Changelog

There is now one way to write an LTP model with an AI: skills/ltp-project, which produces the Reason Commons *.ltp.yaml project file — the same interchange document the app imports and exports as JSON or YAML, version and event history included — from a PDF, a Markdown document, a repository's plans and issues, a roadmap or meeting notes, with a report of every judgment call. It is a byte-identical mirror of the canonical skill in Promise-Foundation/reason-commons; edit it there and copy it here. The older project-ltp skill is deprecated: its ltp-model.yaml was a second, incompatible format that the app cannot import and that carries no history, and its GitHub issue sync is now done by the Reason Commons projection, which writes and reconciles managed issues itself. The skill's read-only dashboard stays as a viewer for the legacy models still bundled here.

The 2R Research Circle project is gone from the dashboard: its model was the July 2026 analysis, superseded by the live Reason Commons space connected to the 2R-Research repository, which exports the current model to that repository as .reason-commons/<space-id>.ltp.json. The throughput track that fed it — counted from this repository's Git history — went with it, so throughput.config.json now holds no tracks and the generator tolerates that. The dashboard was rebuilt and republished; the Second Renaissance and the software-engineer example remain.

The annotation demo now opens on “Can the world switch to clean energy in time?” with three linked IEA reports mapped to specific claims about emissions pace, renewable scale, grids, electrification and nuclear tradeoffs. The closing now leads to an on-page appendix with AI-control evidence from METR, the International AI Safety Report and Eliezer Yudkowsky, plus the original randomized Ctrip remote-work experiment.

A new contribution-proposals skill turns a batch of raw participant contributions — a Google Doc export, a form dump, whatever a room actually wrote — into one reviewable proposal each for how it might extend, support, challenge, or fail to fit an existing tree, and renders the set as a page the group works through together: accept, edit, or reject, with accepted proposals turning from dashed to solid and staying, so by the last one the tree on screen is the original plus everything the room agreed to. Contributions are quoted verbatim, the machine's reading is labelled as the machine's, and nowhere to put it is a first-class result rather than a failure — a contribution the tree cannot hold may be the most useful thing in the batch. Nothing merges: ltp/ltp-model.yaml is read-only to the whole pipeline, and refreshing the page throws the room's decisions away, which is the honest behaviour for a tool with no standing over the tree. The skill builds to a stamped, unpublished test route by default, so a half-read batch can be served and clicked through before anything touches the page that ships. Reading a contribution that arrives in its own vocabulary — a permaculture cooperative, a mutual credit network — is most of the work, and the skill's guidance is about finding the operative claim inside the experience that earned it rather than reaching for the nearest keyword. Ancestors of whatever is under discussion stay at full strength up to the root, because a claim you cannot trace to a goal is not one anyone can review. Built for Demo C of the talk, where it ships with a rehearsal build from nine contributions. Running Demo C is the runbook for presenting it without having built it.

The dashboard now keeps the open project and tree in the URL, so a single view — the Second Renaissance current reality tree, say — can be linked to directly instead of "open the dashboard, pick the project, click Reality", and Back and Forward walk the views visited. Stale links degrade rather than break: a tree the project has not modelled opens its overview, and an unknown project lands on the picker. The shape of the links is in docs/dashboard.md.

The front page now leads with what the site actually is — a better way to think together and to act on it: trees of change and trees of action, evidence and discussion mapped onto them, with the AI tooling as the bet — and offers two co-equal ways in, the series and the demo. The prose that used to fill the page moved into an About section below the fold. The nav went from Goal Aligner · The Series · Dashboard · Blog to Introduction · Demos · Guide · Blog · About; Goal Aligner and the other rough pieces are gathered under Demos, and the Logical Thinking Process now has a plain-language Guide in front of its analysis documents. Goal Aligner moved from /alignment/ to /demos/goal-aligner/. Design notes: docs/plans/2026-09-01-site-positioning-and-landing-redesign.md.

The Wrong Queue and The Forum Doesn't Remember now each have a plain, non-scrolling companion: the same prose as a classic illustrated essay, with the drawings held still — and the passages that were animations shown as two or three stills side by side. Nothing to scroll-trigger, no JavaScript. They exist to be usability-tested against the scroll-driven originals, which are unchanged; both are linked from the series index.

The landing page also had a pass — one primary "read the series" call instead of four equal buttons, the "Project LTP" wording dropped from the visitor-facing copy, and the dashboard build-and-serve instructions moved out to docs/dashboard.md.

A beat-by-beat browser sweep of The Forum Doesn't Remember, and Doesn't Cumulate (formerly Somewhere to Put It) turned up four collisions. Three were phone-only: the "rooted in an actual place" cards in the two timeline stages sat on top of one another, and the "is necessary for" label on the goal-tree stage ran off the right edge. The fourth was on the desktop layout — a caption grew over the diagram at middling widths — and was fixed by tightening the prose rather than moving it. The superseded story/ draft that this piece replaced has also been removed from the repo.

All five scrolling explainers now open on a landing — a title, a scroll cue and a thin bar back to the rest of the site — instead of dropping the reader inside a pinned drawing with no headline and no sign that the page moves. They also respond faster: the second beat of a section used to take a screen and a half of scrolling and now arrives in under half a screen. In Somewhere to Put It, Maren's story stacks down the left of the drawing rather than each beat wiping out the last — or start at The Wrong Queue, which is the way into the series.

A presentation deck at talk/index.html, built so the audience experiences the Logical Thinking Process before hearing its name. Five acts, each opening on an audience-facing question rather than an artefact: why good conversations go nowhere, what better would look like, the conflict in the way, what changes if it dissolves, and what to do next. The LTP names are withheld until the fifth act, where a ledger maps each act back to the tree that produced it and the deck becomes evidence for its own argument.

An unlabelled five-node spine advances act by act in the corner, so the room registers structure without being told what it is; on the reveal slide the same spine fills in with current reality, goal, conflict, future reality, next steps. The visual system carries the act-three tension directly — loose italic conversation at the edges, a hairline reasoning graph drawing itself, one amber node for whatever was newly contributed.

Single self-contained file, fixed 1920x1080 stage that letterboxes rather than reflowing, arrow-key and swipe navigation, #n deep links, and inline text editing behind the E key.

Drove all three explainers in a real browser and fixed what only shows up there — captions sitting on top of the very thing they were pointing at, boxes falling off the edge of their own drawing, a chain of arrows that read as a tangle. Parts two and three now have portrait layouts to match part one, so the diagrams are readable on a phone instead of shrinking to a ribbon, and the series finally uses the real Organic colours and Figtree rather than a palette that was merely close.