Example from HELIX’s own docs. This generated page comes from docs/helix/. Use it to see the method in practice; start with the artifact-type catalog for reusable templates. Historical plans and reports may describe retired architecture.

Source identity (from 06-iterate/metrics-dashboard.md):

ddx:
  id: metrics-dashboard
  authoring:
    home: repo
  review:
    self_hash: 08901f26fcd1422bd385122af0254abacc088a9f0eed2ae7a1df3173e2a5ee5a
    deps: {}
    reviewed_at: "2026-09-16T03:20:33Z"

Metrics Dashboard: HELIX 2026-Q3 (human-outputs branch)

Review Window: 2026-09-10 → 2026-09-15 (feat/helix-human-outputs) Baseline: none. This is the first sample against the PRD’s success metrics (docs/helix/01-frame/prd.md Success Metrics). The previous dashboard measured the execution loop (bead close rate, cycle time, wrapper-CLI green rate); those metrics belong to the runtime and are retired here. Status: Complete

Decision

The branch made three PRD metrics measurable for the first time and measured a fourth surface, present mode, through its own gate. The eval run passes 33 of 36 deterministic checks and scores 57/62 on the rubric; the skill body carries zero runtime-specific commands; the one deck produced so far passes the deliverable gate with 0 blocking findings. The alignment-quality target (fewer than three findings per align run on a healthy artifact set) is not yet measurable: the only sample is a fixture seeded with drift. No metric crossed a degradation threshold because there is no prior reading to degrade from.

Summary

evals/results/20260915-2202/summary.md is the current sample for the skill metrics (commit 20315d5c, nine briefs, about $32 per run); the present-mode sample is the DEL-001 gate record (docs/helix/06-iterate/deliverables/DEL-001-helix-evaluation-deck.md Render section). Authoring quality reads well on the proxy available (template-authored PRDs pass validate-instance.py on the first attempt); the PRD’s comparative study against free-form PRDs has not been run.

Metrics Table

Interpretation Rules

• Deterministic checks are pass or fail; a lost check on any brief is a regression and blocks the change that caused it.
• Rubric scores are a trend signal between runs, not a pass mark; a change is kept when checks stay green and the rubric total moves up on the briefs it targets (evals/README.md Reading a result).
• The alignment target reads only on a healthy artifact set. A finding count on a seeded fixture records the mode’s recall, not the metric.
• A present-mode gate warning is tolerated when the Render section names it and the reason; a blocking finding is not.
• A row marked “not measurable yet” creates a follow-up; a “proxy” row creates a follow-up to replace the proxy with the PRD’s measurement.

Trend Notes

• Second reading (20260915-2202 at 20315d5c) against the first (20260911-0955 at 0756e5fa): checks went from 32/32 over eight briefs to 33/36 over nine; the three losses are gate and wording rules that did not exist at the first run (evals/README.md Runs). Rubric went from 48/54 to 57/62.
• The two rubric points lost on present-investor-deck were evidence visibility (slide bodies and owner dates fall outside the truncated diff the judge sees), not deck content.
• The two points lost on align-baseline were a truncated handoff block and fixture text absent from the evidence, both report shape rather than finding quality; the one point lost on validate-prd was judgment content authored in place.

Follow-Up

• Add a healthy-set align brief (a fixture with no seeded drift, or docs/helix/ itself) so the < 3 findings target has a reading; improvement-backlog.md carries the item.
• Fix the three failed checks (coverage vocabulary and the shape rule in the present mode text; the title stop) and run the eval again.
• Replace the first-review proxy with the PRD’s comparative study (template-authored vs free-form PRDs through the same review).

Review Checklist

• Baseline is explicit (none; first sample, prior execution metrics retired)
• Each metric cites a source (an eval result file, a test lane, or the DEL-001 gate record)
• The summary states the decision implication (checks and portability pass; alignment target not yet measurable)