Visibility 2x2

A local board for arguing about what Cargento should become. It lives at
docs/visibility-2x2/ in this repository and is a development tool, not part of the
shipped plugin.

Start it

Stdlib-only, no dependencies. From the repository root:

python3 docs/visibility-2x2/serve.py --port 8899

It runs in the foreground and prints the URL. Keep that process running while you open the printed
URL with the available browser capability. From another local shell, this command opens it in the
default browser:

python3 -m webbrowser -t http://127.0.0.1:8899/

Use 127.0.0.1, not the hostname spelling: the server binds IPv4 loopback only.

Pass --port N if 8899 is taken. There is no daemon mode and no state file, because this
is a thing you open to have a conversation and then close. Stop it with ctrl-c.

Tell the user what they are looking at

The board carries 57 candidate signals. Each is scored on two axes, and the horizontal one
is the one people misread, so lead with it.

• Vertical is impact: if the user had this reliably and instantly, how much would it
  change what they actually do.
• Horizontal is access: how hard is it for a normal user to get this information today
  without Cargento. Left means a harness already shows it, so building it is convenience.
  Right means Cargento creates access that did not exist.

So the top-right corner is where the product earns its position. Engineering effort is a
separate build score and has nothing to do with the horizontal axis.

Five views along the top: the 2x2 map, by outcome (grouped by the six outcomes any
signal can serve), journey (a story-map of narrative stage against release, where each
column is an item's link to one of the five promises and the Promise row is the
promise map's wording verbatim),
the two open decisions that need a human call, and a sortable table.

Three things to say before anyone quotes a number

The scores are a blind panel consensus rather than one person's estimate: three independent
lenses scored all 57 items without seeing any prior numbers, and the median is shown. That
still leaves three ways to misread the board.

1. Rank on riskAdjustedImpact, not impact. Several items score high on access
   precisely because nothing computes the number anywhere, so their value depends on a
   heuristic being right. That discount is priced per item.
2. Turn on "primaries only" before treating it as a build list. Twenty-one subset
   relationships were found; the board reads as 57 features and is nearer 47.
3. The differentiated cutoff is a convention. Ten items sit within five points of the
   impact-70 line and are marked by the shaded band. Describe them as unsettled, not as
   decided.

docs/visibility-2x2/README.md explains all of this in more depth, including where the
board came from and what is in the audit trail.

Editing during a session

Clicking a dot opens it for editing: sliders for both axes, keep / park / cut, and a notes
field. Everything saves back to items.json, so Export (which copies the whole set as
Markdown) reflects whatever the room decided. Editing items.json directly also works, and
the server picks it up on the next browser refresh.

After changing any score by hand, run:

python3 docs/visibility-2x2/audit/verify-scores.py

It recomputes every value from the raw panel evidence in audit/ and reports anything that
no longer matches. Worth running before quoting a figure at anyone, because the failure mode
here is a stale number surviving into a meeting.

Do not

• Do not rewrite scores to match an opinion without re-running the panel. The numbers are
  defensible because they came from lenses that could not see each other's work, and hand-
  tuning them quietly throws that away. If a score looks wrong, add a note saying so.
• Do not present the differentiated quadrant as a hard set of ten. See the cutoff caveat.
• Do not bind the server to anything but loopback. It is a working document, not a service.