The Cape Ann climate reading

A cape’s climate conversation enters — 27 documents, five voices: the town and its agencies, the people who live here, the marsh and the fisheries, working life, and the study of what may come. 975 compositions. habitat reads how each voice bends away from the room — and where the room’s quiet is made of voices pulling in different directions. A reading, not a finding.

the readings, printed — by the instrument

Every table below is emitted from stored artifacts by a build script; no number on this section was typed by a person. Coupling per named axis is warp — how much that axis pulls the other sixteen — read by the instrument's own function. Displacement is observed warp minus the median of a matched-n permutation null (300 draws, seed 7), each construction against its own null. A quoted warp states raw or ridged: raw is the certificate's convention; ridged is the live field's (the declared 0.01 conditioning constant inside the covariance). The correlation matrices behind every warp value here are published beside this page — reading-data.json, both constructions, its fingerprint in the data island above — and the tables recompute from that file alone. Nothing here is sorted, selected, or thresholded, and nothing here says what any axis means — that resolution is the reader's. town = town & agencies · people = people who live here · marsh = marsh & fisheries · work = working life · study = study of what may come.

observed warp, by voice — raw, ridged, and the gap

Start at the highlighted Boundary row. Its raw variance is the thinnest in the corpus — the declared constant is five times its size — so its gap column shows most clearly what the pair of constructions is for: raw is what the discourse deposited, ridged is what the operable field holds, and the gap is the constant's share.

Observed warp, raw construction, all seventeen axes, canonical order. Full nulls and artifacts by request.

displacement against each voice’s own null

The highlighted rows — Stability, Boundary, Resonance — are where the voices pull in different directions past their own size-baselines; the section below the tables walks them. Then read the study column top to bottom: its lower rows move together, a concentration that survived its registered test.

Displacement against the matched-n null, raw and ridged per voice. Where profiles agree across voices the coupling is invariant — structure; where they differ by voice it is conditioned — and what that means is yours to resolve.

floors, as values

The highlighted last row is the floor itself — each voice's Marchenko–Pastur edge, both constructions. An axis reads against its own voice's floor; the ridged floor carries the declared constant inside it, which is why both are printed.

Floor values as values: per-axis variance and each voice's Marchenko–Pastur edge, both constructions. The ridged floor carries the declared constant inside it; the use statement above quotes the pair and its gap rather than electing one.

the surplus overlay, per axis

Only comparisons within a row or a column are readable here — the two streams' scales differ by construction. The highlighted tail rows carry the one tested reading: their concentration in the study voice survived a registered null. Everything else is values awaiting a reader.

The instrument's surplus overlay, per axis per voice: lead is the engine's own ratio of semantically-mediated to direct variance, surplus the signed difference. A construction note travels with these numbers: the ratio compares streams of different widths and normalisations, so its absolute scale is not interpretable across the two spaces — only comparisons within a column or a row are readable here. Values, not verdicts; what any of it means is the reader's.

where the voices pull apart

Read the Stability, Boundary, and Resonance rows of the displacement table: every voice moves, and the signs disagree — the marsh pulls one way on persistence where the people who live here pull the other. What looks like quiet at the room’s scale is disagreement at the voice scale. That is shown in the raw construction, the certificate’s own; on Boundary the effect thins under the ridged construction, and the table prints both so nothing rests on a choice we made. These axes carry the instrument’s own framing questions — what must persist? who is inside this decision? who will be affected? — printed here as the frame the instrument declares, not as what any voice said. What the disagreement means belongs to the reader.

The fourth quiet axis is different in kind: Boundedness is thin in every voice because of a declared limit of the instrument’s own extraction — a limit written down and predicted, before this corpus was ever read, in a registration that named it corpus-independent. Three silences are the discourse’s. One is ours.

how this was tested

Every number above was emitted from stored artifacts by a build script; a drift gate fails the build if the page and the artifacts disagree. The claims were tested against permutation nulls matched to each voice’s own size — because how much a voice is present is part of what that voice is, not a nuisance to correct away. Predictions were registered before runs, and the registrations kept their failures: a reading we liked — that the least-present voices refract most — was killed by its own registered test. Two more died the same way the day this page was rebuilt. The registrations, the nulls, and every draw are available by request.

how this reading is kept

Each document’s field is frozen with a per-document hash chain. The corpus was built twice, independently, sixteen hours apart: the semantic layer came back byte-identical, 27 of 27 documents. The structural layer did not — so every structural number on this page is quoted from one named build, and that limit is part of the record, not a footnote to it.

what this reading does not say

Seventeen thousand filings, one hundred sixty-four letters, and a town’s own climate record — one instrument, unchanged. The reading is handed back; what it means stays with the town.

← habitat · method