Skip to content

sheawinkler/ContextLattice

v3.20.0 Feature

This release adds 3 notable features for engineering teams evaluating rollout.

✓ No known CVEs patched
Read the diff → Tool health → What is this tool? →

✓ No known CVEs patched in this version

Topics

agent-framework agent-orchestration ai-infra ai-interface context-engineering context-management
+4 more
go long-horizon-agents long-horizon-intelligence rust

Summary

AI summary

New local Utility Ledger records outcome claims, verification events, and bounded analytics with explicit error handling.

Changes in this release

Feature Medium

Adds a new local Utility Ledger joining outcomes, token-impact rows, telemetry, and verification events.

Adds a new local Utility Ledger joining outcomes, token-impact rows, telemetry, and verification events.

Source: llm_adapter@2026-07-18

Confidence: high

Feature Medium

Introduces `contextlattice utility record` command to log outcome claims.

Introduces `contextlattice utility record` command to log outcome claims.

Source: llm_adapter@2026-07-18

Confidence: high

Feature Medium

Introduces `contextlattice utility verify` command for verifiers to record witnessed results.

Introduces `contextlattice utility verify` command for verifiers to record witnessed results.

Source: llm_adapter@2026-07-18

Confidence: high

Feature Medium

Makes observed verified utility yield available without requiring a control experiment.

Makes observed verified utility yield available without requiring a control experiment.

Source: llm_adapter@2026-07-18

Confidence: high

Feature Medium

Adds bounded cohort analytics, task‑class economics, confidence intervals, and advisory policy gates for paid runtime artifacts.

Adds bounded cohort analytics, task‑class economics, confidence intervals, and advisory policy gates for paid runtime artifacts.

Source: llm_adapter@2026-07-18

Confidence: high

Feature Medium

Ensures the ledger is fsync‑backed, restart‑safe, and independently disableable without mutating authoritative ledgers.

Ensures the ledger is fsync‑backed, restart‑safe, and independently disableable without mutating authoritative ledgers.

Source: llm_adapter@2026-07-18

Confidence: high

Feature Medium

Enforces owner‑only lifetime locking so only one writer can access a ledger path, causing subsequent gateway processes to fail closed.

Enforces owner‑only lifetime locking so only one writer can access a ledger path, causing subsequent gateway processes to fail closed.

Source: llm_adapter@2026-07-18

Confidence: high

Feature Medium

Keeps verification receipts authoritative even when reconciliation storage fails; responses report partial failure and preserve `event_recorded:true`.

Keeps verification receipts authoritative even when reconciliation storage fails; responses report partial failure and preserve `event_recorded:true`.

Source: llm_adapter@2026-07-18

Confidence: high

Feature Medium

Provides commands `contextlattice utility analytics` and `contextlattice utility gate` for paid operators.

Provides commands `contextlattice utility analytics` and `contextlattice utility gate` for paid operators.

Source: llm_adapter@2026-07-18

Confidence: high

Feature Medium

Allows disabling the local ledger via `GO_UTILITY_LEDGER_ENABLED` without altering authoritative evidence.

Allows disabling the local ledger via `GO_UTILITY_LEDGER_ENABLED` without altering authoritative evidence.

Source: llm_adapter@2026-07-18

Confidence: high

Full changelog

ContextLattice v3.20.0: Intelligence With Receipts

Context is not valuable because it is large. It is valuable when it helps an agent produce independently verified useful work. ContextLattice v3.20.0 makes that relationship measurable without pretending correlation is causation.

The CLI remains the primary interface. HTTP remains the integration fallback.

What changed

  • The new local Utility Ledger joins Context Pack outcomes, exact token-impact rows, quality telemetry, and session verification events by exact project, agent, session, and sample identity.
  • contextlattice utility record records an outcome claim; a distinct verifier records the witnessed result with contextlattice utility verify.
  • Exact model-visible ContextLattice tokens remain separate from exact wire tokens and observed provider totals. Estimated or missing denominators never enter exact claims.
  • Observed verified utility yield is available without a control. Causal gain requires a leakage-free matched control with identical task, assignment, model, runner, harness, reconstruction identity, exact model-visible token count, and tokenizer encoding.
  • Negative gains, exclusions, failures, cost, latency, and tool calls remain visible instead of being optimized out of the story.
  • Paid runtime artifacts add bounded cohort analytics, task-class economics, conservative confidence intervals, and advisory policy gates.
  • The ledger is bounded, fsync-backed, restart-safe, and independently disableable without mutating authoritative outcome, token-impact, quality, or session ledgers. A configured persistence failure returns a visible 503 instead of acknowledging a memory-only claim, latches writes closed until restart, and binds uncertain bytes to the original source claim during recovery.
  • Owner-only lifetime locking enforces one writer per ledger path, so a second gateway process fails its Utility Ledger closed instead of dropping, conflicting with, or serving stale observations.
  • Verification receipts remain authoritative even when reconciliation storage fails: HTTP and CLI responses report partial failure explicitly, preserve event_recorded:true, and never invite a blind event replay.

Quick start

EVIDENCE_DIGEST=sha256:aaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaa

contextlattice utility record \
  --agent codex \
  --agent-id reporter_agent \
  --project contextlattice \
  --session-id session-123 \
  --context-pack-quality-sample-id sample-123 \
  --outcome-id outcome-123 \
  --utility-value 1 \
  --utility-unit accepted_result \
  --verification-event-id verification-123 \
  --verification-evidence-digest "$EVIDENCE_DIGEST" \
  --verification-passed true \
  --verifier-kind human_review \
  --verifier-id reviewer_agent \
  --pretty

contextlattice utility verify \
  --agent reviewer \
  --agent-id reviewer_agent \
  --project contextlattice \
  --session-id session-123 \
  --sample-id sample-123 \
  --outcome-id outcome-123 \
  --verification-event-id verification-123 \
  --utility-value 1 \
  --utility-unit accepted_result \
  --verification-evidence-digest "$EVIDENCE_DIGEST" \
  --verification-passed true \
  --verifier-kind human_review \
  --verifier-id reviewer_agent \
  --pretty

contextlattice utility status --project contextlattice --pretty

Paid analytics and advisory evaluation:

contextlattice utility analytics --project contextlattice --pretty
contextlattice utility gate --project contextlattice --pretty

Claim boundary

The verifier identity is locally attested, not a remote identity-provider guarantee. The runtime structurally separates the reporting agent from the verification event actor and rejects self-verification, but a malicious local caller with machine access can still lie about its identity. ContextLattice records that trust boundary rather than hiding it.

Rollback

Disable GO_UTILITY_LEDGER_ENABLED to stop recording the derived local ledger without changing authoritative evidence. Paid operators can remove the Utility Analytics routes from the protected path set and recreate only gateway-go; the free local ledger remains independent.

Weekly OSS security release digest.

The CVE patches and breaking changes that affected production tools this week. One email, every Sunday.

No spam, unsubscribe anytime.

Share this release

Track sheawinkler/ContextLattice

Get notified when new releases ship.

Sign up free

About sheawinkler/ContextLattice

Private-by-default memory and context layer for agents with Go/Rust runtime, staged retrieval across fused data backends, and long-horizon context continuity.

All releases →

Related context

Earlier breaking changes

  • v3.17.3 Agent guidance now mandates the `/agents/tasks` route family.
  • v3.17.3 Task worker now checks approval before any execution steps.

Beta — feedback welcome: [email protected]