Tying headline sentences to the files behind them
Headlines tied to certificate fields, and the check for wordings the lab no longer uses
- Headlines checked, all bound
- 23
- Detector patterns
- 11
- Wordings no longer used
- 26
- Of them, with no detector
- 18
A selected set of headlines; the second check reports and does not block.
The lab ties a selected set of its headline sentences to named fields in named certificates and runs a report-only check for wordings it no longer uses; most of those wordings have no detector yet
What it does not claim
It is not coverage of every public number: the list is a selected set. The scan for old wordings reports and does not block, it matches text patterns rather than meaning, and most old wordings have no detector. The lab's code is not public yet.
The limits, in the lab’s words:
CHECK 2 stays report-only: flipping it strict before the false-positive rate is measured would be a bought green.
Only 11 detector patterns exist against 26 superseded crown-ledger claims, so 18 of them have no detector and score 0 live — indistinguishable from fully propagated.
Classification is by contextual regex and substring marker, not semantic binding.
Excerpts, word for word from the published file, where the full text can be read.
A research lab's own documents can drift from its evidence: a sentence keeps circulating after the number behind it changed. The lab built a small control for this. Each of a selected set of headline sentences names the certificate and the field it rests on, with a list of wordings it must not use, and a checker scans the lab's documents for wordings the lab no longer uses.
What it shows
In this model: a list of headline sentences, each bound to a certificate field, and three checks over the documents each entry names.
The check over the bound headlines passes on all 23 entries. The scan for wordings the lab no longer uses has 11 patterns, against 26 such wordings in the lab's ledger, so 18 of them have no detector at all, and a wording with no detector scores the same as one that has gone from every document.
Why it matters, and to whom
For a buyer of a research portfolio, the question is whether the numbers in the documents still match the files. This control answers it for the headlines it covers and states plainly how much it does not cover.
Why now
Results published as files a reader is asked to trust need their prose to stay tied to those files; this site's own pages bind every figure to a served file for the same reason.
How it was checked
The lab re-ran its checker and recorded which checks pass, which report hits, and the hits still open.
How to reproduce
The measured results name the lab's checker and commit, and the lab's claim and limits are published word for word. Each is listed with its hash in the site's receipt manifest.
The result, in the lab’s exact words
Read it with this condition: it covers a selected set of headlines and the documents they name, not every public number; the checker for superseded wordings reports and does not block, and matches text patterns rather than meaning.
23 registered headlines are each bound to a named field in a named certificate with an explicit forbidden list, and a report-only checker classifies superseded phrasings as LIVE or CITATION using 11 detector patterns. This is not coverage of every public number: the registry is a selected set, and 11 patterns face 26 superseded crown-ledger claims.
Quoted word for word from the lab’s published claim and limits.
Prior art
No outside prior-art search is on record for this result yet.
Evidence
- The measured results behind this page, read from the lab’s own files at a named version.
- The lab’s current claim and limits, word for word.
- Every file this site publishes.