Absorb Scoring Rubric: Novel Features
When to read: This file is the scoring rubric for the novel-features subagent (Phase 1.5 Step 1.5c.5). It defines kill/keep checks, the 4-dimension score, the transcendence table format, and reprint verdict rules. The subagent reads this file as part of its operational playbook — the SKILL no longer applies it inline.
Brainstorming flow (personas, candidate generation, gap analysis, cut pass) lives in novel-features-subagent.md. This file owns the static rubric only.
Kill/keep checks
Apply these to every candidate BEFORE scoring. Cut ruthlessly — survivors must be features that can actually ship.
| Check | Kill condition | Keep/reframe action |
|---|---|---|
| LLM dependency | Feature requires NLP, summarization, sentiment analysis, classification, or semantic grouping | Reframe as mechanical: replace "summarize" with "extract top-rated items + stats." Add pipe-friendly output so users can | claude "summarize" themselves. If no mechanical version is useful, cut. |
| External service | Feature requires a service not in the spec (e.g., scraping a website, calling a third-party API not in the brief) | Cut unless the service is free, public, and has no auth. An enrichment API documented in the brief (like OMDb for movie-goat) is fine. |
| Auth the user doesn't have | Feature requires write access, OAuth scopes, or paid tiers the user hasn't confirmed | Gate behind an auth check, or cut if the feature is useless without it. Read-only features using the same auth as other commands are fine. |
| Scope creep | Feature is really an application, not a command. Would take >200 lines to implement, needs a TUI, or requires persistent background processes. | Descope to the one-command version. "Dashboard" → "summary stats." "Monitor" → "poll once with --watch." If the one-command version isn't useful, cut. |
| Verifiability | Feature can't be tested in dogfood. No way to verify the output is correct without manual inspection or domain expertise. | Flag as low-confidence. Keep only if the value is high enough to justify manual QA. |
| Reimplementation | Feature synthesizes API responses locally instead of calling the API. Hand-rolled response builders, hardcoded JSON returned as an "API result," endpoint stubs that return constants, or aggregations computed in-process when the API has an aggregation endpoint. | Cut or rewrite. A printed CLI that pretends to call the API is strictly worse than the API call it replaces. Local SQLite commands (stale, bottleneck, health, reconcile) are local-data commands, not fake API calls. Curated static-reference commands must use // pp:novel-static-reference. Commands that call a real API through a wrapper dogfood cannot see must use // pp:client-call; never use that directive for hardcoded payloads or fake stubs. Dogfood's reimplementation_check enforces this at generation time. |
| Credential getenv | Novel command that calls the API reads the auth env var with os.Getenv instead of config.Load + AuthHeader() / novelAuthHeader |
Rewrite. Endpoint commands already use that seam; getenv misses credentials saved by auth login / auth set-token. Check resp.StatusCode before decoding the body so a plain-text 401 is not reported as a JSON syntax error. Dogfood flags os.Getenv("<AUTH_ENV>") in novel command files. |
Buildability proof. For each surviving feature, write one sentence:
"This uses [specific API endpoint or local data] to compute [specific output] with no external dependencies."
If you can't write that sentence, the feature fails the vet.
Score (4 dimensions, raw 0-10)
Apply to candidates that survive the kill/keep checks.
| Dimension | Points | Scoring |
|---|---|---|
| Domain Fit | 0-3 | 3=core to this API's power users, 2=useful but niche, 1=tangential, 0=wrong domain |
| User Pain | 0-3 | 3=research surfaced explicit demand (community complaints, competitor gap), 2=implied need, 1=speculative, 0=no evidence |
| Build Feasibility | 0-2 | 2=API endpoint + local data covers it, 1=needs minor data model additions, 0=requires new infrastructure |
| Research Backing | 0-2 | 2=evidence from 2+ sources in Phase 1/1.5 research (web search, community issues, MCP source, DeepWiki analysis each count as 1 source), 1=evidence from 1 source, 0=invented |
Normalize: score_10 = round(raw / 10 * 10). Include features scoring >= 5/10.
Transcendence Table Format
Survivors render as rows in the absorb manifest's transcendence table:
| # | Feature | Command | Score | Buildability | How It Works | Evidence | Long Description |
|---|---------|---------|-------|--------------|--------------|----------|------------------|
| N | Player comparison | compare "LeBron" "Curry" | 8/10 | hand-code | Joins player_stats + team + season tables in local SQLite | ESPN community requests, espn_scraper lacks cross-player queries | Use this command to compare two players. Do NOT use it for team-level comparisons; use 'team compare' instead. |The "How It Works" column is the buildability proof — one sentence showing the specific API endpoint or local data that powers the feature.
The "Buildability" column tags whether the generator auto-emits the feature
from the spec (spec-emits) or whether the agent must hand-write the Cobra
file plus root.go wiring after generate (hand-code). See Pass 3 question 5
in novel-features-subagent.md for the
classification rules. The Phase Gate 1.5 prose showcase counts hand-code
rows from this column when reading out the hand-code commitment.
The "Evidence" column MUST cite specific findings from Phase 1 or Phase 1.5 research. "Power users would love this" is not evidence.
The "Long Description" column is none unless the feature overlaps with a
sibling command closely enough that an agent may choose the wrong tool. When it
is populated, it is implementation guidance for the hand-coded Cobra Long
field and must reference only command names that survived the Pass 3 cut.
Reprint verdict rules
Reprints occur when a prior research.json exists at
$PRESS_LIBRARY/<api>/research.json (provenance) or under
$PRESS_MANUSCRIPTS/<api-slug>/*/research.json (archived runs). The subagent
loads prior novel_features (planned) and novel_features_built (shipped),
re-scores each against the current personas, and tags every prior feature with
exactly one verdict:
| Verdict | When | Pool action |
|---|---|---|
| Keep | Persona fit, score ≥ 5/10, buildable | Add with prior command reused so the reprint stays compatible |
| Reframe | Right idea, wrong shape — persona fit exists but command/scope drifted | Add with a new command/description; flag the rename |
| Drop | No persona fit, score < 5/10, or unbuildable now | Exclude; record one-line reason for the manifest |
Surface in the manifest. Reprint runs add a Source column to the
transcendence table: prior (kept), prior (reframed from <old-command>), or
new. Below the table, list dropped prior features with their one-line
justifications so the user can override the drop at the Phase 1.5 gate review.
Prior features are never silently absorbed and never silently dropped.