Hiro development journal

Recommend stopping current Daylab and integrating a bounded evidence collector

2026.08.10 at 16:05:00 America/Los_Angeles Design recommendation recorded; runtime unchanged Machine-readable JSON

Executive summary

The current standalone Daylab loop should be stopped rather than connected directly to Stage 6.

Its useful component is adaptive evaluation evidence collection, but its current 30-minute workflow repeatedly rereads response envelopes, reruns probes, regenerates generic curiosity and journal artifacts, and emits completion notifications without a novelty requirement.

A narrower Daylab-derived collector should eventually become an upstream Stage 6 input. It should submit an evidence packet only when a new failure signature or materially stronger evidence crosses explicit thresholds.

Stage 6 must remain the sole authority for candidate construction, validation, authorization, and promotion. Daylab should never acquire direct promotion authority.

No process, scheduler, notification preference, or Stage 6 control was changed while making this recommendation.

Work completed

Stop-versus-integrate decision

Complete

Compared the loop's valid evaluation output with its repeated diagnostics and artifact generation.
Rejected direct integration of the existing loop because it lacks novelty gating and would create continuous pressure on a safety-sensitive promotion lane.
Recommended stopping the standalone loop while preserving its historical evidence and reports.

Bounded Stage 6 integration design

Proposed

Retain adaptive suite selection and isolated evaluation execution as the collector's core responsibilities.
Deduplicate by failure signature and opportunity key before creating an evidence packet.
Require either a novel failure or a material increase in independent supporting observations.
Apply explicit daily model-call and packet-count budgets and a backoff interval after clean or duplicate runs.
Remove personality journaling, generic curiosity generation, repeated envelope analysis, and routine Telegram completion notices from the collector path.
Hand qualifying evidence to Stage 6 without granting the collector construction or promotion authority.

Decisions and reasoning

Stop the current standalone Daylab loop when implementation is authorized.
Do not integrate the current full workflow into Stage 6.
Reuse only the adaptive evaluation collector behind novelty, deduplication, budget, and backoff controls.
Keep Stage 6 as a separately authorized consumer with exclusive promotion authority.
Make no runtime change in response to a design-advice question.

Validation and evidence

Check	Status	Result
Recommendation basis	confirmed	The prior live inspection found 16 cycles, 252 evaluation calls, 352 probes, 12 specs over five opportunity keys, and a latest clean cycle with no new spec.
Authority boundary	confirmed	The current Daylab implementation is proposal-only and has no direct promotion authority.
Journal tests and production build	passed	npm run test:hiro passed, generated-output validation passed for 91 entries, and npm run build completed successfully.

Current state

Daylab remains active at its existing cadence because this session produced a recommendation rather than an execution authorization.
Stage 6 remains disabled and is not consuming current Daylab output.
No notification or runtime behavior changed.

Next steps

Obtain explicit authorization to stop the current Daylab process.
Specify and test the bounded evidence-packet contract, novelty rules, deduplication keys, budgets, and backoff behavior.
Resolve the separate Stage 6 disabled state before activating any new evidence handoff.
Reactivate the collector only after tests demonstrate that duplicate and clean cycles do not create packets or notifications.