rank-sim-idea

name

rank-sim-idea

description

Triage a proposed Clowder simulation idea with the V×F×R×C×H rubric before writing code. Use whenever the user proposes adding a new system, feature, mechanic, creature, item, or behavior to the sim. Produces a score, bucket, and recommendation anchored against the shadowfox calibration — prevents narratively-cool, ops-expensive ideas from slipping in unpriced. Trigger phrases - "add X to the sim", "what if cats could Y", "new system for Z", "I want a feature that...", "we should have a <creature/item/mechanic>". Do NOT fire on balance tweaks to existing systems, bug fixes, or refactors.

Rank a Sim Idea

Clowder's backlog gets hit constantly with new-feature proposals. Ideas are not equal-cost (Tynan Sylvester, Designing Games). A rubric that prices only implementation cost lets shadowfox-class ideas — narratively brilliant, operationally expensive forever — slip in unpriced. This skill runs the triage before code gets written, so the user sees the true cost while they still own the decision.

When to fire

Fire when the user proposes something new for the simulation:

• "Add a crow species that steals shiny items"
• "What if cats could dig burrows"
• "I want a weather event where…"
• "We should have a [new creature / item / mechanic / behavior]"
• Any docs/systems/*.md stub that doesn't yet exist

Do NOT fire when:

• The user is tuning an existing system ("bump mentor_warmth_diligence_scale") — that's balance work, covered by CLAUDE.md §"Balance Methodology".
• The user is fixing a bug or refactoring.
• The idea is already in docs/systems/*.md or docs/open-work.md — that work is priced by the existing indexes. Redirect them there.
• The idea is about the AI substrate (cluster A1–A4). That has its own planning surface in docs/open-work.md.

The rubric

Five axes, each scored 1–5. Multiplied, not summed — a low score on any axis sinks the idea, because one-dimensional wins don't ship.

Implementation cost (C) and simulation-health tax (H) are separated deliberately: cheap to build and expensive to live with are independent failure modes, and conflating them hides the second one until it's too late. Shadowfoxes are the case study.

V — Value

How much this moves the needle on the three continuity canaries (project-vision.md §"Continuity canaries": generational, ecological variety, mythic texture) and §5 sideways-broadening (grooming, play, courtship, burial, preservation, generational knowledge).

• 5 — Directly lights a currently-zero canary, or unblocks multiple §5 axes.
• 4 — Strong §5 alignment; lights a partially-passing canary.
• 3 — Supports the thesis without moving a canary.
• 2 — Orthogonal polish; helps one subsystem but doesn't change what the world looks like.
• 1 — Tooling or diagnostic only; no in-world effect.

F — Fit

Concordance with project-vision.md: honest world, no director, ecology-with-metaphysical-weight, emergent complexity, survival-lock as a bug.

• 5 — The idea is the thesis (e.g. The Calling, generational knowledge). A DF-style beer-cats-puke-depression chain reaction.
• 4 — Clear thesis fit, no tension.
• 3 — Compatible but could be misbuilt into a director-ish system.
• 2 — In tension with the thesis if not tuned carefully.
• 1 — Fights the thesis (would need to be reframed to ship).

R — Risk (higher = safer)

Probability it works as predicted without regressing canaries on the first implementation.

• 5 — Isolated; no regression surface; observable outcome.
• 4 — Extends an existing subsystem; regression scope bounded.
• 3 — Touches scoring or GOAP; requires A/B verification per CLAUDE.md §"Balance Methodology".
• 2 — Architectural; could produce flipper or second-order effects.
• 1 — Hypothesis-level; unclear whether the desired behavior is reachable with the current substrate.

C — Implementation Cost (higher = cheaper)

One-time engineering effort, inverted.

• 5 — ≲300 LOC, one file, no new messages/components.
• 4 — 300–700 LOC, extends existing systems.
• 3 — 700–1.2k LOC, new ECS subsystem but no GOAP touch.
• 2 — 1.2k+ LOC, GOAP or coordination rework.
• 1 — Multi-cluster; gated on A1 IAUS refactor or requires new architecture above GOAP.

H — Simulation-Health Tax (higher = lower tax)

The ongoing cost of living with the feature. Tuning cycles, canaries it forces us to maintain, interaction surface with existing scoring, destabilization of unrelated canaries. The shadowfox axis.

• 5 — Zero ongoing tax. No new canary, no scoring interaction, no tuning slot in balance soaks.
• 4 — Local tax. One or two constants to keep tuned; no new canary.
• 3 — Moderate. A new measured metric but not a hard gate; touches one existing balance axis.
• 2 — High. Forces a new canary or regularly destabilizes an existing one; shows up in every balance diff.
• 1 — Shadowfox-class. Dedicated canary + defense pipeline + permanent presence in every tuning session.

Shadowfox calibration anchor

Every score is read relative to shadowfoxes, which shipped and are expensive forever. Scoring shadowfoxes retrospectively:

Axis	Score	Reason
V	5	Mythic-texture canary pillar; fog-bound corruption-born predator is the thesis in one creature.
F	5	Ecology-with-metaphysical-weight in one line.
R	2	Fear/ward/flee interaction with scoring destabilized mortality; required a bespoke canary to even detect misbehavior.
C	3	Built, but significant — the ambush + corruption-spawn pipeline contributed meaningfully to wildlife.rs (2.5kloc).
H	1	Dedicated canary (ShadowFoxAmbush ≤ 5 in CLAUDE.md), defense pipeline stub (docs/systems/shadowfox_wards.md), perpetual tuning slot. Maximum ongoing tax.

