Fieldpacket

A research library

Fieldpacket Deep research for real questions.

A growing library of sourced, revisable research packets on whatever deserved a proper answer: technology, home and tools, buying decisions, places, investigations. A packet is published only when the investigation clears the publishing bar.

Packets published
2
Sources cited
185
Sections
2
Last updated
2026-10-07
All 2 packets

FP-0003TechnologyEvidence reviewTaste = the bottleneck

Making AI Coding Agents Design Genuinely Good Web UI

Why AI-built interfaces all look the same, and the skills, context files, MCP servers and feedback loops that fix it.

Highest-leverage moveVisual feedback loop, 3 CITL cycles
+17.8%
Root causeRLHF mode collapse, α ≈ 0.57–0.65
Typicality bias
Slop prevalenceof 1,590 Show HN pages (+32% mild)
22% heavy
The toolkitSkills → context → libs → MCP → prompt → loop
6 layers
Sources auditedacross 10 parallel subtopics
95

Updated 2026-07-0718 sections10 agents95 sources~7,400 words Open the packet

Sections

Browse by section

The full library

Technology

1 packet

Tools, models and the systems around them, researched past the launch posts.

  1. Making AI Coding Agents Design Genuinely Good Web UI2026-07-07
All in Technology

Choices that need real evidence: specs, testing, where reviewers disagree, and a verdict.

  1. Carbon Steel Pans: A Deep-Dive Buying Guide2026-06-21
All in Buying Decisions

Why a packet exists

The publishing bar

A packet is published because the investigation produced something worth publishing, never researched because a keyword looked worth publishing. Plenty of runs stay private.

The bar

It has to earn a public page

A packet goes public only if it does at least one of these:

Finds
something surprising
Reconciles
sources that conflict
Assembles
information that doesn't exist together anywhere else
Beats
conspicuously bad search results
Decides
a question that needed real evidence
Tabulates
a table or taxonomy worth keeping as a reference

The labels

Say how it was checked

Before anything goes public it gets a publication audit: original contribution, primary-source coverage, single-source dependence, claim strength against evidence, staleness. It comes out publish, keep private, or another research pass.

Evidence
Evidence review: a structured review of published sources
Field
Field tested: the editor has used the thing
Revised
Packets get revision logs, not quiet date bumps
Every factual claim carries an inline citation
Required
Where sources disagree, the conflict is stated, not averaged away
Required
Recency is checked; a stale rave is a lead, not a fact
Required
Thin, article-shaped output
Rejected

Part 2: the tool

The workflow behind every packet

The research-to-report workflow.

Brief your coding agent the way you'd brief a researcher. It decomposes the question, sends parallel agents after sources, audits what came back, and synthesizes a decision-grade report, then publishes it as a designed, self-contained document in a library you host yourself: on your own machine, an intranet, or a site like this one.

Agent research defaults to shallow sourcing and a wall of prose. Fieldpacket assumes you'll use the document to make a decision, possibly months later, possibly on a phone.

Open source · agent-driven

API keys
0
Sources per run
40–100+
Per report
1 file
To install
3 cmds
Method
Parallel research agents
Runs on
Claude Code · Codex
Requires
Node 24+, an agent
Output
Standalone HTML · PDF
License
MIT

Research → report

Two halves, one pipeline

  1. Topicyour brief
  2. Decomposedistinct angles, before any search
  3. Parallel agentsone per subtopic
  4. Auditcounts, quality, contradictions
  5. Synthesizeby theme, cited inline
  6. Buildagainst the house standard
  7. Polisha second agent attacks it
  8. Publishindex, then verify

Half one — the research

An army of research interns

The heavy lifting. It scopes the job by tier, breaks the topic into distinct angles before searching anything, then runs one agent per subtopic — each doing broad searches, deep-reading primary sources, and closing its own gaps.

Scope
Focused (5–7 subtopics) · Standard (7–9) · Comprehensive (9–12)
Angles
Current state, technical depth, comparisons, real-world use, hard data, criticism, economics, future
Audit
Source count and quality, contradictions, thin subtopics — with supplementary agents where it's short
Output
4,000–10,000+ words, 40–100+ cited sources, synthesized by theme rather than stapled together

Half two — the packet

Designed, then adversarially checked

A build pass renders the markdown against a house design standard. Then a second agent attacks the result — and it is genuinely allowed to reject the first agent's work.

Build
The headline finding first, key figures, method metadata, numbered sticky contents, decision components
Polish
Tables usable at 360px, every row and citation checked against the source, AA contrast, reduced motion, print CSS
Publish
Rebuilds the library index, then verifies — a report isn't done until verification passes
Format
One self-contained HTML file. No framework, no build step, nothing to rot

Real output, not mockups

What comes out the other end

Everything below is lifted from the Carbon Steel Pans packet in the library above: its rows, its verdicts, its sources. Designed, self-contained, and useful at a glance.

PanTier / Price (2026)GaugeHandleWatch-out
Matfer Bourgeat Black SteelMid / $100–$145~3.2mmWelded, rivetlessOlder stock warps; 2023 FR recall (contested); price rose
De Buyer Mineral B PROMid / $110–$1453.0mmRiveted stainless hollowHeaviest; still riveted
Lodge Carbon SteelBudget / $39.95–$44.90~2.39mmRiveted steel (flat)Rough, sticks 2–3 wks; heavy; thin gauge
Darto N27 / N30Value-premium / ~$70–$90 +ship3mm / 4mmWelded, rivetlessShips from Argentina; no warranty; season at home
Decision components. Comparison matrices, verdict cards, and itemised figures with dotted leaders, chosen to fit the content rather than applied as decoration. Here: four of the ten pans and five of the twelve columns in the master table. See the whole table

Overall winner: Matfer Bourgeat Black Steel (11" or 11¾")

Buy if you want one do-everything pan and cook on any surface. Don’t buy if you need a cool-touch handle out of the box, or the contested 2023 French recall bothers you.

