Based on the detailed changelog and technical specification text you provided, here is a structured summary of the release notes, architectural changes, and protocol updates for the ARS (Academic Research System) suite, organized by version.
Note: The suite release is tagged v3.6.8 to maintain CHANGELOG monotonicity.
- Schema 13.1 (
shared/sprint_contract.schema.json):- Extends Schema 13 with two new
modeenum values:writer_fullandevaluator_full. - Introduces optional top-level fields:
pre_commitment_artifacts(writer-only) anddisagreement_handling(evaluator-only). - Adds 12
allOfbranches enforcing reviewer/writer/evaluator conditional gates. - Zero-touch Promise: Existing reviewer contracts remain byte-equivalent under Schema 13.1.
- Extends Schema 13 with two new
-
Shipped Templates:
- Writer:
shared/contracts/writer/full.json(Templates D1–D7, F1, F4, F2, F3, F0). - Evaluator:
shared/contracts/evaluator/full.json(Templates D1–D5, F1, F2, F3, F6, F4, F5, F0). - Promoted from design-time artifacts to live shipped status atomically with the upgrade.
- Writer:
-
Two-Phase Split:
- Phase 4: Split into 4a (writer paper-blind pre-commitment) and 4b (paper-visible drafting + self-scoring).
- Phase 6: Split into 6a (evaluator paper-blind pre-commitment) and 6b (paper-visible scoring + decision).
- Phase-numbered outputs (
<phase4a_output>,<phase6a_output>) mirror the v3.6.2 reviewer pattern.
-
Lint Count Summary:
- Writer: 3+4 | Evaluator: 5+5 | Reviewer: 5+6 (Reviewer remains zero-touch).
-
SKILL.md Updates: Added a
## v3.6.6 Generator-Evaluator Contract Protocolblock (101 lines). - Agent Files: Added 47 lines to
draft_writer_agent.mdand 57 lines topeer_reviewer_agent.md. - Validator Extensions:
scripts/check_sprint_contract.pynow audits SC-* mode-gating (SC-5, SC-11 reviewer-only; SC-9 extended).- Test Suite: 17 new tests brought unit-test count from 54 to 71 (including schema-branch negative tests and mode-gating tests).
- CI/Manifest:
scripts/check_v3_6_6_ab_manifest.pyenforces manifest schema and git-tracked invariants..github/workflows/spec-consistency.ymlnow iterates writer/evaluator templates alongside reviewer loops.
Date: 2026-04-30
* Hardening: Protected 13 of 18 documented hallucination/drift patterns across synthesis_agent, research_architect_agent (survey-designer), and report_compiler_agent (abstract-only).
* PATTERN PROTECTION Block: Added to all agent prompts.
* Reference Files: Added 4 new operational contract references (irb_terminology_glossary.md, psychometric_terminology_glossary.md, protected_hedging_phrases.md, word_count_conventions.md).
* Audit Template: New shared/templates/codex_audit_multifile_template.md with mandatory Section 4(f) checks.
* Static Lint: scripts/check_v3_6_7_pattern_protection.py and mutation suite (29 tests).
* Codex Review: 7 rounds of review reached SHIP-OK with zero P1+P2 findings.
Date: 2026-04-27
* Consumer Wiring: Activated deep-research/agents/bibliography_agent.md and academic-paper/agents/literature_strategist_agent.md.
* Protocol: Follows five-step "corpus-first, search-fills-gap" flow with four Iron Rules.
* Pre-Screened Reproducibility: Reports enumerate included/excluded/skipped entries with F3 zero-hit notes.
* Fallback: Parse failures trigger external-DB-only flow with [CORPUS PARSE FAILURE] surface.
* Citation Compliance: Deferred to v3.6.6+.
Date: 2026-04-25
* Schema 9 Update: Added optional literature_corpus[] input port for user-owned literature.
* Entry Format: CSL-JSON compliant (authors, year, title, source_pointer, private optional abstract/user_notes).
* Adapters: Provided 3 Python adapters (folder_scan.py, zotero.py, obsidian.py) as starting points.
* Rejection Log: Contract added at shared/contracts/passport/rejection_log.schema.json.
* CI Gates: Schema validation and adapter sync checks added.
Date: 2026-04-23
* Feature: Opt-in reset boundary (ARS_PASSPORT_RESET=1).
* Behavior:
* ON: Promotes every FULL checkpoint to a context-reset boundary; systematic-review mode makes reset mandatory at every FULL checkpoint.
* OFF: Preserves pre-v3.6.3 behavior byte-for-byte.
* Resume: New resume_from_passport=<hash> mode allows resuming from the Material Passport ledger alone.
* Ledger: Schema 9 gains reset_boundary[] (append-only) with boundary and resume kinds.
Date: 2026-04-23
* Gate Mechanism: Forces reviewers to pre-commit scoring plans before reading.
* Schema 13: Defines sprint contracts with panel_size, acceptance_dimensions, failure_conditions, etc.
* Two-Call Gate: Phase 1 (blind) output wrapped in <phase1_output> to narrow self-injection surface.
* Templates: Ship full.json (panel 5) and methodology_focus.json (panel 2).
* Reserved Modes: reviewer_re_review, reviewer_calibration, reviewer_guided (retain pre-v3.6.2 behavior).
- v3.5.1: Opt-in Socratic Reading-Check Probe (
ARS_SOCRATIC_READING_PROBE=1). - v3.5.0: Added
collaboration_depth_agent(4th agent); scores user-AI collaboration on a 4-dimension rubric. - v3.4.0: Added
Compliance Agent(PRISMA-trAIce 17 items + RAISE 4 principles). - v3.3.6: Documentation overhaul (Added
ARCHITECTURE.md,SETUP.md,PERFORMANCE.md). - v3.3.3: Release prep, lint hardening (YAML frontmatter), and showcase link fixes.
- v3.3.2: Added
data_access_levelandtask_typemetadata fields. - v3.3.1: Spec consistency patch (aligned mode counts and skill versions).
-
v3.3: Integrated PaperOrchestra techniques (Semantic Scholar API, Anti-Leakage Protocol, VLM Figure Verification).
-
Schemas:
shared/sprint_contract.schema.json,shared/sprint_contract/...,shared/contracts/writer/full.json,shared/contracts/evaluator/full.json. - Validators:
scripts/check_sprint_contract.py,scripts/check_v3_6_6_ab_manifest.py,scripts/check_corpus_consumer_protocol.py. - CI Workflows:
.github/workflows/spec-consistency.yml,.github/workflows/spec-consistency.yml. - Tests:
tests/fixtures/v3.6.6-ab/,scripts/adapters/tests/.
This structure captures the evolution from input handling (v3.6.4) through reset mechanics (v3.6.3) to the complex contract orchestration and paper-blind phases introduced in v3.6.6 and v3.6.7.