Canon Evaluation
The rubric that replaces "this looks good." Every dimension is scored 1–5
against stated anchors, every score carries a one-line piece of evidence, and
the whole thing survives an adversarial pass that is actively trying to kill it.
Two rules make this worth running:
- Evidence or no score. A dimension scored without a quoted line, a file
path, or a command output is unscored. Write
unscored — no evidence rather
than guessing; an honest gap is information, an invented 4 is noise.
- The adversary goes last and wins ties. If the refutation pass lands a real
hit, the artifact is revised. You do not get to out-argue it in the summary.
The ten dimensions
Score 1–5. Anchors are deliberately unsentimental: 3 is competent, 5 is rare.
| # |
Dimension |
1 |
3 |
5 |
| 1 |
Canon fit |
Contradicts LOCKED canon |
Consistent, sits beside existing systems |
Consistent and makes an existing locked system make more sense |
| 2 |
Mechanism |
Vibes; nothing works a particular way |
Rules stated, mostly consistent |
A reader can predict outcomes and be right; the rule constrains the author too |
| 3 |
Cost |
Power with no price |
A price is named |
The price is paid by someone specific, on the page, and it hurts |
| 4 |
Consequence |
Changes nothing |
Changes one situation |
Reframes something the reader already believed |
| 5 |
Naming |
Register-wrong or collides |
Register-correct, collision-checked |
Register-correct, etymology carries meaning, sounds inevitable aloud |
| 6 |
Voice |
AI-slop present; register blended |
Clean, in-register |
Prose a reader would quote from memory |
| 7 |
Viewpoint |
Single omniscient assertion |
One narrator, acknowledged bias |
Three viewpoints that disagree, and the disagreement is the point |
| 8 |
Mystery budget |
Everything explained, or nothing is |
Withheld deliberately, logged |
Withheld, logged, and the withholding actively generates story |
| 9 |
IP distance |
Recognizable as franchise X, renamed |
Architecture borrowed, bricks original |
Synthesized across sources into something with its own logic |
| 10 |
Story surface |
A closed fact |
Implies one scene |
Implies conflicts, factions, and questions others can build on |
Hard gates — any one of these blocks staging regardless of composite:
- Dimension 1 scores 1 → contradicts the vault. Not a revision; a rewrite.
- Dimension 9 scores 1 → IP exposure. Rewrite or drop.
- Dimension 5 scores 1 → renaming touches everything downstream. Fix before anything else.
node .claude/ci/lore-lint.mjs <files> exits non-zero → fix first, score after.
Composite bands (sum of 10 dimensions, max 50):
If any dimension is unscored, the verdict is incomplete — evidence required.
Do not compute a composite, do not stage, do not promote. Otherwise "evidence or
no score" becomes a way to buy a band: omit evidence for the one weak dimension,
sum the other nine, and a 38/45 reads like a 38/50. An unscored dimension is a
missing answer, not a zero and not an exemption.
| Band |
Verdict |
| 42–50 |
Stage it. Nominate for /lock-decision if it closes a known gap. |
| 34–41 |
Stage as STAGING with the weak dimensions named in the STAGING LOG. |
| 25–33 |
Revise before staging. Name the two lowest dimensions and fix those. |
| ≤24 |
Not ready. Return to phase 2 of the gate and pick a different pattern. |
The adversarial pass
After scoring, argue the other side. Four lenses, each asked to succeed at
killing the work, not to be fair to it:
- The canon lawyer — find the contradiction. Not with the vault (the lint
covers that), but with STAGING content, with the
book/ corpus, with
something established three files away. Grep before concluding there is none.
- The IP auditor — name the franchise this is closest to, and state the
specific element. If the honest answer is "this is X's mechanic with new
nouns," say so. Closeness to one source is the danger; synthesis across
several is the defense.
- The bored reader — why would anyone care? What question does this make
them want answered? If the answer is "it's lore, it's just true," the artifact
is inert and dimension 10 is a 1 regardless of what was scored.
- The continuity auditor from three years out — what does this make
impossible later? What does it force every future writer to work around?
