Skip to content
Merged
Show file tree
Hide file tree
Changes from all commits
Commits
File filter

Filter by extension

Filter by extension


Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
19 changes: 19 additions & 0 deletions CHANGELOG.md
Original file line number Diff line number Diff line change
@@ -1,5 +1,24 @@
# Changelog

## 1.12.0

- **Papers drive the work; the record names no paper** (method rule 8, model section 4). A lab that writes
papers keeps one overview page of the whole trail and one page per active paper, written as that paper's
argument and opening with a scope table: audience, thesis, experiments included, excluded or superseded, and
open. The papers' open rows and their reviewers' criticisms say what is proposed, ordered, parked or dropped,
and a result revises the overview and every paper it touches. So far gives the current finding, never the
story of the bugs on the way. Adding, merging or retiring a paper is direction.
- **`[papers]` in `lab.toml`** names the overview and the active paper pages, as library paths. It is optional:
a lab without it is unchanged. With it, each paper page marks its scope cells `data-scope="included"`,
`"excluded"` and `"open"`, and `lab check` refuses an id that is not in the lab, a step under included that is
not done or is void, a step under open that is finished, a question or goal under open that is done, a parked
or dropped question or goal in either, and a goal, question or step listed in two rows of one page. It notes a
paper page committed before a result about something it lists under included or open.
- **`lab now` lists the papers**: each active paper with the open work its scope names, then the live steps
that no paper lists, by themselves or by their question or goal.
- The skills `lab-set-up`, `lab-work`, `lab-talk` and `lab-review` and the method changed: run `lab init` in a lab
after upgrading.

## 1.11.1

- **A run in progress in another checkout no longer fails `lab check`.** A run stays in the `.lab/running` of the
Expand Down
2 changes: 1 addition & 1 deletion README.md
Original file line number Diff line number Diff line change
Expand Up @@ -56,5 +56,5 @@ where folio serves the lab, which brings `folio-kb` 0.5.0 or later, and list `la
## Read more

- [docs/model.md](docs/model.md): the contract. What you read, how a number becomes a result, who decides, the commands, the files.
- [method/METHOD.md](method/METHOD.md): the seven rules every lab keeps.
- [method/METHOD.md](method/METHOD.md): the eight rules every lab keeps.
- [skills/](skills/): lab-set-up, lab-work, lab-review, lab-talk and lab-mission. [agents/](agents/): the reviewer.
65 changes: 62 additions & 3 deletions docs/model.md
Original file line number Diff line number Diff line change
Expand Up @@ -255,7 +255,7 @@ in points. When a named result belongs to a replaced experiment, the page says s
You decide two things: **direction** and **resources**. You decide them in a discussion with the agent,
not by approving requests.

- **Direction.** The goals and the questions.
- **Direction.** The goals, the questions, and the papers they serve.
- **Resources.** Everything outside the lab that it may touch: what it may use, how much, and what it
may do there.

Expand Down Expand Up @@ -376,6 +376,36 @@ is a resource and a cap, nothing else. A lower cap is the agent's to write. A hi
rule, a new resource or a larger grant comes out of a decision: one that lists that experiment, or
that resource, under `about:`.

**Papers drive the work; the record names no paper.** A lab that writes papers keeps two layers apart. The
record (`goals/`, `decisions/`, `runs/`) is the trail: every question, run, failed attempt and replaced
experiment, so that every number has an origin. No card names a paper, because papers change, split, merge
and are dropped while the record stays as it is and is read by each of them. Above the record, the lab's
library holds one **overview** page that keeps the whole trail in one place, whatever paper it serves: the
words on one running example, the workflow, every related-work row, every established finding, every
criticism and what answers it. Beside it, each **active paper** has its own page, written as that paper's
technical argument and nothing else: it never talks about the lab or how it works. A paper page opens, under
its thesis, with a **scope table** of five rows: the target audience and venues, the core thesis, the
experiments included (the clean runs each claim rests on), the experiments excluded or superseded (a run
that hit a harness bug, a subset a full run replaced, work that belongs to another paper), and the open
experiments the paper still needs. Then come its words on one running example, its pipeline, its
comparison with the published work of its community, its established results citing only included work,
the criticisms a reviewer in that community will raise with what answers each, and what remains to run.

