MAQS-Way
A path-layer standard for LLM long-term memory systems that bridges the gap between compliance policy and engineering implementation. It provides executable schemas, lifecycle and pattern specifications, and includes eight immutable principles.
Timeline (last 90 days)
MAQS-Way's feedback mechanism stipulates: if there is a misreading in an audit report, it can be corrected by filing an issue, the correction lands as a second commit, and what is discovered is never silently rewritten.
The entire verification process of MAQS-Way can be executed with a single command `python3 schemas/maqs_way_audit.py all`, whose output includes passing information such as schema 4/4, manifest 12/12, declared 8/8 as well as the printed gaps (such as B4), with the gate result being GATE GREEN.
MAQS-Way's gate mechanism loudly lists known gaps while keeping the light green; the green light certifies compliance with the release rules (including 'every gap is listed'), not the absence of problems; its design philosophy is that trust in a standard is built on transparency rather than perfection.
LoCoMo focuses on fact recall and consistency in multi-turn long conversations, while LongMemEval covers a broader range of long-term memory tasks, including temporal reasoning, conflict resolution, and forgetting verification, and both are used in MAQS-Way as measurable verification means.
The MAQS-Way team used the same checklist to conduct a static audit of three open-source memory systems, Mem0, Graphiti, and Letta MemFS, adopting file:line evidence and a constructive tone, and rated exemplary items and gap items for each system.
In Mem0, the tenant filter reaches the vector store before top-k (main.py:1471), and isolation is proven through 'attempted violations'; Mem0 is exemplary in tenant isolation, but partially meets standards on provenance, aging, deletion residue, and write atomicity, and has gaps in trust types and built-in evaluation tools.
The remaining gaps in MAQS-Way (including B4 write atomicity, normalization of E10/E11, I10–I12, I14, I15, P9/P10, dead code rule 0.5a and drift threshold 0.6a, twelve in total) were all rescheduled to v0.3, each has an owner, and none was quietly handled.
MAQS-Way author Aleksandr Kossarev comes from Jõgeva, Estonia, and his project stance is: "Don't check whether a mechanism exists, check whether it actually works."
MAQS-Way v0.2 has been officially released, the first minor version of a practical standard for AI long-term memory, and at release it proactively disclosed 12 known gaps, each with an owner and a target version, listed directly in the specification document.
MAQS-Way is positioned as the pathway layer between the policy layer and the audit layer, comprising 8 principles (P1–P8), a data model based on OMPS, lifecycle management, and measurable verification means (LoCoMo, LongMemEval, deletion lattice, negative tests).
8 more timeline events
All Facts (18)
MAQS-Way's feedback mechanism stipulates: if there is a misreading in an audit report, it can be corrected by filing an issue, the correction lands as a second commit, and what is discovered is never silently rewritten.
50%UnverifiedMAQS-Way is positioned as the pathway layer between the policy layer and the audit layer, comprising 8 principles (P1–P8), a data model based on OMPS, lifecycle management, and measurable verification means (LoCoMo, LongMemEval, deletion lattice, negative tests).
50%UnverifiedThe way-v0.1.3 frozen by MAQS-Way v0.2 is a specific version snapshot of OMPS (schema_version is way-v0.1.3, independent of the release version number v0.2), and all v0.2 changes are at the wording and test framework level, not touching the data model.
50%UnverifiedThe entire verification process of MAQS-Way can be executed with a single command `python3 schemas/maqs_way_audit.py all`, whose output includes passing information such as schema 4/4, manifest 12/12, declared 8/8 as well as the printed gaps (such as B4), with the gate result being GATE GREEN.
50%UnverifiedMAQS-Way's gate mechanism loudly lists known gaps while keeping the light green; the green light certifies compliance with the release rules (including 'every gap is listed'), not the absence of problems; its design philosophy is that trust in a standard is built on transparency rather than perfection.
50%UnverifiedLoCoMo focuses on fact recall and consistency in multi-turn long conversations, while LongMemEval covers a broader range of long-term memory tasks, including temporal reasoning, conflict resolution, and forgetting verification, and both are used in MAQS-Way as measurable verification means.
50%UnverifiedIn Mem0, the tenant filter reaches the vector store before top-k (main.py:1471), and isolation is proven through 'attempted violations'; Mem0 is exemplary in tenant isolation, but partially meets standards on provenance, aging, deletion residue, and write atomicity, and has gaps in trust types and built-in evaluation tools.
50%UnverifiedThe remaining gaps in MAQS-Way (including B4 write atomicity, normalization of E10/E11, I10–I12, I14, I15, P9/P10, dead code rule 0.5a and drift threshold 0.6a, twelve in total) were all rescheduled to v0.3, each has an owner, and none was quietly handled.
50%UnverifiedMAQS-Way author Aleksandr Kossarev comes from Jõgeva, Estonia, and his project stance is: "Don't check whether a mechanism exists, check whether it actually works."
50%UnverifiedMAQS-Way attempts to fill the path layer missing from LLM long-term memory compliance, which specifies which schemas, lifecycles, and patterns should be used and how to prove they are effective.
50%UnverifiedOMPS is a vendor-neutral portability data model project that provides memory-item.schema.json and deletion-marker.schema.json.
50%UnverifiedThe MAQS-Way lifecycle defines a complete state machine: ACTIVE→STALE→CONTESTED→SUPERSEDED→EXPIRED→ARCHIVED→DELETED→TOMBSTONE→COMPACTION, together with dual timestamps.
50%UnverifiedThe MAQS-Way tombstone record format is deletion_log(request_id, timestamp, scope, layer_counts, requestor), with zero PII, Ed25519 signatures, and a hash-chained structure.
50%UnverifiedThe MAQS-Way verification layer defines a minimal golden set of metrics and target values: retrieval R@5 ≥0.85, temporal reasoning ≥0.8, abstention ≥0.9, duplication ≤2%, contamination 0, deletion residue 0%, isolation leakage 0, poisoning blocking 100% ASI06.
50%UnverifiedMAQS v1.6 is an audit checklist that outputs three levels of verdicts, CRIT/IMP/REC; MAQS-Way is a governance contract that cites MAQS's 0-J sections as implementation evidence.
50%UnverifiedThe MAQS-Way repository contains the machine-readable 8 principles, 7 test vectors under the conformance/valid/invalid directories, and an evals/ directory for running LoCoMo/LongMemEval.
50%UnverifiedMAQS-Way v0.2 has been officially released, the first minor version of a practical standard for AI long-term memory, and at release it proactively disclosed 12 known gaps, each with an owner and a target version, listed directly in the specification document.
50%UnverifiedThe MAQS-Way team used the same checklist to conduct a static audit of three open-source memory systems, Mem0, Graphiti, and Letta MemFS, adopting file:line evidence and a constructive tone, and rated exemplary items and gap items for each system.
50%