 Oct 06:26
v0.1.12
8d236bf

KRY 0.1.12 fixes repeated credit in the optional TEE and SEV-SNP measurement paths.

A measurement ID containing a slash or whitespace could fail to match its existing receipt. A second presentation with different evidence bytes could then mint the same saving again. Both entry points now require a nonempty string fully matching [^/\s]+, including operator overrides and dry runs. Direct promote_to_tee rejects invalid IDs with ValueError; it does not return the same result as “no prior receipt.”

A related lookup bug affected otherwise valid IDs containing JSON-escaped characters. Receipt matching now decodes JSON before comparing complete IDs, preserving quotes, backslashes and non-ASCII IDs. Invalid CLI input exits nonzero.

Validation includes 90 new cases: 82 exposed failures on 0.1.11 and eight retained passing controls. The full integration suite passed 2,573 tests with 14 optional skips and one existing dependency warning. The separate native-crypto suite passed 173 tests with no skips; the JavaScript corpus passed 402 cases. The signed tag's full release proof is included as a release asset.

Existing receipt formats and vector verdicts are unchanged. These changes prevent new duplicate credits through the covered paths; they do not rewrite historical receipts or automatically repair earlier balances. Operators who used affected IDs should review the retained ledger before relying on its totals.

The deferred v0.2.0 decisions remain unchanged: usd_equivalent verification, the pairwise multiplier set, sequence-based anchor lookup and the import name. Nitro/SNP remain optional. The package contains the core library and public CLIs; optional evidence scripts require the checkout or full source archive. Passing these checks does not establish the truth of a measurement, provider charges, cash savings or settlement.

v0.1.11

Choose a tag to compare

@github-actions github-actions released this 23 Oct 05:12
v0.1.11
e809ae2

KRY 0.1.11 hardens receipt generation, evidence reconciliation and optional verification paths. It also adds opt-in tools for examining supplied usage records while keeping source authenticity, actual billing and payment settlement separate.

Changes include:

  • Correct rounded attestation totals and receipt sequence numbers, reject ambiguous settlement grant IDs, and harden optional TEE and post-quantum input handling.
  • Preserve imported logs as self-reported estimates with source floor zero. Tighten per-request research and artifact grading to require exact unique request IDs and successful reconciliation.
  • Make LiteLLM callback deduplication durable across updated writers sharing the same intact receipt log; reject contradictory usage counts and report uncertain persistence.
  • Add optional evidence assessment, row commitments, Anthropic usage estimates/projection and private LiteLLM observations in the repository checkout.
  • Require hash-locked CI dependencies and execute optional crypto tests with real ML-DSA operations. Keep Nitro hardware demonstrations optional.

Upgrade with python3 -m pip install --upgrade kry-attest==0.1.11. The PyPI distribution contains the core library, kry-try and kry-verify; additional scripts and optional proof-of-concept tools require the repository checkout or full GitHub source archive.

Migration: attestation formats and historical vector verdicts are unchanged. Re-evaluate previous research/artifact packets under the stricter acceptance rules, regenerate T1 manifests when needed, and supply matching unique request IDs for required per-request records. Token-only matches remain diagnostic. Request-ID digests support comparison; they do not authenticate or anonymize records.

Verification establishes receipt integrity and supported benchmark arithmetic. It does not establish source truth, actual account charges, cash savings, ownership by default or payment settlement. The historical 52-record research bundle remains unchanged and represents zero-dollar self-traffic. See the changelog and claims boundary for the full scope.

Validation: the final local integration suite passed 2,483 tests with 14 optional-dependency skips and one existing dependency warning. The isolated release suite passed 2,475 tests with 17 optional skips in each of ten reproduction rounds; all 12 release checks passed. Optional native crypto passed 173 tests with zero skips, and the JavaScript corpus passed 402 with zero failures. Fresh wheel and source installs passed receipt verification and tamper rejection. These are software checks, not independent customer validation.

