---
title: "Decision patterns from the community (with fit verdicts)"
type: community
source_tier: community
tags: [ideas, patterns, index, fit-verdicts, community]
created: 2026-09-20
updated: 2026-09-21
confidence: medium
sources:
  - raw/community/diogo-almeida-coding-agent-notes.md
  - raw/x/MiguelriosEN-2101029313906987422.md
  - raw/nate/jev-shaped-problems.md
  - raw/x/INDEX.json
  - raw/x-repos/INDEX.json
  - raw/x/k2sbhai-2101657436696547773.md
jev_version: "jev-1.13.0"
summary: "Index of all 37 community decision patterns (ID, primitive, fit verdict, page), routing by project type to one of four domain pages, retrofit order, and anti-patterns."
---

# Decision patterns from the community (with fit verdicts)

> **TL;DR** 37 decision shapes people published for Jev, digested from curated X posts and the repos they link. This page is the index: find the rows that look like judgments your project already makes, then open the ONE domain page that holds them for the full state / ask / code / fit / map detail. Pattern IDs are permanent.

## Route by project type

| Your project is mostly... | Open | ~Tokens |
|---|---|---|
| an AI agent, coding tool, context or memory layer, model/tool/skill routing | [[ideas/patterns-agents]] (P01-P11, P35+) | 3-4k |
| browser, desktop, mobile or voice automation; a judgment inside a product UI | [[ideas/patterns-interface]] (P12-P14, P34) | 1.5k |
| evals and guardrails, search/rerank, corpus labelling, documents, transcripts, games/IoT, markets | [[ideas/patterns-data]] (P15-P18, P24-P27) | 3k |
| marketing, sales, GTM, content, support, approvals, back-office ops | [[ideas/patterns-business]] (P19-P23, P28-P33) | 4.5k |

Spans two? Read the one holding most of your candidate rows; open the second only for a specific Pxx.

## Route by where Jev sits

A second way in, from [Nate B. Jones's member guide](https://unlock-ai.natebjones.com/guides/jev-shaped-problems) (used with his permission): four placements of a classifier in a system. He also publishes a build-ready recipe-card prompt for each, plus a read-only audit prompt; they are his paid material, so follow the link rather than expecting them here.

| Placement | Our patterns |
|---|---|
| 1. Intake router: between messy incoming items and software that already knows what to do next | P22, P19, P23, P26, P16 |
| 2. Bulk classification as a tool for a reasoning model | P18, P27, P15, P20, P21 |
| 3. Outer loop of an agent; an LLM is one of the actions it can pick | P04, P12, P13, P14, P03, P02, P35, P36, P37 |
| 4. Inside an interface element | P34 |

## All patterns

**Fit** is our verdict against [[concepts/jaggedness-jev-1-13]] and [[concepts/system-one]]: `strong` = bounded options, semantic judgment, code owns the rest; `workable` = real, with a stated mitigation; `poor` = a documented weakness (numbers, dates, time series, generation, huge state) sits in the middle of it. `strong*` = strong only in the narrow per-item form; the domain page says which form is poor. **Page**: agents = [[ideas/patterns-agents]], interface = [[ideas/patterns-interface]], data = [[ideas/patterns-data]], business = [[ideas/patterns-business]].

