certificate concept

Definition	A claim, made at a time, that something is safe to act on. A clustering, a benchmark score, a freshness guarantee, and a rank bound are all certificates. Chapters 9 and 13.
Example	A certificate cleared 1000 decisions and the witness disagreed on 206 of them, a false-clear rate of 0.206.
Book	Data Mining as Observation, draft 0.2, commit `f3914f0`; entry id `certificate`, kind concept.
Status	measures [demonstrated]. Corrections: 1 item(s), see below.
Defining equation	Book equation 0.26.
Assumptions and scope	A certificate is graded only where a witness exists, and where none exists it is unassessed rather than safe. A certificate is bound to what it certified, the bytes, the anchor sample, the footprint, and the time, and carries none of it to another corpus, workload, or moment. A vacuous certificate proves nothing about the object. It does not prove that no code preserves rank or that no structure exists, only that this test could not certify one. The rank certificate’s floor is a guarantee in its strict setting and an estimate in its percentile setting.
Prior art	none recorded
Evidence	`geometric-observation/claims/LEDGER.md:65`, `geometric-observation/experiments/GO3-certificate-vacuity-v3-NOTES.md:1-60`, `geometric-observation/claims/LEDGER.md`, `geometric-observation/chapters/ch19_the_certificate_that_ages.md:1-95`, `observation-theory-campaigns/experiments/FRESHNESS-PROGRAM.md:1-40`, `observation-theory-campaigns/experiments/RADIO-FRESHNESS-TRACK.md:33-39`, `lean/DataMiningAsObservation/Certificate.lean`
Reviewed	not yet reviewed; generated 2026-09-10 from records at the commits on the provenance page.

A claim about a read at a time, and a witness that checks it later.

Equation

Book equation 0.26.

\[\mathrm{FC}=\Pr\big[W\ \text{refutes}\ \big|\ \mathcal C_t\ \text{clears}\big],\qquad \text{coverage}=\Pr\big[\mathcal C_t\ \text{clears}\big].\]

Book equation 9.3.

\[\hat\mu=\frac{\bar s_{\mathrm{true}}-\bar s_{\mathrm{distr}}}{\sigma_{\mathrm{distr}}},\qquad \mu_{\mathrm{crit}}=\mathbb E\Big[\max_{N-1}\mathcal N(0,1)\Big],\qquad \rho=\frac{\hat\mu}{\mu_{\mathrm{crit}}},\qquad \rho=1\ \text{vacuous}.\]

Book equation 11.2.

\[\begin{gathered} r=\frac{d_{\mathrm{compressed}}}{d_{\mathrm{exact}}},\qquad \kappa_{\text{strict}}=\frac{\max r}{\min r},\qquad \tau\ \ge\ 1-2\hat\mu(\kappa_{\text{strict}}),\qquad \rho_S\ \ge\ 1-3\hat\mu(\kappa_{\text{strict}}), \\ \kappa_{97.5/2.5}=\frac{q_{97.5}(r)}{q_{2.5}(r)}\ \text{ gives the same two expressions as estimates, not floors.} \end{gathered}\]

Book equation 13.1.

\[\mathrm{FC}=\Pr\big[W\ \text{refutes}\ \big|\ \mathcal C_t\ \text{clears}\big],\qquad d_O(\Delta)=\operatorname{tr}\big(P_C\,M_{\mathrm{drift}}(\Delta)\big),\qquad M_{\mathrm{drift}}(\Delta)=\mathbb E\big[\delta_\Delta\delta_\Delta^{\top}\big].\]

Conditions

A certificate is graded only where a witness exists, and where none exists it is unassessed rather than safe.
A certificate is bound to what it certified, the bytes, the anchor sample, the footprint, and the time, and carries none of it to another corpus, workload, or moment.
A vacuous certificate proves nothing about the object. It does not prove that no code preserves rank or that no structure exists, only that this test could not certify one.
The rank certificate’s floor is a guarantee in its strict setting and an estimate in its percentile setting.

Conditions are curated in entries.toml rather than read from a record.

Ledger

measures. GO-3 [demonstrated]. The certificate’s vacuity threshold predicts where single-stage retrieval dies. geometric-observation/claims/LEDGER.md:65.

First stated

Volume 14, chapter 19 for the certificate that ages and chapter 9 of Data Mining as Observation for the clustering certificate. The rank certificate is specified in turboquant-pro/docs/CERTIFICATE_SPEC.md, DOI 10.5281/zenodo.20660087.

Measurements

Where the book states it	Numbers, as the book’s sources table records them	Source
chapter 9 section 9.3	the margin certificate, mu crit as the expected maximum of N minus 1 standard normals, rho, death at 0.948 within 6 percent, Spearman 0.991 vs 0.873, fourteen corpora, six gates, the v1 to v3 path, the standing correction	`geometric-observation/experiments/GO3-certificate-vacuity-v3-NOTES.md:1-60`; `geometric-observation/claims/LEDGER.md` row GO-3
chapter 13 section 13.2	the grammar, certificate, witness, refresh floor, false-clear rate, vacuity, the witness table	`geometric-observation/chapters/ch19_the_certificate_that_ages.md:1-95`; `observation-theory-campaigns/experiments/FRESHNESS-PROGRAM.md:1-40`

Failures and corrections

observation-theory-campaigns/experiments/RADIO-FRESHNESS-TRACK.md:33-39 at f7b7c77. ⚠️ XPROTO-AICSI scope correction (2026-08-25). The v1 seal stands for what it tested, but the reconstruction-vs-consumer dissociation does not survive the community-standard substrate. On real 3GPP CDL-C with a CsiNet-class codec the NMSE-optimal codec reconstructs near-perfectly (NMSE ≈ 0.03) and false-clears 0.0 on all seeds; the pre-registered kill fired. See analysis/aicsi/PREREG-XPROTO-AICSI-V2.md (REFUTED AT SHAKEDOWN, NOT SEALED, kept negative). This is a scope correction, not a retraction. Do not headline the AICSI row; WCNC §III-C is dropped.

Invariance envelope

none declared

Machine checked

lean/DataMiningAsObservation/Certificate.lean, theorems falseClear_mul_coverage, coverage_empty, falseClear_mem_unit, minOverStrata_passes_iff, minOverStrata_le_weighted_mean, at observation-data-mining f3914f0; what the check covers is stated in the book’s appendix C.

Used in

Data Mining as Observation chapters 0, 1, 9, 11, 12, 13, 14.

witness; false-clear rate; coverage; rank certificate; vacuity threshold; coherence time.

Status

Generated 2026-09-10 by encyclopedia/generate.py; book at observation-data-mining f3914f0; the commit of every record is listed in the encyclopedia’s provenance.

← central limit theoremchallenge set →