v0.1.10

Choose a tag to compare

@github-actions github-actions released this 18 Oct 10:22
v0.1.10
b5f89af

What's Changed

Other Changes

Full Changelog: v0.1.9...v0.1.10

v0.1.9

Choose a tag to compare

@github-actions github-actions released this 18 Oct 03:42
v0.1.9
8e277a5

What's Changed

Other Changes

  • fix(scripts): leave committed evidence alone unless asked, answer --help, fail the demo on INVALID by @thequantumfalcon in #98
  • docs: make every public claim match what the verifiers check by @thequantumfalcon in #99
  • docs: state the research_grade count its committed evidence holds by @thequantumfalcon in #100
  • fix(verify): reject a forged JS action witness, fail closed on hostile input, bound JSON parsing by @thequantumfalcon in #101
  • release: v0.1.9 — reject a forged JS action witness; verifiers fail closed and bound JSON parsing by @thequantumfalcon in #102

Full Changelog: v0.1.8...v0.1.9

v0.1.8

Choose a tag to compare

@github-actions github-actions released this 17 Oct 17:18
v0.1.8
d365ec5

What's Changed

Other Changes

  • fix(verify): close the null by_tier floor bypass and JS prototype-key forgeries by @thequantumfalcon in #96
  • release: v0.1.8 — close the null by_tier floor bypass and JS prototype-key forgeries by @thequantumfalcon in #97

Full Changelog: v0.1.7...v0.1.8

v0.1.7

Choose a tag to compare

@github-actions github-actions released this 17 Oct 05:37
v0.1.7
9d223a8

What's Changed

Other Changes

  • ci(attribution): check commit messages and identities, not only tracked files by @thequantumfalcon in #87
  • fix(artifact): explain FAIL gate rows on a zero-exit internal artifact by @thequantumfalcon in #88
  • docs(readme): quickstart notes on publishing the anchor and optional crypto warnings by @thequantumfalcon in #89
  • docs(contributing): separate AI disclosure from AI attribution by @thequantumfalcon in #91
  • ci(deps): bump the actions group across 1 directory with 7 updates by @dependabot[bot] in #90
  • deps: bump ruff from 0.16.7 to 0.16.8 in the python group by @dependabot[bot] in #84
  • fix(quickstart): run on a stock Python and call the sample data sample by @thequantumfalcon in #93
  • release: v0.1.7 — quickstart on a stock Python, sample wording, anchor line by @thequantumfalcon in #94

Full Changelog: v0.1.6...v0.1.7

v0.1.6

Choose a tag to compare

@github-actions github-actions released this 19 Sep 17:31
v0.1.6
fd63db4

Compatibility: verification is stricter

Some documents accepted by 0.1.5 are intentionally rejected by 0.1.6. Verification now rejects overflowing economic values, malformed JSON, nonnumeric trust summaries, wrong-profile documents, noncanonical raw-literal hashes and incorrectly rounded totals. Producers must correct and reseal invalid documents; the verifier does not repair them.

Equivalent float spellings normalize consistently across Python and JavaScript, and correctly rounded large finite receipts are accepted. Existing vector bytes and verdicts are preserved, with no hash-version change.

The stranger CLI reports VERDICT: PARSE_ERROR for malformed or overflowing JSON and retains exit code 1. Semantic failures and unreadable files still report INVALID. Consumers matching output text must recognize all three outcomes.

Read the compatibility decision before upgrading.

Changes

  • Close the token-detail privacy bypass before public bundle inputs are copied.
  • Preserve top-level billing tiers in savings reports; exclude conflicting or nonstandard tiers. Short-context pricing is a stated basis, not a lower bound on net cache savings.
  • Add OpenAI prompt-cache reporting, refresh output-price entries, exclude batch traffic, and support LiteLLM routing-savings receipts. Provider prompt-cache estimates remain report-only and self-reported.
  • Add a hardened, pinned LiteLLM integration CI job and receipt-level regressions, including genuine rounding ties, large values and Windows console encodings.
  • Expand the conformance corpus to 65 vectors and 75 checks, adding 27 vectors without modifying pre-existing vector documents. Require both verifiers to accept every fuzz base before mutation.
  • Correct unsupported agreement claims and update the spec and live verifier documentation.

Evidence and limits

The tagged release workflow verifies the signing key and runs the full release gate, including ten reproducibility rounds, before publication. Release assets include the gate report, SHA-256 checksums and build provenance. The wheel and source distribution uploaded to PyPI are checked against those same release checksums.

Passing the tests and sampled differential fuzz is bounded evidence, not proof of agreement on every input. The stdlib-only core and the integrity/magnitude/veracity distinction remain unchanged; no new claim of externally verified real-world savings is made.

Full changelog