Improvement Backlog — HELIX 2026-Q3

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/improvement-backlog.md):

ddx:
  id: improvement-backlog
  authoring:
    home: repo
  depends_on:
    - metrics-dashboard
  review:
    self_hash: 952df3e5bbb6e967cd0e3d485ae7106893c464008ecda327e9c0c6a33527e2cd
    deps:
      metrics-dashboard: 486cd18a01712b15850ce7dd808d432a8a52cc56d1db22a8939ed31b633191c0
    reviewed_at: "2026-09-16T03:20:07Z"

Improvement Backlog — HELIX 2026-Q3

Iteration: 2026-Q3 (post-v0.12.0; the human-outputs branch feat/helix-human-outputs) Source Learnings: the 2026-09-10 top-to-bottom evaluation that produced the human-outputs plan (docs/helix/01-frame/prd.md non-goals and R-4; docs/helix/00-discover/product-vision.md success definition), the first present run (docs/helix/06-iterate/deliverables/DEL-001-helix-evaluation-deck.md Assumptions and gaps), and docs/helix/06-iterate/metrics-dashboard.md.

Prioritization Rules

Rank by authority leverage: items that keep the skill inside the PRD’s non-goals (content plus one skill, no execution engine, no imposed technology) rank above cosmetic improvements.
Rank by evidence leverage: items that make a success metric measurable rank above items that add surface.
Rank by public-surface impact: items that make the published reference more accurate rank above internal cleanups.
Items without tracker references stay at the bottom until they are filed; the backlog does not retain unsourced ideas.

Backlog Items

Priority	Item	Evidence	Tracker Ref	Why Now	Status
P1	Build the evaluation the PRD promises: fixed corpus, headless run per host, `validate-instance.py` plus a rubric, results published; retire the 922-file `family-test/` scaffold that does not run	`evals/results/20260915-2202/summary.md`: nine briefs at commit `20315d5c`, 33/36 checks, rubric 57/62, about $32 per full run; `family-test/` removed	this branch	The present mode and the deliverable gate gave the eval something concrete to grade	done
P1	Rasterize `.pptx` renders in the deliverable gate on the authoring host (LibreOffice, or PowerPoint automation permission)	`skills/helix/scripts/deck-qa.py` rasterizes through LibreOffice and writes a contact sheet; DEL-001 Render section records the inspection	this branch	A deck nobody looked at is not finished; the gate says so	done
P2	Sub-agent fan-out in align, review, and converge: one agent per review dimension or artifact family, fan-in through `workflows/modes/_report.md`	`workflows/modes/align.md` Fan-out section; review and converge likewise	this branch	The report shape exists, so fan-in has a contract	done
P2	Read governing artifacts that live in a connector-backed tool (Google Docs, Notion, Jira) directly, instead of only through the checkout stub	`workflows/conventions.md` Reading through a connector; `authoring.connector` in the schema	this branch	Hosts expose document connectors; read-only access removes a manual copy step	done
P2	Consolidate the install index and six host guides (1,946 lines across seven files on `main`) into one guide plus per-host deltas, and test `.github/copilot-instructions.md` against the router	`docs/install/README.md` (886 lines across three files); `tests/validate-install-consistency.sh` keeps the Copilot file a pointer	this branch	The router split changed what every guide must describe	done
P2	Port Sloptimizer’s slide target into the deliverable gate: `shape.slop` and `restatement` checks on bullets, labels, captions, and verdicts, kept in sync with the upstream fixture	`skills/helix/scripts/check-deliverable.py`; `tests/validate-headline-sync.sh`; DEL-001 Render, Gate	this branch	The headline port left shapes unchecked; the deck rebuild showed the same slop below the titles	done
P2	Add a healthy-set `align` brief (a fixture with no seeded drift, or `docs/helix/` itself) so the PRD’s fewer-than-three-findings target has a reading	`docs/helix/06-iterate/metrics-dashboard.md` alignment row: the only sample is the seeded `recipe-app/baseline` fixture (28 findings)	pending bead	The eval runs; one brief closes the only PRD metric still unmeasured	open
P3	One-pagers and briefs render end to end: `scripts/render-doc.js` builds a masthead-plus-introduction document flow (the title unit’s Body is authored introduction prose, not a field dump) and HTML section patterns for `table`, `two-column-comparison`, `stat-callout`, `claim-evidence` visualized as `panels`/`icon-list`, and `risk-matrix`; a one-pager packs into a weight-balanced two-column grid with wide units spanning full width; output is HTML plus PDF via a local Chrome/Chromium headless print-to-pdf, no native dependency	`skills/helix/scripts/render-doc.js`; `workflows/modes/present.md` §Kinds; `workflows/activities/06-iterate/artifacts/deliverable/prompt.md` §Writing a document	this branch	The deck pipeline showed the shape; the gap was a real renderer and real authoring guidance, not a research question	done
P3	Kind-aware limits in `check-deliverable.py`: per-pattern word/bullet caps now scale up for a document instead of applying the slide’s own numbers, `consecutive_same_pattern_max` is relaxed for a document, and a one-pager’s word budget (`theme.yml`’s `max_words_one_pager`) and a brief’s estimated page count (a weight heuristic shared with `render-doc.js`’s one-pager grid balancer) are checked at gate time, all as warnings — the actual render-and-look step stays the ground truth	`skills/helix/scripts/check-deliverable.py` (`is_doc`, `WORD_SCALE`/`COUNT_SCALE`, `unit_weight`, `DOC_WEIGHT_PER_PAGE`)	this branch	The HTML render path landed; a gate that still checked every kind against slide caps was the obvious next gap	done
P3	`.docx` renderer for one-pagers and briefs	`skills/helix/scripts/render-doc.js` (HTML/PDF only)	pending bead	The HTML path is done and in daily use; `.docx` is a separate, narrower, unrequested format	open
P3	Prune overlapping catalog types: fold `test-suites` and `test-procedures` into `test-plan`; decide whether `market-analysis` stays separate from `competitive-analysis`	03-test carries six types; discover carries four market documents	pending bead	The catalog table and graph regenerate cleanly now, so a prune is mechanical	open
P3	Sloptimizer adapter section for the `human-facing` profile	`easel-skills` `skills/sloptimizer/references/adapters-helix.md` § “Deliverable Titles (`human-facing` profile)” ties the adapter to `workflows/voice.yml`’s `human-facing` profile by name instead of the prior `artifact-signal` hard-code	easel-skills `eb13ce6`, `f5de158`	A rewrite now lands inside the deck voice instead of fighting it	done
P3	Add `ddx.type` to every catalog `example.md` so type resolution never falls back to the directory heuristic	`tests/validate-instance.sh` asserts the heuristic; every example emits a `frontmatter.type` warning	pending bead	Cheap once, removes 53 warnings	open

