Foreshock Docs / Docs / Operate / Manual Spectrometer

Manual Spectrometer

GUIDE·9 min read·UPDATED Wed Jul 15 2026 00:00:00 GMT+0000 (Coordinated Universal Time)

The Manual Spectrometer (the Spectrum Analyzer, at /client/query) is Foreshock's one-shot instrument. Paste a specimen, score it, and get back the full geometric breakdown — no sensor, no ongoing monitoring, just one paste and one answer.

What it's for

Most of Foreshock runs continuously against a connected source. The Manual Spectrometer is the opposite: a single specimen, scored on demand. The two most common uses are:

  • Research and synthetic examination — testing how a phrase, a paragraph, or a whole document scores across the spectrum, one specimen at a time.
  • One-off tonality checking — pasting a draft reply, a press release, or a script before it goes out, to see its likely emotional impact ahead of time.

Every score spends one request from your monthly manual-query allotment (Free: 150/mo, Tectonic: 500/72h rolling, Seismologist: 1,500/72h rolling — see Pricing). The specimen itself is scored and immediately discarded; only the resulting numbers are kept.

!!!SCREEN SHOT: The Manual Spectrometer dashboard view, showing the specimen dock open, the quota meter, and the scatter/spectrum charts populated with a scored example.

Scoring a specimen

Open the specimen dock (the Specimen button), paste text, and press Score specimen (or ⌘/Ctrl+Enter). You can also upload a small .txt file instead of pasting (capped at 8 KB in the dock; a direct paste can run up to 32,000 characters before the server rejects it).

A few query parameters change how the scorer behaves, all exposed as toggles above the Score button:

  • Limit — cap the number of rows returned; leave blank for everything above threshold.
  • Min Similarity / Min Dot Product — gate out anything below a similarity or dot-product floor.
  • Auto-Kneedle (?elbow=1) — ask the scorer to auto-truncate the spectrum at its knee, returning only what the Kneedle considers signal.
  • Strip code tokens (?strip_code=1) — useful when scoring specimens that mix prose with code fences or identifiers.
  • Protect Shunts (?protect_shunts=1) — excludes your own account's Shunt Studio shunts from the request, so a public-facing tool built on this endpoint can't be used to fingerprint what your community privately watches for.

Every score is backed by a receipt: a UUID permalink that anyone holding the link can view — the same idea as a payment receipt link. A receipt never shows the scored text, only its SHA-256 fingerprint, its length in characters and words, and the full set of results.

Two ways to view a result

  • Dashboard — the charts described below, meant for reading at a glance.
  • Raw — every returned row as a sortable table (key, sensor, group, similarity, distance, dot product, sim %). This is the power-user view, and the key column is exactly what you'd quote in a bug report or paste into the API reference. Results can be exported as JSON from here.

Each graph, explained

  • Latent Multivariate Gaussian Population (scatter) — every scored key plotted by similarity and distance, colored by group (density probe, saturation key, gradation, shunt tripwire). Drag-select the legend to isolate one group.
  • Centroid Group Density Distribution — one bar per saturation ("%-key") scalar, with a 20% floor line marking the minimum for a viable research signal.
  • Emotional Valence — Nearest Centroid Classifiers — the graduated spectrum, grouped by pillar, with the base/+/++ intensity tiers shown side by side.
  • Signed Centroid Indexes (Kneedle-Immune) — outlook (negative ↔ optimistic) and action (suppressed ↔ mobilized), each shown as a beam whose width is the evidence spread. These never get truncated by the Kneedle, because their sign is the signal.
  • Dense Concept Saturation (Background Tonality) — the pragmatic scalars (sarcasm, gratitude, sycophancy, and similar meta-concepts), which are never Kneedle-ignored and are read as noise below the 20% floor regardless of curve shape.
  • Monolithic Shunts — your account's mean-pooled shunts on a similarity/distance scatter, with a detection box; a marker inside the box is a live trip.
  • Intrinsic Shunts — the higher-magnitude, LCE-eligible shunts, tuned rather than live-edited, scored on their own scale.

See How Aldous scores language for what these three sensor families actually measure, and Shunts & Latent Concept Erasure for what a shunt trip means.

!!!SCREEN SHOT: The Raw results table, sorted by distance, with the JSON export link visible.

Next

Want the same scoring loop from your own code instead of the browser? See the API reference — it's the identical POST /api/v1/score endpoint underneath.