Knowledge systems and governance

Organisational Memory + OMC

Repository-centred project memory with explicit authority, provenance and deterministic retrieval.

Status
active
Checked
23 October 2026
Evidence
1 public link
Type
Knowledge systems and governance

Current status: Repository governance is active, with Stage 8 complete and Stage 9 under way. OMC's deterministic retrieval was built and then parked; no implementation work is authorised while it is paused, and resuming it needs a fresh decision.

The problem

Conversational memory was not reliable enough for long-running technical work. It could collapse current state, history, proposals and decisions into one plausible-sounding narrative. Recency could be mistaken for authority, and a retrieved statement could arrive without enough provenance to judge it.

Organisational Memory was created to make project knowledge durable, reviewable and governed outside the conversation that happens to use it.

The system Mick designed

Organisational Memory is a repository-centred authority system, not a claim that one repository automatically outranks every other source. Knowledge is stored with structure that separates current state from history, proposals from accepted decisions, and project-local context from broader authority. The applicable authority boundary determines which committed source owns a fact.

The current system deliberately supports split authority: responsibilities that have been explicitly migrated can be owned by the successor repository while excluded responsibilities remain with their prior authoritative source until a separate cutover decision.

OMC — the Organisational Memory Compiler — is a deterministic, read-only interface over explicitly pinned committed repositories. Its retrieval layer was built to select bounded attributable context for another tool to use. It reports relevance and provenance; it does not become the decision-maker simply because it can retrieve the record.

Constraints that mattered

  • Committed sources outrank conversational recollection within their accepted authority scope.
  • Current state, history, proposals and decisions remain distinct.
  • Retrieval must preserve source identity and provenance.
  • Registry membership alone does not prove current project status.
  • Newer material does not automatically outrank accepted authority.
  • Split authority must remain explicit until a separately approved cutover.
  • Consequential decisions remain human decisions.

What proved difficult

The central difficulty was preserving meaning across ingestion, compilation and retrieval. A technically successful search can still be wrong if it drops lineage, merges superseded material, treats relevance as authority, or presents a proposal as an accepted decision. Evaluation therefore had to examine not only whether a result was returned, but whether the consuming workflow could still determine where that result came from and what authority it actually carried.

How failure was handled

The work used staged controls and deterministic evaluations. Retrieval defects and ambiguous authority were treated as system problems, not patched over with better-sounding prose. Where automation could not safely decide between competing records, the system preserved the conflict or stopped the affected work for human adjudication instead of inventing a winner.

Implemented today

The repository system provides structured organisational knowledge, explicit authority and approval boundaries, durable provenance, deterministic validation and bounded successor ownership. The evidence-gated manual successor work stands at Stage 8 complete, with Stage 9 under way: remaining manual responsibilities are being cut over pack by pack, and each pack needs its own accepted decision.

OMC Stage 4 was the implemented compiler baseline: it can deterministically retrieve bounded exact context from an explicit set of pinned repositories while remaining read-only and network-independent. It is now parked. The locally available models could not separate retrieval defects from raw model capability failures, so further retrieval work would have optimised the wrong bottleneck. Resuming it requires a fresh decision.

The private operational corpus is deliberately not published. The linked governance reference is a sanitized, synthetic, non-authoritative architectural example of the principles behind the work. It is not independent proof of the private system’s current operational state.

Validation evidence

Validation has included deterministic repository and compiler checks, black-box retrieval evaluation, provenance and authority tests, and controlled fresh-context workflows. The public governance repository shows the evidence-first and human-authority model without exposing private project memory.

Current boundary

The core repository system is in use. OMC Stage 4 was built and evaluated, but the compiler is parked and takes no part in current operation. Full successor cutover is not implied, and retrieval remains downstream from the authority sources it reads.

Private memory, internal operator procedures, source locations and unpublished project records remain outside the public boundary.

Public boundary: This is an architectural account only. It does not reproduce private Organisational Memory records, internal paths, operator material or private repository content.