Selection for Next Iteration

Closed this iteration: P3 — the Sloptimizer adapter section for the human-facing profile landed in easel-skills (eb13ce6, f5de158), so hosts with Sloptimizer installed now get the same title rules the HELIX gate ports; see the Backlog Items row above for evidence.
Next candidate: P2 — a healthy-set align brief, the only PRD metric still unmeasured (see Backlog Items).

Review Checklist

Each item cites evidence
Tracker references are included (or marked as “pending bead” where not yet filed)
Ordering is deterministic (priority + position within priority)

Innsigle seal: model-primary by HELIX

The signature covers the markdown source of this page, not these HTML bytes. This page quotes that seal; verify it against the source file.

Composition: model-primary
Issuer: HELIX helix
Signing key: ed25519:b0865d76d834a52c48506414d16f4e5a (build key)
Signed source: artifacts/improvement-backlog.md
Signed: 2026-09-23T14:11:58Z
Content digest: sha256:d2bf4d19…d40a394c

This build key is endorsed by the human key for build signing; the signature is not a detector and not a truth guarantee.

Raw attestation JSON

{
  "payload": {
    "innsigle": "1",
    "type": "https://innsigle.dev/claim/colophon/v1",
    "issued_at": "2026-09-23T14:11:58Z",
    "issuer": {
      "id": "helix",
      "name": "HELIX",
      "key_id": "ed25519:b0865d76d834a52c48506414d16f4e5a",
      "key_url": "https://documentdrivendx.github.io/helix/.well-known/innsigle/keys.json"
    },
    "subjects": [
      {
        "uri": "https://documentdrivendx.github.io/helix/artifacts/improvement-backlog/",
        "digest": {
          "alg": "sha256",
          "value": "d2bf4d199ad75c00f41ee5a015d4242e9b73cf16da8ac92480d6ddb0d40a394c"
        }
      }
    ],
    "colophon": {
      "schema_version": "1",
      "composition": "model-primary",
      "ingredients": [
        {
          "kind": "model",
          "name": "Claude",
          "role": "draft"
        },
        {
          "kind": "tool",
          "name": "sloptimizer",
          "role": "rewrite"
        },
        {
          "kind": "human",
          "name": "operator",
          "role": "structure-edit"
        }
      ],
      "notes": null
    }
  },
  "payload_encoding": "json",
  "signatures": [
    {
      "key_id": "ed25519:b0865d76d834a52c48506414d16f4e5a",
      "alg": "ed25519",
      "sig": "hMaBHDbe_-R568h96ZRoQSy9Agqp3HWdTpYl6PGrStL46HnKkb1mWmXuG17rVjFVc-iTaoXzFr7lZIXDnJGqAg",
      "signed_at": "2026-09-23T14:11:58Z"
    }
  ]
}

attestation file · issuer keys