The control runs from the papers to the lab, not the other way. Which questions and experiments are
proposed, ordered, parked or dropped follows the open rows and the criticisms on the active paper pages: a
decision says which paper row it serves, and a step that serves none is a candidate to park or drop. When a
result lands, the overview and every paper it touches are revised; a run that a newer one supersedes moves
from included to excluded. Adding, merging or retiring a paper is direction, so it is decided with you. A
question's or an experiment's So far gives the current finding, not the story of the bugs on the way: the
replaced experiments and the reasons they were replaced keep that story.

A lab opts in by naming the pages in `lab.toml` (section 9). Then `lab check` holds each scope table:
included lists only finished work that is not void, replaced or dropped, open only work still to do, nothing
parked or dropped is in either, and a goal, a question or a step sits in one row; a paper page committed
before a result about something it lists under included or open is noted. `lab now` lists each paper's open
experiments, and the live steps no paper lists, by themselves or
by their question or goal.

## 5. What the agent uses

- **To look:** `lab now`, `lab show <id or path>`, `lab brief <question>`, `lab since`, `lab define <term>`. Each shows one level,
Expand Down Expand Up @@ -449,7 +479,7 @@ Five skills: `lab-set-up`, `lab-work`, `lab-review`, `lab-talk` (tell you where
a decision to you, record what was decided), and `lab-mission` (save, launch, resume and close a mission). One
role: the reviewer.

Seven rules:
Eight rules:

1. Say what would refute it before you run.
2. A number counts only as a reviewed result of a locked plan.
Expand All @@ -462,11 +492,13 @@ Seven rules:
6. Record what was decided and why, in the lab's words, and show the operator the text as recorded.
Never record a decision that was not discussed with them.
7. Say what you are on. When you start on a card, log what you are doing and why now.
8. Papers drive the work; the record names no paper. The papers' open rows and criticisms say what to
propose, order, park or drop, and a result revises the overview and every paper it touches.

`lab check` holds what can be held mechanically: ids and links, the goal's vocabulary being terms of the glossary, a title and a description on every
card, short cards that write only their own front matter, a test on every line, a reason on every
replacement, unchanged locks, sealed runs, results that still re-derive and were reviewed, use inside
the resources' amounts, a decision behind every goal, question and metered amount, an experiment
the resources' amounts, what each paper's scope table lists when the lab names its papers, a decision behind every goal, question and metered amount, an experiment
behind every answer, the method and the skills as the kit installed them, and a history that only
grows. What it cannot read as this page describes (a mistyped folder, a list written as one value, a
file that is not text) is a problem it reports; no command stops on it.
Expand Down Expand Up @@ -903,6 +935,33 @@ holds the line `launched`, else next, and `lab check` asks for the status to be
reports a mission that breaks any of this, one whose `carries:` is not a decision, and a current mission whose
decision is still open.

**Papers.** `lab.toml` names the library's paper pages in a table, as library paths (the path under
`content/`, as `see:` writes it):

```
[papers]
overview = "entries/where-the-results-stand"
active = ["entries/paper-a", "entries/paper-b"]
```

`overview` is the page that keeps the whole trail; it is never an active paper. Each page under `active` is
in the library and marks the three cells of its scope table that list work, `data-scope="included"`,
`data-scope="excluded"` and `data-scope="open"`, empty when it lists nothing:

```
<tr><th>Experiments included</th><td data-scope="included"><a href="/lab/G1.Q1.4/">G1.Q1.4</a>, ...</td></tr>
```

