LearnNewsExamplesServices
Frontmatter
id14454
titleGolden Path route-attribution diagnostic — same-run rejection-bucket ledger (closes the #14422 OQ4 gate)
stateClosed
labels
enhancementaiarchitecture
assigneesneo-gpt
createdAtJul 2, 2026, 10:18 AM
updatedAtJul 2, 2026, 12:03 PM
githubUrlhttps://github.com/neomjs/neo/issues/14454
authorneo-fable
commentsCount2
parentIssue14472
subIssues[]
subIssuesCompleted0
subIssuesTotal0
contentTrust
projected
quarantined0
signals[]
blockedBy[]
blocking[]
closedAtJul 2, 2026, 12:03 PM

Golden Path route-attribution diagnostic — same-run rejection-bucket ledger (closes the #14422 OQ4 gate)

Closed v13.1.0/archive-v13-1-0-chunk-8 enhancementaiarchitecture
neo-fable
neo-fable commented on Jul 2, 2026, 10:18 AM

Graduates from Discussion #14422 (OQ4, RESOLVED_TO_AC as this leaf; AC shape converged on-thread by @neo-gpt + author). Standalone, pre-epic: runs before the concept-graph epic exists, because its output is what closes #14422's instrumentation gate and narrows OQ1 with data. Refs #14304 (Lane 4 context).

Problem — snapshots empirically cannot characterize this system

Two independent read-only probes of the same nodes, minutes apart (2026-07-02): 4 GUIDES edges @ struct_score 4.1x vs 1 GUIDES edge @ 0.00. Consistent with frontier churn; conclusive that only same-run instrumentation can explain why the rendered handoff shows Structural: 0.00 on top-ranked items. The route chain has at least four traced exclusion layers — type gate (type ∈ {ISSUE, DISCUSSION}), label/actionability gate (117/277 = 42.2% of OPEN items excluded by {epic, needs-design, needs-re-triage, not-code-ready} — intentional, must be accounted, not "fixed"), blocker gate, and a render-ordering artifact (handoff reads :649 before the frontier GUIDES writes at :762) — none of which are observable together today.

Acceptance Criteria

  • AC1 — the same-run ledger: one full Dream/Golden-Path cycle emits, in a single run, per candidate: semantic presence + score · OPEN/state check · type-gate result · label/actionability bucket (with per-label counts) · blocker-gate result · pre-ranking structural components per edge type as (nodeId, semantic, structural, final) · post-selection frontier —GUIDES→ item writes · rendered component values.
  • AC2 — committed artifact: the ledger lands as learn/agentos/measurements/golden-path-route-attribution-<date>.md with the run's config snapshot (thresholds, pool size, frontier source).
  • AC3 — read-ordering fix rides along: handoff reads occur AFTER the cycle's GUIDES writes, or the ordering is explicitly documented as correct with the reason.
  • AC4 — acceptance fork: structural ≠ 0.00 observed for ≥1 top-10 item in-run, OR the dead-write / cold-start defect is confirmed with the exact rejection bucket named and routed to a fix ticket.
  • AC5 — schema discipline: any logged edge attributions carry the #14422 OQ6 four-axis contract properties (authority-ref / fidelity / extractionProvenance / lifecycle) — no bespoke schema.
  • AC6 — scope note: log/artifact only; no new node classes minted (the #14426 post-sync canary is N/A and this AC records why).

Sequencing / who this unblocks

Per the goal-scoping map (2026-07-02): this is the linchpin's first physical move — its output unblocks (a) #14422 §6.2 graduation → Golden-Path-v2 epic, (b) business-engine Leaf 2 (#14442/#14446 chain), (c) the cold-start disposition (#14422 OQ4 / #14453 OQ5), (d) #14447's ranking-input contract. Unassigned by design — implementation-tier, claimable.

tobiu referenced in commit 1666a3d - "feat(ai): add golden path route attribution ledger (#14454) (#14458) on Jul 2, 2026, 12:03 PM
tobiu closed this issue on Jul 2, 2026, 12:03 PM