Score: 5 × 5 × 2 × 3 × 1 = 150 → "expensive but valuable" bucket. Matches lived experience: shipped, narratively load-bearing, permanently expensive to keep tuned.

Any proposal scoring V=5/F=5 but H=1/R=2 is asking to become the next shadowfox. Surface that explicitly in the output.

How to estimate H at proposal time

H is the hardest axis to score and the whole reason this skill exists. Triangulate three sources — state which drove the score in the output.

(a) Structural tells

The proposal is at high risk of H=1–2 if it exhibits two or more of:

• Introduces a new entity type that reads and writes existing scoring axes (fear, wards, flee, hunger, bond) rather than living in its own isolated axis.
• Probabilistic rare-event existence with cascade consequences ("spawns occasionally, but when it does, the colony reshapes").
• Creates a feedback loop coupling two or more existing subsystems (shadowfox: corruption → spawn → ward-building → corruption pushback → spawn rate).
• Forces a bespoke canary because the failure mode isn't legible to existing canaries (shadowfox's ShadowFoxAmbush ≤ 5 exists because generic Starvation = 0 wouldn't have caught it).
• Tunables cross-coupled with other systems' tunables — its constants will show up as variables in future balance threads for unrelated work.

Conversely, an isolated new axis with its own canary-free scoring contribution is H=4–5 (e.g. a new leisure action that scores on an existing mood axis and doesn't feed back).

(b) Memory of priors

Query mcp-memory-service with retrieve_memory, searching for tag clowder + ongoing-tax plus the proposal's structural features. Each shipped system should leave a pattern memory so the next triage is sharper:

name: ongoing-tax-<system>
tags: [clowder, ongoing-tax, pattern]
type: pattern
content:
  System: <name>
  Predicted H at proposal: <n>
  Observed H after N sim-years of tuning: <n>
  Structural tells that fired: [list]
  Tuning iterations consumed: <n> (count from docs/balance/)
  Bespoke canary required: yes/no

Cold start today — the rubric gets sharper with each system priced. After running this skill, if the user proceeds with the idea, commit the memory as part of the idea's landing commit so the prior is captured.

(c) Retrospective grep — concrete evidence beats priors

Before committing H, grep docs/balance/*.md for the nearest-relative shipped system and count iteration threads:

Grep docs/balance/*.md for terms matching the proposal's analogue
(e.g. "fox", "corruption", "ward", "scent", "detection").
Count distinct iteration headers ("Iteration 1", "Iteration 2", ...).

A proposal analogous to a system that burned 3+ balance iterations is probably H≤2. Name the specific balance thread(s) in the output.

Score buckets

• >1000 — cheap win. Pick up next session.
• 300–1000 — worthwhile; plan carefully.
• 80–300 — expensive but valuable; earn the slot. Requires hypothesis + prediction per CLAUDE.md §"Balance Methodology".
• <80 — defer unless a dependency forces the hand, or reframe to raise a low axis before reconsidering.

Output format

Produce exactly this block — the user will paste it into docs/open-work.md if the idea lands:

## Rank: <idea name>

| V | F | R | C | H | Score | Bucket |
|---|---|---|---|---|-------|--------|
| n | n | n | n | n | NNN   | bucket |

**Justifications**
- V: ...
- F: ...
- R: ...
- C: ...
- H: ... (H-source: structural tells / memory / balance-grep; cite specifics)

**Dependencies:** gated on A1 IAUS refactor? other stubs?

**Canary hit:** which currently-zero continuity canary this lights
(generational, ecological variety, mythic texture), or "none".

**Shadowfox comparison:** one sentence — would this become the next
permanent-tax item? Point to the structural tells that fired or didn't.

**Recommendation:** pick up / plan carefully / earn the slot / defer.
If "earn the slot", name the hypothesis + prediction the user will
need to write per balance methodology.

What this skill does NOT do

• Score ideas already stubbed in docs/systems/ or docs/open-work.md. Those are priced by the existing backlog indexes. Redirect the user to those files. If the user explicitly requests a retroactive score (audit), that's fine — note it as retrospective in the output.
• Score refactors of the AI substrate (A1–A4). Those have their own planning surface (docs/systems/ai-substrate-refactor.md, docs/open-work.md §§5).
• Make the decision. Produce the triage artifact; the user decides whether to proceed, reframe, or defer.

References for the skill (kept here so the rubric doesn't rot)

• docs/systems/project-vision.md — §5 sideways-broadening, the continuity canaries, the ecological-magical-realist thesis. V and F axes quote from here.
• CLAUDE.md §"Balance Methodology" — the hypothesis / prediction / observation / concordance format required for ideas in the 80–300 bucket.
• CLAUDE.md §"Long-horizon coordination" — where to file the output if the idea lands.
• docs/wiki/systems.md — current stub inventory; cross-check before scoring to confirm the idea isn't already priced.
• docs/open-work.md — in-flight thread index; same cross-check.