A cell lists what its `/lab/<id>/` links name, written from the site's root; any other text in it is prose. A
cell whose closing tag is left out ends where the next cell or row begins. The rules hold for the ids written in
the cells. Under included, a step is done and not void, and a question or a goal is not parked or dropped;
under open, a step is open, waiting or running, and a question or a goal is not done, parked or dropped. A goal,
a question or a step has one row on a page: where another row needs to mention it, it is named in words,
without its link. A question written in one row and a step under it written in another is allowed: the open
question with its one finished step included is the usual case. A decision may be cited in any row. In
`lab now` and in the note on a stale page, a goal or a question written under included or open stands for
every step under it. A lab without `[papers]` has none of this.

**Worked-out states.** A goal is open until `reached: true` makes it done or `dropped:` drops it. A
question is waiting while the decision it came from is open, then open, then done with its answer; it
is parked while the last settled decision that parks or brings it back parks it. A step is waiting while
Expand Down
15 changes: 12 additions & 3 deletions method/METHOD.md
Original file line number Diff line number Diff line change
@@ -1,6 +1,6 @@
# The method

Seven rules every lab keeps. Each says what holds it: a command that refuses, a check, or a reviewer's
Eight rules every lab keeps. Each says what holds it: a command that refuses, a check, or a reviewer's
judgement. The lab's own `AGENTS.md` adds only what is local.

1. **Say what would refute it before you run.** An experiment states what it compares, what would
Expand All @@ -25,13 +25,14 @@ judgement. The lab's own `AGENTS.md` adds only what is local.
4. **Write for a reader who was not there.** Every page stands on its own: it says what each thing it
mentions is, names each arm of an experiment and what it is, names another page by its exact
title, and uses a term only as the glossary defines it, the one place a word is defined. A number is said once: it lives in its result row, and a card names it. A state is
worked out, never written, and So far says what was learned, never where a step stands. Every
worked out, never written, and So far says what is learned now, never where a step stands and never the story of
the bugs on the way: the replaced experiments keep that. Every
card has one of the same states: open, waiting, running, parked, done, replaced or dropped. A
number's label says what it counts and of how many, in one set of cases named once: "of the 20
checks the solver leaves open", not "of 20". *The numbers and the budgets are held by `lab check`; whether a page
can be read cold is judged, and the reviewer reads for it.*
5. **Decide everything that is not direction or resources, and never wait idle.** Direction is the
goals and the questions. Resources are what `resources.toml` lists and how much. For those two,
goals, the questions and the active papers. Resources are what `resources.toml` lists and how much. For those two,
open a decision that says what was seen, the options and what you recommend, and work on what it
does not hold. What a decision settled needs no further yes. Everything else is yours: decide,
record it, go on. A mission (`missions/`) carries out what a settled decision settled across sessions, and approves
Expand All @@ -49,3 +50,11 @@ judgement. The lab's own `AGENTS.md` adds only what is local.
should never have to ask you. When you start on a card or move to another, log it:
`lab log <id> doing "<what>" "<why now>"`. The line stands on the front page until a result, a
stuck line or a choice about the same card ends it. *Judged.*
8. **Papers drive the work; the record names no paper.** The record is the trail, and no card names a paper.
When the lab writes papers, its library keeps one overview of the whole trail, and one page per active
paper that is that paper's argument and nothing else. A paper page opens with its scope table: audience,
thesis, the experiments included, those excluded or superseded, and those still open. What you propose,
order, park or drop follows the papers' open rows and their reviewers' criticisms; when a result lands,
revise the overview and every paper it touches, and move what a newer run supersedes to excluded. Adding,
merging or retiring a paper is direction. *Held by `lab check` when `lab.toml` names the papers; otherwise
judged.*
2 changes: 1 addition & 1 deletion pyproject.toml
Original file line number Diff line number Diff line change
Expand Up @@ -5,7 +5,7 @@ build-backend = "hatchling.build"
[project]
dynamic = ["readme"]
name = "lab-kit-cli"
version = "1.11.1"
version = "1.12.0"
description = "A research lab's record, its method, and the command that keeps both."
license = "MIT"
authors = [{ name = "Piergiuseppe Mallozzi" }]
Expand Down
3 changes: 2 additions & 1 deletion skills/lab-review/SKILL.md
Original file line number Diff line number Diff line change
Expand Up @@ -44,7 +44,8 @@ A failed plan is fixed by its author and read again.
the first sentence and folds the rest.