Cheap now, expensive forever is the failure mode worth catching early.
A lens that lands a hit produces a revision, not a footnote. Record hits that
were considered and deliberately accepted — with the reason — so the next
reviewer does not re-litigate them.
Output format
### Canon Evaluation — <artifact name>
**Composite: NN/50 — <verdict band>**
Lint: `lore-lint.mjs` exit 0 (N files)
| # | Dimension | Score | Evidence |
|---|---|---|---|
| 1 | Canon fit | 4 | Uses the locked Starweave Gate (852 Hz) without redefining it — CANON_LOCKED.md:62 |
| … | | | |
**Adversarial pass**
- Canon lawyer: <hit, or "no hit — checked FACTIONS.md, STORY_ENGINE.md, book/legends">
- IP auditor: <closest source + specific element + why the distance holds>
- Bored reader: <the question it makes a reader want answered>
- Continuity auditor: <what it forecloses>
**Verdict**: <stage / revise / rewrite> — <the two lowest dimensions and the fix>
Scoring the world, not the artifact
For whole-system reviews — "how good is Arcanea's magic system now" — use the
twelve-dimension world scorecard in docs/worldbuilding/research/SYNTHESIS.md
§2 instead, which benchmarks against Tolkien, Elder Scrolls, FF, Marvel, HP,
Warcraft, and the anime canon. That file lands in a follow-up PR; until it does,
score per-artifact with the rubric below and defer the whole-system verdict
rather than improvising a scorecard. That scorecard is the strategic instrument; this
rubric is the per-artifact one. Re-score the world scorecard when a Ten Upgrade
lands, and record the movement — an upgrade that does not move a dimension
either was not needed or was not finished.
1---2name: canon-evaluation3description: Score any piece of Arcanea lore against a ten-dimension rubric with an adversarial refutation pass, before it is staged or promoted. Use when evaluating a new character, god, godbeast, creature, artifact, faction, place, era, magic mechanism, Library text, legend, or parable — and whenever asked whether lore is good, canon-consistent, on-voice, ready to stage, or ready for /lock-decision. Produces a scored verdict with per-dimension evidence, not an opinion. Invoked as phase 5 of the lore-release-gate.4---56# Canon Evaluation78The rubric that replaces "this looks good." Every dimension is scored 1–59against stated anchors, every score carries a one-line piece of evidence, and10the whole thing survives an adversarial pass that is actively trying to kill it.1112Two rules make this worth running:1314- **Evidence or no score.** A dimension scored without a quoted line, a file15 path, or a command output is unscored. Write `unscored — no evidence` rather16 than guessing; an honest gap is information, an invented 4 is noise.17- **The adversary goes last and wins ties.** If the refutation pass lands a real18 hit, the artifact is revised. You do not get to out-argue it in the summary.1920## The ten dimensions2122Score 1–5. Anchors are deliberately unsentimental: **3 is competent**, 5 is rare.2324| # | Dimension | 1 | 3 | 5 |25|---|---|---|---|---|26| 1 | **Canon fit** | Contradicts LOCKED canon | Consistent, sits beside existing systems | Consistent *and* makes an existing locked system make more sense |27| 2 | **Mechanism** | Vibes; nothing works a particular way | Rules stated, mostly consistent | A reader can predict outcomes and be right; the rule constrains the author too |28| 3 | **Cost** | Power with no price | A price is named | The price is paid by someone specific, on the page, and it hurts |29| 4 | **Consequence** | Changes nothing | Changes one situation | Reframes something the reader already believed |30| 5 | **Naming** | Register-wrong or collides | Register-correct, collision-checked | Register-correct, etymology carries meaning, sounds inevitable aloud |31| 6 | **Voice** | AI-slop present; register blended | Clean, in-register | Prose a reader would quote from memory |32| 7 | **Viewpoint** | Single omniscient assertion | One narrator, acknowledged bias | Three viewpoints that disagree, and the disagreement is the point |33| 8 | **Mystery budget** | Everything explained, or nothing is | Withheld deliberately, logged | Withheld, logged, and the withholding actively generates story |34| 