The instrument is now calibrated and verified.
Three things have landed since Issue #4 shipped. The engine now carries a
per-spec discriminator that retires the Gatsby-SPA-catchall false-positive class.
The eleven dogfooded kineticgain.com documents are now ed25519-signed and verified
— the first verified row in any Pulse measurement. And the universe has grown to
941 domains, including all forty-seven .kineticgain.com operator surfaces
that the portfolio actually deploys. This page is the locked-in baseline against which
Issue #5 (August 2026) reports the first true quarterly delta.
The baseline numbers, locked in
A full v0.4 crawl on the 899-domain universe at hardening close. The raw is committed to
procurement-pulse-engine/data/issue-4-v04-full.json. The numbers below are not
a re-publish of Issue #4 — they are the calibrated instrument's current reading, with the
discriminator and signing fixes both applied.
.kineticgain.com operator-surface CNAME the Kinetic Gain portfolio deploys (47).
The "every deploy enters universe" rule is now a v2 strategy invariant — measured properties grow
with the portfolio, not behind it.
The discriminator — what the false positive looks like, after the fix
Issue #4 flagged corporate.charter.com as a 82/100 publisher. It wasn't — the site is
built with Gatsby, and its catch-all returns the 404 page-context payload as application/json
for any /.well-known/* path. Engine v0.4 (commit
5892d88)
now requires each of the eleven Suite paths to carry its canonical *_version field —
the minimal Suite-spec shape, not just "valid JSON". Re-probed live, corporate.charter.com
drops 82 → 0; kineticgain.com holds 100/100. The instrument now distinguishes a
Suite document from a SPA's internal page-rendering context.
| Spec | Path | Discriminator (v0.4) |
|---|---|---|
| AEO | /.well-known/aeo.json | aeo_version |
| Agent Card | /.well-known/agents/index.json | agent_card_version |
| Prompt Provenance | /.well-known/prompts/index.json | provenance_version |
| Evidence Bundle | /.well-known/evidence/index.json | evidence_version |
| MCP Tool Card | /.well-known/tool-cards/index.json | tool_card_version |
| Tutor Card | /.well-known/tutor-cards/index.json | tutor_card_version |
| Student-AI Disclosure | /.well-known/student-ai-disclosure.json | disclosure_version |
| Classroom AUP | /.well-known/classroom-aup.json | aup_version |
| Clinical AI Card | /.well-known/clinical-ai-cards/index.json | clinical_ai_card_version |
| Incident Card | /.well-known/incident-cards/index.json | incident_card_version |
| Decision Card | /.well-known/decision-cards/index.json | decision_card_version |
application/json, and any "valid JSON of the wrong shape" response.
What this still doesn't catch: a vendor that hand-crafts a JSON file with the right discriminator
field but otherwise empty payload. That would be intentional misrepresentation, not a SPA artifact —
a different problem, and the publishable evidence already includes the full document for inspection.
The first verified row
The Issue #1 commitment around ed25519 signature verification has now closed.
kineticgain.com's eleven dogfooded /.well-known/ documents are signed
against a public key published at
kineticgain.com/.well-known/pulse-signing.json.
The engine probe now reports { verified: 11, unsigned: 0, invalid: 0 } on the dogfood row —
the first time any row in any Pulse measurement has carried a verified signature.
All eleven Suite documents under kineticgain.com/.well-known/
are signed with the same ed25519 keypair. The signature covers the canonical serialization of each
document with the signature block recursively removed and remaining keys sorted.
By default the engine trusts the embedded public_key (tamper-evident).
With --verify-key-fetch it instead fetches the key from signing_key_url
(provenance). Both pass on the dogfood row. The probe pattern mirrors
hash-attestation-rs.
For any future vendor that signs their Suite docs and publishes a key, the engine will roll the
verification state up into signatures.verifiedRate on the issue dataset — the August
read will report this as a leading indicator separately from the publication rate.
What Issue #5 (August 2026) reports against
The August quarterly delta is the first time the Pulse runs against a calibrated instrument. Issue #1 was the zero-baseline calibration; Issues #2 and #3 expanded and stress-tested it; Issue #4 published the first universe-scale finding alongside the false-positive that shaped the v0.4 fix. Issue #5 has nothing left to calibrate — its only job is to measure the field's movement (or lack of it) across three months of buyer pressure and vendor reaction.
- Same universe, recrawled. No editorial pivots between now and August. Numbers move because vendors move, not because the lens did.
- The publication rate — currently
1 / 941 (0.11%)— is the baseline. Anything above it is news. - The verification rate is reported separately. If even one new domain joins
kineticgain.comin theverifiedcolumn, that's a structural shift, not noise. - The drift artifact (
data/issue-5-drift.json) will surface every per-domain change: newly publishing, stopped publishing, score and per-spec changes. Reproducible, not editorial.
Pre-registration — the read-locked plan for August
To keep the instrument honest, the August read is pre-registered here, not adjusted after the fact:
- Crawl date: the first weekday on or after 2026-08-15.
- Universe: whatever
universe.csvcontains at crawl time on themainbranch ofprocurement-pulse-engine. The "every deploy enters universe" rule keeps this honest — surfaces are added when they ship, not chosen to flatter the numbers. - Engine: v0.4 or whatever
maincarries — any further engine changes between now and August will be disclosed in the issue itself, alongside the v0.4 baseline for comparison. - Drift report: generated against
data/issue-4-v04-full.json(this baseline). The drift JSON ships in the issue folder.