Description
Owner 2026-10-24: the private demo’s telemetry reads as synthetic (identical values across sources and functions, very low volumes, incoherent status mix, AWS stores idle while functions run, placeholder resource names) and panel titles carry proof caveats. Make it production-scale and plausible against the private flow design, and make the dashboards look like a real operations team built them. Work lands in the private source and the demo stack only; any generic engine capability it needs (per-entity weights, diurnal shape, per-instance profiles) lands in synthkit as a generic feature.
Acceptance Criteria
- #1 Live panels show distinct, plausible per-source and per-function values at production-like volume with a diurnal shape; status mix, rejection fraction and route latencies agree with each other and with the private flow design
- #2 Queue, table and function metrics move coherently with request volume (stores are not idle while functions run)
- #3 Dashboard and panel titles, resource names and descriptions read as a real operations estate; no placeholder or caveat titles; browser screenshots at baseline and during the headline incident recorded
Definition of Done
- #1 just check (fmt-check, lint, gen-check, env-check, docs-check, test, race, hygiene, ui-check, compose-check, helm-test, lab-check, signal-fidelity)
- #2 just gen (only if a blueprint field, construct/workload config struct, or a skill under plugins/synthkit/skills/ changed)
- #3 just dump — inventory diffed against signals/
Implementation Plan
Compose the retained three-file private pass-three source onto current infrastructure main without reverting sibling runner changes; include paired restart-hash values and reconcile identity docs. Final image and any new generic fields wait for public candidate acceptance; root gates, snapshots rollback, deploys and proves browser/Assistant stories.
Loop60: static private draft preparation after private preflight/audit; final runtime/image/deployment acceptance remains gated on accepted causal source, evaluation cost, native review and exact proof.
Implementation Notes
loop57: landed engine f5b1685..83d69fc; private blueprint rkps-awsinfra 0c401b9 live on main-311d472; dashboards/rules/5 SLOs/Assistant rules applied; browser renders baseline + headline (codex/loop57/browser/); KG enabled with custom rules.
Reopened 2026-10-25 for loop58: causal incidents (children .15 to .25), private blueprint on the causal model, dashboards and alerts, and browser plus Assistant proof.
loop58 closeout (2026-10-25): parked unlanded; see /Users/rob/repos/synthkit/codex/closeout-2026-10-25-loop58.md. Private demo pass 3 committed locally on rkps-awsinfra branch loop58-pbp (dd99992), not pushed or deployed; owner: no live work 2026-10-25.
Owner closeout: Static private source draft only. Resume final engine, cost bounds, duration/rejection/private wiring, native review, infrastructure gate and deploy/browser/Assistant admission. No deployment.
Loop60 closeout: Static private draft remains unchanged; repaired verifier passed native review and offline negative controls only. No final-source Go build, actual capture, whole-runner cold/catch-up/fault/RSS, image, infrastructure gate, deployment, browser or Assistant acceptance this run. Resume only after owner-authorized cost resolution and accepted public engine; retain existing rollback deployment.
Owner approved bounded performance repair and actual whole-private-runner proof before deployment. Independent senior resumes only to prepare new local-only whole-runner cold/catch-up/fault/RSS tooling, preserving all source numbers, consumers, horizon, real runner invocation boundaries and prior full-facts oracles. Root executes exact final source later; preparation is not runtime, image, deployment or task completion.
Owner scope amendment: a separate session now owns demo dashboard authoring and application. This loop must not create, update, delete or restore dashboards or mutate their folder. Existing alert-rule, SLO and Assistant-rule authority remains. Root records old/new names for any changed dashboard-queried metric, label or log field in machine-local coordination notes; no dashboard completion is inferred from source/cost work.
Second private-task convergence stop: initial whole-runner proof tool falsely accepted primer-only dynamic coverage and could abandon a child. Escalated repair fixed coverage and ordinary cleanup, but native delta review proved a second termination during cleanup entry still bypasses group kill/reap. No Go build, real proof child or engine measurement ran. Owner reopening required before any further supervisor/tool repair or execution; preserve all archived packets/probes and keep existing demo image/blueprint. No renamed bypass or dashboard mutation.
loop61: resumed under the owner’s Q04 reopening (charter) after the realism set deployed; needs-owner removed.