A failed results review is final for that experiment. Say in the note what a replacement must do
differently.
differently. Its numbers can then be cited nowhere: if a paper page lists it under included, `lab check`
says so, and the author takes it off.

When you read a plan, also read the card as someone who was not there. If an arm is named and not
explained, or the card points at something it does not say, say so in the note: it does not fail the
Expand Down
10 changes: 8 additions & 2 deletions skills/lab-set-up/SKILL.md
Original file line number Diff line number Diff line change
Expand Up @@ -43,8 +43,14 @@ You set the lab up in one discussion with the operator, so that everything it se
here first.
11. The site: the lab's `docs/` folder is a folio library whose charter lists `lab_kit.site` under
`views:`. Set it up with folio's own skill if it is not there.

12. Commit. Git is what holds the record: nothing is held before the first commit, and
12. If the lab will write papers, ask the operator which papers it aims at now, for whom and where (it is
direction, and may stay empty). Write the overview page (for example `entries/where-the-results-stand`) with folio's
write skill: the words on one running example, the workflow, the related work, and empty sections for the
findings and the criticisms. For each paper, write `entries/paper-<slug>`: its thesis, then the scope table
(target audience and venues, core thesis, experiments included, excluded or superseded, open), with the
last three cells marked `data-scope="included"`, `"excluded"` and `"open"`. Name them in `lab.toml` under
`[papers]` (`overview`, `active`). No card names a paper (method rule 8).
13. Commit. Git is what holds the record: nothing is held before the first commit, and
`lab check --strict` says so.

Done when `lab check --strict` passes and `lab now` shows each question as open.
13 changes: 12 additions & 1 deletion skills/lab-talk/SKILL.md
Original file line number Diff line number Diff line change
Expand Up @@ -33,7 +33,9 @@ from the record, shortest first. A decision is a discussion, not a request for a
comes up, give its definition from `lab define` in a few words. Never use a word in its everyday
sense when it is also a lab term. Compare items in a table.
- For "where are we": each goal in a line, each open question with what is known so far, what is
running, what is next, what there is to discuss. No more than a screen.
running, what is next, what there is to discuss. No more than a screen. When the lab names its papers,
lead with them: each active paper by its title, with what its open row still needs, then the steps
serving no paper.
- `lab brief` is the question's page as text. Give it as it is, cut to what they asked. If what you
want to say is not in it, the card is missing something: fix the card, then answer.
- If the record does not say, say so. Never fill a gap from memory.
Expand Down Expand Up @@ -64,6 +66,15 @@ that resource or that experiment under `about:` before it is decided: `lab new d
--about <resource or id>`. What a locked decision is about cannot be changed afterwards, and a
resource's new amount names the decision in `from`.

## Papers

When the lab names its papers, they set the direction: the options you bring follow their open rows and
their reviewers' criticisms, and an order of work (`order:`) is built from the open rows, the paper the
operator cares about first. Adding, splitting, merging or retiring a paper is a decision like a question's:
`lab new decision "<what it is about>"`, with the paper named in Seen and Decided by its page's title. Then
change the pages and `[papers]` in `lab.toml` (the lab-work skill, The papers and the overview). A
retired paper's open work is parked or dropped in the same decision, or moved to another paper's open row.

## When they decide something nobody raised

`lab new decision "<what it is about>" --about <id>`, fill **Seen** with what prompted it, then
Expand Down
Loading
Loading