9 | **IP distance** | Recognizable as franchise X, renamed | Architecture borrowed, bricks original | Synthesized across sources into something with its own logic |35| 10 | **Story surface** | A closed fact | Implies one scene | Implies conflicts, factions, and questions others can build on |3637**Hard gates** — any one of these blocks staging regardless of composite:3839- Dimension 1 scores 1 → contradicts the vault. Not a revision; a rewrite.40- Dimension 9 scores 1 → IP exposure. Rewrite or drop.41- Dimension 5 scores 1 → renaming touches everything downstream. Fix before anything else.42- `node .claude/ci/lore-lint.mjs <files>` exits non-zero → fix first, score after.4344**Composite bands** (sum of 10 dimensions, max 50):4546**If any dimension is `unscored`, the verdict is `incomplete — evidence required`.**47Do not compute a composite, do not stage, do not promote. Otherwise "evidence or48no score" becomes a way to buy a band: omit evidence for the one weak dimension,49sum the other nine, and a 38/45 reads like a 38/50. An unscored dimension is a50missing answer, not a zero and not an exemption.5152| Band | Verdict |53|---|---|54| 42–50 | Stage it. Nominate for `/lock-decision` if it closes a known gap. |55| 34–41 | Stage as STAGING with the weak dimensions named in the STAGING LOG. |56| 25–33 | Revise before staging. Name the two lowest dimensions and fix those. |57| ≤24 | Not ready. Return to phase 2 of the gate and pick a different pattern. |5859## The adversarial pass6061After scoring, argue the other side. Four lenses, each asked to *succeed* at62killing the work, not to be fair to it:63641. **The canon lawyer** — find the contradiction. Not with the vault (the lint65 covers that), but with STAGING content, with the `book/` corpus, with66 something established three files away. Grep before concluding there is none.672. **The IP auditor** — name the franchise this is closest to, and state the68 specific element. If the honest answer is "this is X's mechanic with new69 nouns," say so. Closeness to *one* source is the danger; synthesis across70 several is the defense.713. **The bored reader** — why would anyone care? What question does this make72 them want answered? If the answer is "it's lore, it's just true," the artifact73 is inert and dimension 10 is a 1 regardless of what was scored.744. **The continuity auditor from three years out** — what does this make75 impossible later? What does it force every future writer to work around?76 Cheap now, expensive forever is the failure mode worth catching early.7778A lens that lands a hit produces a revision, not a footnote. Record hits that79were considered and deliberately accepted — with the reason — so the next80reviewer does not re-litigate them.8182## Output format8384```markdown85### Canon Evaluation — <artifact name>86**Composite: NN/50 — <verdict band>**87Lint: `lore-lint.mjs` exit 0 (N files)8889| # | Dimension | Score | Evidence |90|---|---|---|---|91| 1 | Canon fit | 4 | Uses the locked Starweave Gate (852 Hz) without redefining it — CANON_LOCKED.md:62 |92| … | | | |9394**Adversarial pass**95- Canon lawyer: <hit, or "no hit — checked FACTIONS.md, STORY_ENGINE.md, book/legends">96- IP auditor: <closest source + specific element + why the distance holds>97- Bored reader: <the question it makes a reader want answered>98- Continuity auditor: <what it forecloses>99100**Verdict**: <stage / revise / rewrite> — <the two lowest dimensions and the fix>101```102103## Scoring the world, not the artifact104105For whole-system reviews — "how good is Arcanea's magic system now" — use the106twelve-dimension world scorecard in `docs/worldbuilding/research/SYNTHESIS.md`107§2 instead, which benchmarks against Tolkien, Elder Scrolls, FF, Marvel, HP,108Warcraft, and the anime canon. That file lands in a follow-up PR; until it does,109score per-artifact with the rubric below and defer the whole-system verdict110rather than improvising a scorecard. That scorecard is the strategic instrument; this111rubric is the per-artifact one. Re-score the world scorecard when a Ten Upgrade112lands, and record the movement — an upgrade that does not move a dimension113either was not needed or was not finished.
frankxai