| ID | Decision pattern | Primitive | Fit | Page |
|---|---|---|---|---|
| P01 | Decisions wearing a generation costume | Choice/Noul | strong | agents |
| P02 | Model and effort routing | Choice | strong | agents |
| P03 | Tool-risk / irreversible-action gate | Noul/Score | strong | agents |
| P04 | Loop controller: done, stuck, recover | Choice/Noul | strong | agents |
| P05 | Worker routing and branch pruning | Choice/Score | strong | agents |
| P06 | Confidence-gated human queue | Noul | strong | agents |
| P07 | Compaction: what to drop, and when | Noul | strong | agents |
| P08 | Skill, rule and context selection per turn | Choice/Noul | strong | agents |
| P09 | Memory write/update/delete policy | Choice | workable | agents |
| P10 | Staged diff review and semantic lint | Noul/Score | workable | agents |
| P11 | Test-output interpretation | Choice/Noul | workable | agents |
| P12 | Next browser action from a code-built menu | Choice | strong | interface |
| P13 | Picking a site-exposed tool instead of a click | Choice | strong | interface |
| P14 | Desktop, mobile, voice and form control | Choice | workable | interface |
| P15 | Jev-as-a-judge for agent evals | Noul/Score | strong | data |
| P16 | Screening what goes in and what comes out | Choice/Noul | workable | data |
| P17 | Per-candidate relevance: rerank, SQL predicate, graph edge | Choice/Noul | strong | data |
| P18 | Map-reduce labelling over a corpus or a timeline | Choice/Noul/Score | strong | data |
| P19 | Lead and ICP scoring, fit kept apart from intent | Choice/Score | strong | business |
| P20 | Signal detection and reply prioritisation | Noul/Score | strong | business |
| P21 | Creative scoring, ad teardown and content dedupe | Choice/Noul | strong* | business |
| P22 | Inbound triage fan-out: tickets and email | Choice/Noul/Score | strong | business |
| P23 | Policy-bounded approval triage (refunds, incidents, discounts) | Choice/Noul/Score | workable | business |
| P24 | Action selection from simulator or device state | Choice/Noul | workable | data |
| P25 | Per-block buy/sell decision | Choice | poor | data |
| P26 | Document classification and packet splitting | Choice | strong | data |
| P27 | Moment finding over a transcript | Noul/Score | workable | data |
| P28 | Next-best action on a deal or account | Choice | workable | business |
| P29 | Churn and retention intervention triage | Choice/Noul/Score | workable | business |
| P30 | Draft QA against voice, brand and claim rules | Noul/Score | strong* | business |
| P31 | Brief, idea and news-response triage | Choice/Noul/Score | strong | business |
| P32 | SEO page review and internal-link checks | Noul/Score | workable | business |
| P33 | Metric-driven ad, budget and timing decisions | Choice | poor | business |
| P34 | Classifier inside an interface element | Choice/Noul/Score | strong | interface |
| P35 | Security-aware routing | Choice/Noul | workable | agents |
| P36 | Background read-only task triage | Choice/Noul/Score | strong | agents |
| P37 | Self-compiling workflow: agent does the job, then compiles it | Choice/Noul | strong | agents |

## Retrofitting an existing agent

- **Start order** ([@k2sbhai](https://x.com/k2sbhai/status/2101657436696547773)): smallest repeated decisions first — context pruning (P07), model routing (P02), tool/skill selection (P08), supervision (P04), review gates (P10), browser/computer actions last (P12, P14). Measure cost, latency and error rate before and after; cheaper with more bad decisions is a failed optimisation.
- **Business workflows**: find the expensive queue of incoming information and put Jev at the front of it (Ryan Vogel / Greg Isenberg, [@startupideaspod](https://x.com/startupideaspod/status/2101029801821941933)) — P22, P19.
- **Vet community repos first**: see the caveats in [[ideas/community-repos]].
- Read-only audit prompt for an existing codebase: [[reference/agent-skill]].

## Anti-patterns seen in the wild

- **"It can't hallucinate."** It cannot break your schema; it can still pick a wrong valid option confidently — [[concepts/jaggedness-jev-1-13]], [[concepts/confidence]].
- **Asking Jev for its own confidence.** `confidence` comes back on Choice and Score; Noul has none. Walkthroughs also invent the types `boolean`/`categorical` and a `context` field; the real fields are `state` and `questions` — [[reference/http-api]], [[reference/openapi-schemas]].
- **One global threshold, or a Noul threshold reused on a Choice.** Structural invariants between question types are not guaranteed — [[concepts/jaggedness-jev-1-13]].
- **Overlapping Choice options.** @cyrilXBT reports about a third of answers were inconsistent until four overlapping categories were rewritten — criteria, not model ([[guides/writing-instructions-and-criteria]]).
- **Review velocity, "last 30/60/90 days", settlement windows.** Date ordering is a documented failure mode; extract parts, compute in code ([[cookbooks/date-extraction]]).
- **Dumping a session, sitemap or corpus into one state.** 64k per request, 32k for state plus the longest question, and accuracy falls as irrelevant detail grows ([[concepts/state]]).
- **Asking it to explain, summarise or write the reply.** Not trained to generate; chaining choices to fake it is slow and bad ([[concepts/jaggedness-jev-1-13]]).
- **Counting.** "How many of these failed or mention X" — one question per item, add up in code.

## Related

- [[ideas/consult]] — running the "could Jev help this project?" conversation
- [[ideas/patterns-agents]], [[ideas/patterns-interface]], [[ideas/patterns-data]], [[ideas/patterns-business]] — the patterns themselves
- [[ideas/field-reports]] — independent measurements behind the numbers quoted in patterns
- [[ideas/community-repos]] — index of the repos named in *Seen* lines; [[ideas/overview]] — what this tier is
- [[patterns/overview]], [[cookbooks/overview]], [[concepts/use-case-map]], [[guides/choosing-a-primitive]]

## Sources

Post and repo links are inline in each pattern's *Seen* line on the domain pages; each domain page lists its raw captures in frontmatter (`raw/x/`, `raw/x-repos/` in the private repo; indexes `raw/x/INDEX.json`, `raw/x-repos/INDEX.json`).