Living Memory Wiki — Index

Bundle OKF 0.2 · 0 conceitos · andrewcrenshaw/remember-okf-sample-bundle

Open source Repository Open in the app JSON README (API)

About

# Living Memory Wiki — Index

Total synthesized lessons: 2
Last compiled: 2026-09-05T13:55:31.652Z

## Top Claims
- [[./claims/asserted-fixture-lesson-demonstrating-the-human-actor--d-003.md|Asserted fixture lesson demonstrating the human actor form]] 

## Methods

## Facts
- [[./facts/okf-v0-2-conformance-fixture-lesson-e-001.md|OKF v0.2 conformance fixture lesson]]

Details

Kind: OKF bundles
Topic: AI, RAG & memory
Publisher: andrewcrenshaw
Origin: okf_github
Category: dados
Version: 0.2
Last push: 2026-09-05T14:34:00Z
Repository state: ativo
License: NOASSERTION
Added: 2026-09-08 09:02:43
Updated: 2026-09-08 09:02:43
Origin id: andrewcrenshaw/remember-okf-sample-bundle:bundle/INDEX.md

README

# remember — OKF v0.2 sample bundle (deletion semantics)

A small, fixture-driven OKF v0.2 bundle emitted by the *remember* wiki compiler
(Lexenne), published so the deletion-semantics conventions discussed on
[knowledge-catalog #207](https://github.com/GoogleCloudPlatform/knowledge-catalog/issues/207)
can be tested against real producer output rather than prose.

## What this is, and is not

- **Is:** the unedited output of our production emitter (private repo
  `lexenne-remember`, compiler commit `77b283f8b2406f24bf573da21b196ea989a06847`)
  run over a fixture lesson,
  two log write-claims, and one tombstone. Every byte under `bundle/` came out
  of the compiler; nothing was hand-written.
- **Is not:** an export of our live corpus. That is a larger artifact and
  separate work. This bundle proves the emission path end to end, no more.

## What is planted in it (expected findings)

| case | where | expected reading |
|---|---|---|
| write-claim that landed | `bundle/log.md` → `okf-conformance-fixture-001` | resolves to the concept in `bundle/facts/` |
| actor-class routing (OKF v0.2 section 7) | `bundle/claims/` asserted lesson vs `bundle/facts/` observed lesson | asserted emits `generated.by: human:<8-hex>` (pseudonymous, derived from the session id when the lesson carries one, otherwise from its stable lesson id); observed emits the process form `remember/0.2` |
| basis mapping (#159 enum) | same two lessons | observed emits `basis: live-source`; asserted OMITS `basis` - who stood behind it travels in `generated.by`, and omission carries no negative claim |
| write-claim that never landed | `bundle/log.md` → `okf-conformance-never-landed-999` | `never_landed` — recorded, nothing behind it |
| deliberate removal | `bundle/tombstones.md` | `presence: removed`, reason `curation`, successor named |
| dangling lifecycle edges | concept frontmatter `supersedes` / `contested_by` targets | ordinary §6.1 tolerance — references, not assertions |

Presence is sparse per the #207 convention: `removed` and `never_landed` are
the only values; the default carries no marker.

`bundle/tombstones.md` is the human narration of a removal. **The machine-readable
location for absence records is `bundle/.manifest.json`**, under its `absences` key:
every removed and never-landed concept is filed there against its stable
`remember://lesson/<id>`, so a checker never has to parse prose. A `removed` record
carries the tombstone `date`, the `reason`, the attributing `recorded_by`, and a
`successor` where one exists; a `never_landed` record carries none of those, because
nobody recorded that absence — it is inferred by reconciling `log.md` write-claims
against the bundle. The sibling `absenceReconciliation` key states whether that
reconciliation pass actually ran (`ok` with a `parsedCount`, `failed`, or `not_run`),
so an empty never-landed set is never mistaken for a clean corpus that was never
examined. `entries` and `absences` are mutually exclusive: a currently removed
concept is excluded from `entries` and has no generated page.

**One identifier form for the exclusivity check.** `entries` is keyed by bare lesson
id for page addressing, but every `entries` value now carries an `id` field in the
same `remember://lesson/<id>` form that keys `absences`. A checker asserts
DC-F2 exclusivity by intersecting the `id` VALUES of `entries` with the `id` values
(or keys) of `absences` — one identifier form, no normalisation step — so an empty
intersection is a real assertion, not an accident of two disjoint key namespaces
(knowledge-catalog #207, leesharks run 08; producer fix PCC-6668). Legacy manifests
without the field are backfilled on load. `absenceReconciliation` also now carries
`falseRemovalClaims` (write-claims that named a removed concept), empty here.

## For checker authors — two things you need about our format

1. **Our log prose.** Write-claims in `log.md` look like
   `* **Lesson created**: lesson <id> `. The compiler's own reconciliation
   pass parses them with:

   ```
   /^\*\s\*\*[^*]*created[^*]*\*\*:\s+lesson\s+(\S+)/i
   ```

   Note the claim verb precedes the id.

2. **Our identity scheme.** The stable id lives in each concept's frontmatter
   `id:` key (`remember://lesson/<id>` in edge references); **filenames are
   summary slugs and do not contain the id.** A checker that resolves claims
   against filename stems will not resolve ours — it needs to read `id:` from
   frontmatter. This is the stable-id-versus-path question from
   knowledge-catalog #120 showing up as tooling.

## Integrity

`SHA256SUMS` covers every file under `bundle/`. Verify in a shell:

```
shasum -a 256 -c SHA256SUMS
```

## License

CC0 1.0 Universal (see `LICENSE`) — same terms as the deletion conformance
fixture this is meant to be tested alongside.

Living Memory Wiki — Index

About

Details

README

More