feat: Engraphis 1.7 release — full changeset - #186
Merged
Conversation
The 1500-char prose context was actively harmful on Qwen 3.7 Flash (bench_v1/V12: -0.067 substring score vs no-memory baseline) and only marginally helpful on larger models. Smaller models get distracted by dense technical context and lose focus on the actual question. Switch the default to a 300-char "[n] first-sentence" compression of the upstream MCP context. The compression prefers the explicit "[n]" markers that engraphis_session already emits; falls back to period-splitting for unnumbered prose. Always fails open: returns the original context unchanged on any parse failure. Users who want the original 1500-char prose can opt in with ENGRAPHIS_HOOK_FORMAT=prose. Tests cover both modes, the env override, the fail-open behaviour, and the per-fact budget cap. 19/19 tests pass (10 existing + 9 new). ruff check clean.
… slider The classic dashboard shipped a `data-graph-scale="10"` attribute on the Line-width slider that multiplied the displayed value by 10 (min="2" max="45") but never divided it back before reaching the engine. Picking the `communities` preset put `linkw=7.2` into the engine, which renders edges at `0.82 * 7.2 ~= 5.9` pixels - ten times thicker than the `GRAPH_PRESETS.communities.linkw = 0.72` the engine was designed for. The Ledger already ships the raw range (min="0.1" max="2" step="0.01") and a 0.72 default, so the classic dashboard now matches it. Removing the `data-graph-scale` keeps the round-trip lossless: the slider value the user sees is the value the engine receives. A regression test parses `GRAPH_PRESETS` from `classic_assets/dashboard.js`, asserts each preset's `linkw` is reachable on the slider, and asserts the `data-graph-scale` attribute never re-appears on the `linkw` input. Cache busters on the served CSS/JS bumped to v=20260902-classic-linkw-slider-1. Co-authored-by: CommandCodeBot <noreply@commandcode.ai>
Version 1.7 across pyproject.toml, __init__.py, commercial_manifest.json, marketplace.json, plugin.json, and CHANGELOG.md. Core changes: semantic_confidence_calibration default True, ProfileConfig arm_candidate_k_default=200, graph_seed_fallback opt-in, store id-kind whitelist with allow_unsafe escape hatch, tightened valid_to<=valid_from with quarantine-marker exception, scored LIKE fallback, audit hash-chain prev_hash + AUDIT_ACTORS, per-version schema_migrations, secure_erase_impact, _CAPABILITY_INDEX TTL/LRU, MCP confirmed attestation for destructive tools, CLI --workspace alias. Includes 7 new release test files, grounded distractor dataset, and .gitignore fix for dashboard_slider_test.png. All 12 CI/eval gates verified: ruff clean, pyright 0 errors, pytest all pass, eval harness/ablation/reinforcement/adversarial/ grounded/code_arm all PASS.
|
You have reached your Codex usage limits for code reviews. You can see your limits in the Codex usage dashboard. |
CI's `npm audit --omit=dev` gate fails on fast-uri 3.1.5 (GHSA-5jgf-p345-68v8, GHSA-f65p-4m7j-42xc, GHSA-fph4-wmhf-6fwf, GHSA-jqff-g426-hqxp — high) and qs 6.15.3 (moderate). `npm audit fix --package-lock-only` moves fast-uri to 3.1.7 and qs to 6.16.0 with no API change; verify+integration suites pass locally (21/21) and audit now reports 0 vulnerabilities.
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.
Engraphis 1.7 release changeset
77 files, +5680/−3427. Local
mainfast-forwards ontoorigin/mainvia this PR.Release slices
pyproject.toml,engraphis/__init__.py(_SOURCE_VERSION),engraphis/commercial_manifest.json, plus.claude-plugin/plugin.json+marketplace.json(hashes re-issued).CHANGELOG.mdretitled[1.7] - 2026-09-03.semantic_confidence_calibrationdefault-on (kills singleton-cosine false 1.0); profile-drivenrerank_blend/consolidation_bonus/arm_candidate_k_default=200; explicit arm-cap override clamps both pages; always-on telemetry (candidate_k_used,arm_counts,rerank_changed); graph-seed fallback is opt-in (ProfileConfig.graph_seed_fallback, default False) with mention+link gate — published retrieval evidence stays byte-stable.ids.new_idkind whitelist (allow_unsafeescape) + prefix asserts; explicitvalid_to<=valid_fromrejected with quarantine-marker exception; scored LIKE fallback; audit hash-chainprev_hash+AUDIT_ACTORS; per-versionschema_migrationsrows;Store.secure_erase_impact();migrate_to_v2dim-histogram +--resume;require_exact_backendsforced inENGRAPHIS_ENV=prod+backend_health().forget()= retire-with-history copy;secure_erase(confirmed=True)gate + impact report + chained audit; peer tombstones quarantine-first (tombstones_held, marker recorded, bytes retained for review;erase_device_allowlist); MCPconfirmedattestation on destructive tools;_CAPABILITY_INDEXTTL/LRU; SYNC.md rotation runbook.v2_api.py+inspector/app.pythreadconfirmed.--workspacealias +delete-namespaceflag-or-positional + defaults matrix;$PORTprecedence docs; SKILL Setup init/DB-pin; MCP_TOOLS smart schemas; offline first-run callouts.Included concurrent work (verified green pre-commit)
dashboard_app.py), galaxy damping/rotation balance (engraphis-graph.js), ledger CSRF retry, asset cache-bust20260903-rotation-balance-1+ e2e spec bumps.Tests/evals/CI
eval/grounded.pydistractor + quarantined-probe;release.yml--website-root(dir-conditional).offline-fixtures-v1.json, sha8a74e9f4…propagated to README/BENCHMARKS/2 SVGs): grounded eval now 5 grounded + 6 abstain incl. 1 quarantine probe, 71.1% context reduction unchanged.Verification
ruffclean;pyright0 errors; commercial-manifest OK; CSP/asset gate OK.Deferred (unchanged, per plan): v1 deletion, classic_assets retirement, classic_mcp/
_pull_legacyremoval, SSO/SLA, GPU image.Tagging note:
release.ymltag==version guard passes once merged; tagv1.7post-merge. Pi package stays on its ownpi-v*line.