Skip to content

Add shared BGE text map and verified graph provenance - #935

Draft
realmarcin wants to merge 13 commits into
mainfrom
fix/embedding-graph-900
Draft

realmarcin wants to merge 13 commits into
mainfrom
fix/embedding-graph-900

Conversation

@realmarcin

@realmarcin realmarcin commented Sep 15, 2026

Copy link
Copy Markdown
Contributor

CommunityMech gains a semantic-text PaCMAP view of all 378 records in the validated source snapshot, including four isolates. It uses the fleet's pinned BAAI/bge-large-en-v1.5 revision d4aa6901d3a41ba39fb536a557fa166f842b0e09, 1,024-dimensional normalized vectors and a 512-token window. Existing taxon-graph PaCMAP and Graphviz sfdp views remain available; both represent 373 of 374 communities, with the remaining omission and excluded isolate population explicitly recorded.

Graph matching measures coverage over unique requested taxa and preserves partial coverage. Missing vectors no longer imply fabricated host classifications. Each regenerated graph binds the actual streamed source, current YAML bytes, matching ledger, matrix, reducer settings and output checksums. Publication validates both current graph receipts and freshly exported common-map inputs before staging. The runtime also rejects symlinked or misnamed generation paths. The common text maps share the pinned BGE model; their coordinates are fitted independently and are not aligned across repositories.

Validation against captured main f7c8fae387add00eff6686716fcf06e29455bf2a: the current common build reused all 376 previous semantic vectors and encoded only the two incoming communities. Independent source and complete numerical reviews verified all 378 records, exact input/profile/cache bindings, and byte equality of the three map artifacts built with the prior canonical runtime and validated with the corrected runtime.

Independent reviews accepted both 373-point graph layouts and all 417 installed site files, including 383 fresh renderer outputs. Every common and specialty destination resolves. The two new communities retain their actual 100% and 75% requested-taxon coverage; one separate no-vector omission and the four excluded isolates remain explicit. The exact local merge passed all 12 affected commands: strict validation, writer audit, fresh docs reproduction, paired publication, all 53 affected tests, schema/ontology/GTDB validation for both added records, and history validation. The previous maintained full QA remains evidence for unchanged implementation and prior records: 3,264 tests passed (99 skipped, 8 deselected), plus all configured ontology/term checks. Final-head GitHub CI passed on 91550e96300f805e08e8fb2a394a6abc7f33c9f7: all seven validation checks succeeded; the optional repair-suggestion job was skipped.

Fixes #900, #905, #907, #909, #910, #911 and #936. Implements CommunityMech's portion of CultureBotAI/culturebotai-claw#426 and the reviewed runtime correction in CultureBotAI/culturebotai-claw#440.

Fleet rollout retains released pin c8b8f89ecf29574d7e24f6c65cfd925aefbdd007. Additive governed files are verified against reviewed canonical candidate 16ce0e452af975cdfe4acc181efb6bad4688f5b7. After all ten integration PRs land, CultureBotAI/culturebotai-claw#432 can register the complete fleet. Separate pin-only PRs will use its actual main-reachable squash merge commit after governed-byte verification. This draft remains for coordinated review.

Refs #900; full verified graph regeneration remains tracked separately.
Addresses #900 and #905. Historical arrays remain unverified; full artifact regeneration follows this reviewed source checkpoint.
Addresses #910. Full 372-record semantic JSONL export is byte-identical.
Publish complete current-corpus graph coordinates, matching/omission ledgers, actual reducer provenance and output-bound receipts.
Addresses #911 and CultureBotAI/culturebotai-claw#435. Check both complete graph output sets and current community corpus without source vectors or projection libraries, and trigger publication checks on their validation dependencies.
@github-actions

Copy link
Copy Markdown

Network integrity findings

Warnings only — a member with no interaction yet, or a participant matched by ontology id rather than by name, or one on a community-level interaction that resolves to no member. Reported, but does not fail the build.

Network Integrity Audit Report
================================================================================

0 error, 13 warning across 4 records with findings
Only error-severity findings fail the build.

Bacillus_G12_Y4_X25_Tobacco_Biocontrol_SynCom
--------------------------------------------------------------------------------
  [warning] UNKNOWN_TARGET: Target taxon 'tobacco' not found in taxonomy section (community-level scope)
  [warning] UNKNOWN_TARGET: Target taxon 'tobacco' not found in taxonomy section (community-level scope)

Total: 2 issues (0 error, 2 warning)

Bacillus_siamensis_vallismortis_HT_Masson_Pine_SynCom
--------------------------------------------------------------------------------
  [warning] UNKNOWN_TARGET: Target taxon 'Fusarium oxysporum' not found in taxonomy section (community-level scope)
  [warning] UNKNOWN_TARGET: Target taxon 'Pinus massoniana' not found in taxonomy section (community-level scope)
  [warning] UNKNOWN_TARGET: Target taxon 'Pinus massoniana' not found in taxonomy section (community-level scope)

Total: 3 issues (0 error, 3 warning)

Sclerotinia_Sclerotia_12Strain_Biocontrol_SynCom
--------------------------------------------------------------------------------
  [warning] UNKNOWN_TARGET: Target taxon 'Sclerotinia sclerotiorum' not found in taxonomy section (community-level scope)
  [warning] UNKNOWN_TARGET: Target taxon 'Sclerotinia sclerotiorum' not found in taxonomy section (community-level scope)

Total: 2 issues (0 error, 2 warning)

Space_Habitat_SevenMember_Stress_Tolerance_SynCom
--------------------------------------------------------------------------------
  [warning] DISCONNECTED: Taxon 'Acinetobacter radioresistens 50V1' has no interactions
  [warning] DISCONNECTED: Taxon 'Micrococcus luteus W4-Al-K1' has no interactions
  [warning] DISCONNECTED: Taxon 'Propionibacterium cyclohexanicum TA-12T' has no interactions
  [warning] DISCONNECTED: Taxon 'Pseudomonas antarctica CMS 35T' has no interactions
  [warning] DISCONNECTED: Taxon 'Staphylococcus capitis K1-2-2-23' has no interactions
  [warning] DISCONNECTED: Taxon 'Streptococcus halotolerans HTS9T' has no interactions

Total: 6 issues (0 error, 6 warning)

The full report is attached to the workflow run as an artifact.

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

Stop classifying missing graph vectors as hosts and report actual taxon coverage

1 participant