The one to think twice about: Made In

Buy if out-of-box readiness, DTC support, and a 60-day trial outweigh the last 15% of performance for you. Don’t buy if you’ll season a raw pan.

The packet. It opens with a clear verdict and the research trail behind it, not a throat-clearing introduction, and every pick says who it is for and who should walk away.

Measured handle temperatures

De Buyer Mineral B Pro, stainless hollow
79°F
De Buyer Blue
88°F
De Buyer Mineral B, iron
95°F
Matfer & Made In, flat iron/steel
88–95°F
Itemised figures. Label, dotted leader, figure: the estimate line the house standard calls for.
At 390px. The polish pass rejects any report whose tables overflow on a phone. This is the same document, unmodified — because a reference you can't read in the car isn't a reference.
  1. 01
    America’s Test Kitchen: The Best Carbon-Steel Skillets independent testing, gauge/design criteria, Made In “Recommended with Reservations,” Mauviel top pick.
  2. 02
    Prudent Reviews: Carbon Steel Cookware Pros and Cons limitations, acid reactivity, weight/heat data.
  3. 03
    Made In: Carbon Steel vs. Cast Iron material comparison and use-case split.
The audit trail. Underneath every packet: how it was researched and how many sources stand behind it, each annotated with what it contributed. All 66 entries
  • Tablesrestack as label and value cards on a phone, first column pinned on wider screens
  • Contentscollapses to a tap-open drawer
  • Motionhonours prefers-reduced-motion
  • Printown stylesheet, source URLs expanded

How every report is built

The design standard

The design brief isn't a suggestion — it's a spec with rejection criteria, which is why output stays consistent instead of drifting into whatever the model felt like that day.

Rejection criteria

A report goes back to polish if it reads as a plain article, if it carries fewer than two decision components, if a table overflows on a phone, or if any source link, price, or row from the research went missing. Thin output is a failure even when the content is correct.

Distinct components beyond plain type and tables
8 minimum
Responsive floor — tables must stay usable
360 px
Source substance preserved from the research
Preserved
External runtime dependencies in the output
None

No API keys, ever

Install and brief it

Your coding agent is the engine — Claude Code, Codex, whatever you already pay for. There's no LLM SDK, no key to paste, and no per-report cost beyond the subscription in your pocket. Requires Node 24+.

$ git clone https://github.com/buddyh/fieldpacket.git
$ cd fieldpacket
$ ./install.sh

Links the skills and workflows into your agent, creates ~/fieldpacket.

Then talk to it like a person

"I want a deep dive on carbon steel pans for a home cook. I care about induction compatibility, how much seasoning babysitting they really need, and whether the expensive ones are meaningfully better than a $40 Lodge. Skip anything that reads like an affiliate listicle — I want real sources and I want to know where they disagree. Publish it to my library when it's done."

The more you say about what you actually care about, the better the report. It scopes the job, shows you the plan, and you steer from there — adjust the angles, push back, ask for another section, all in the same conversation.

What the CLI handles

The deterministic half

The agent researches and designs. Everything that should be repeatable rather than improvised is a plain command.

Command
What it does
fieldpacket init
Create the library
fieldpacket index
Rebuild the browsable index from the metadata sidecars
fieldpacket serve
Serve the library locally, loopback by default
fieldpacket list
List published reports, newest first
fieldpacket verify
Check the library, or one report, is well formed
fieldpacket pdf
Render a faithful single-page PDF, exactly as it looks
fieldpacket doctor
Check the environment

Your library is just files

~/fieldpacket/
├── source/<slug>.md      the research markdown
├── reports/<slug>.html   the published field packet
├── meta/<slug>.json      sidecar the index reads
├── pdf/                  rendered PDFs
└── index.html            the browsable library

Static files, so you can host them anywhere — or keep the whole thing on your own machine and never publish a word of it.

Your library index has live search, tag chips with counts, and a density toggle — all client-side, and it still works with JavaScript off.

Questions

Reasonable questions

Do I need an API key?

No — and there's nowhere to put one. Fieldpacket runs inside your coding agent's session, so the research and the design are done by the subscription you already have. Nothing calls an LLM API directly.

What does it actually install?

Two skills (the research method and the publishing workflow) linked into your agent, two workflow scripts, and a small TypeScript CLI. No build step and no runtime dependencies — Node 24 runs the TypeScript directly.

Is the research any good, or is it confident nonsense?

It's as good as the sources it finds, and the method is built to make that auditable rather than to hide it. Every claim is cited inline, the source list is annotated with what each one contributed, and where sources conflict the report says so instead of splitting the difference. You can check its work — that's the point of the audit trail.

What topics can it handle?

Anything you'd research properly: buying decisions, technical landscapes, places and itineraries, market briefs, playbooks. The research skill adapts its subtopics to the question; the design standard doesn't care what the subject is. The library above is the mixed bag to prove it.

Does every run end up in this library?

No. Your runs stay in your own library unless you choose to publish them. The packets here are the editor's, and each one had to clear the publishing bar first; most research never does.

Does my research leave my machine?

Only insofar as your agent searches the web, same as any research session. The library itself is local files. The built-in server binds to loopback, and nothing is published anywhere unless you choose to upload it.

Read one before you install anything.