Before and after

The retired company name sat inside ten sentences of the business-model copy: who mobilises the kit, who runs the survey, who holds it between seasons, who does quality control. Those are the sentences that describe the partnership itself, so the old brand was doing load-bearing work in the part of the page most likely to be read closely. All ten now read Island Lab, with the surrounding prose checked sentence by sentence rather than token-swapped blindly.

Two fixes in one bump. The credit prices were placeholders that our own sourced research had already overtaken: carbon moved from 40 to 24 EUR per tonne and biodiversity from 200 to 150 EUR per hectare per year, which takes the stacked figure from about 216 to about 160 and removes roughly a quarter of the modelled revenue. A visible provenance note now states the bands, that the entire published blue-carbon price series is mangrove rather than seagrass, and that no Posidonia carbon credit has ever issued anywhere. The brand was fixed in the same pass, including both the English and Italian dictionaries.

Fourteen visible strings carrying the retired brand, including the browser tab title, the map attribution, three layer-source links, two hex tooltips and the footer credit in both languages. Sixty-seven other matches in the same file were left alone on purpose, because they are JavaScript globals, storage keys, asset paths and version-history comments: renaming those is how a previous sweep shipped a syntax error to production. Two links also pointed at the retired hostname and now point at the canonical one.

Gained the survey data cycle as a fifth section, with a tool lane and a pipeline lane per step, so the actor-to-tool matrix is no longer the only way in.

The most consequential change of the week, and the one with no "before" URL because this app is not file-versioned. A point-in-polygon test treated the 2017 mapped extent as ground truth and discarded every observation falling outside it, which on the committed snapshot deleted 7 of 13 cells. The product exists to find where the meadow is not where the map says, so that evidence was being deleted before it could be drawn. Now the landmask decides what may paint and the extent only classifies: a cell outside the 2017 polygon renders as disagreement, tagged, and is counted as disagreement rather than reported as excluded.

What a municipality actually runs, and what it costs. An accountability matrix across eleven functions and five parties, the annual cadence as seven steps, the technology as nine layers with honest status, and the cost as two halves: comparable public programmes that are sourced, and our own four lines which are marked pending rather than estimated. It states three uncomfortable things rather than smoothing them, including two published restoration costs that disagree by two orders of magnitude.

The whole survey cycle in ten seconds, as a synthetic animation plus a still for slides. Built by a new concept-demo pipeline whose point is that a demo comes before a build, not after it.

The deploy hook is not firing. The newest commits are on the main branch and production is serving an older build. Until that is resolved, committing is not the same as publishing, which is the assumption most of this week's plan rests on.

Four surfaces still carry the retired brand in visible text: the data tracker's logo, the impact model, the stakeholder map and one project brief. They were left because none of them is in the demo set, and each fix costs a version bump. They are known, not missed.

Sign-off has no owner. The role that makes an untrained operator's data defensible carries authority in the schema and does not exist in the data model. It is drawn as unresolved on the operating model rather than quietly assigned.

Every reading on every map is synthetic. The protocol is real and tested; the fleet has been dark since 8 July. Each surface says so at the point of display, and that should stay true.