Artificial Wasteland · the recorded chamber
Pull one bell. A real passage from the published parliamentary record arrives, followed by the record's own response. Then open the drawer and go to the source.
Odd pulls visit a declared cabinet of 12 named specimens. Even pulls check it against a deterministic equal-probability sample of 256 rows from the adjusted 207,028-token XML census.
Mr. Quibell · immediately before
It did not seem to be very noticeable while I was upstairs.
the record answers
[An HON. MEMBER: “You were asleep.”]
named cabinet specimen · 1936
The notation as a lexicon
Parliament's Hansard staff wrote in a 2019 account that the report has “a small repertoire of stage directions”. They did not quantify that repertoire. This page does. We are counting a reporting instrument, not the House.
The digital text really does contain [An hon. Member made some observation, which was inaudible.]
, [HON. MEMBEKS: “Scandal.”]
, and [Interruptian.]
. The first two are quoted specimens named by the research brief; the bell cabinet uses only rows for which this release can carry its retained source fields.
1919 to 2026, every file in the frozen ParlParse XML frame: 20,084 files, 0 parse failures. Of the exact forms, 88.69% occur exactly once.
One exact form, Interruption., carries 44.40% of all adjusted tokens. Concentration rises quickly:
The plots are separate because the dissociation is the result. Token volume moves around; the inventory of exact forms contracts toward a narrow set of stamps. The comparison remains about the recorded notation under changing reporting and data practices.
Surface means surface. Interruption., Interruption ., and Interruption stay separate. The table is rebuilt from the shipped form counts.
| rank | exact surface | tokens | share |
|---|---|---|---|
| 1 | Interruption. | 91,920 | 44.40% |
| 2 | Laughter. | 9,301 | 4.49% |
| 3 | Interruption . | 7,682 | 3.71% |
| 4 | HON. MEMBERS: "Oh." | 5,393 | 2.60% |
| 5 | HON. MEMBERS: "No." | 4,680 | 2.26% |
| 6 | Interruption | 4,003 | 1.93% |
| 7 | HON. MEMBERS: "Hear, hear." | 3,949 | 1.91% |
| 8 | Laughter . | 1,710 | 0.83% |
| 9 | HON. MEMBERS: "No!" | 1,564 | 0.76% |
| 10 | No. | 1,366 | 0.66% |
The XML path finds bracketed editorial cues and whole anonymous-attributed speeches. Their inventories are shown separately before the union.
| device | tokens | exact types | type/token | top 1 share |
|---|---|---|---|---|
| bracketed editorial cue | 191,334 | 28,425 | 14.86% | 48.04% |
| anonymous attributed speech | 15,694 | 4,552 | 29.00% | 8.70% |
| union | 207,028 | 32,975 | 15.93% | 44.40% |
The prediction we did not make
Marc Alexander published a periodisation in 2023 from publishing history, never from reaction notation. We fixed his six boundaries as inputs, then tested three-year windows on either side using exact bracket surfaces per sampled sitting day.
3 of 6 supported: 1889 at p = 0.0013, 1898 at p = 0.00006, and May 1899 at p = 0.00424. Not supported by this test: 1877 at p = 0.533, March 1892 at p = 1.000, and 1909 at p = 1.000.
The 1898 and 1899 windows overlap and both contrast with the unusually dense 1897 sample. They are not independent confirmations.
1803 to 1918 is a seeded sample of up to 15 sitting days per year, seed 20260903. It emitted 8,473 candidates. 1919 to 2026 is a census of the frozen 20,084-file XML frame, with 0 parse failures. The 1919 seam stays visible. The change to the Official Report at 1909 is a second hard seam inside the sample.
On the fixed 120-file XML check, bracket-route agreement is 98.64% Jaccard. Anonymous attributed speech falls to 47.18% Jaccard. The low route value is not averaged away.
The corrected 27-sitting-day control has 327 shared of 335 union candidates, or 97.61%. Its first run reported 89.85% because a filename bug skipped unsuffixed 2003 XML files. The control went red because the control was defective.
The conservative audit marks 8,554 of 224,023 rows, or 3.82%, as definite false positives under declared rules. OCR-equivalence sensitivities collapse the raw device-surface inventory by 0.103% to 0.308%. That interval is not a general OCR error rate.
A cue is not an acoustic event. Adjacency is not causation. A laughter marker does not certify humour or intent. This is not one complete 1803 to 2026 census, not proof that the chamber became quieter or more orderly, and not a claim that OCR ghosts such as HON. MEMBEKS, HON. MEMBEES, or Interruptian. were spoken or printed that way. Hansard is an edited, substantially verbatim report, not a phonographic transcript.
Loading the shipped payload for an in-browser check.
The gift
As of 3 September 2026, under the search declared on this page, we could not find a public, machine-readable concordance of UK parliamentary reaction cues that types each cue and carries its source provenance.
Use the Hansard Reaction Notation Concordance. The schema, frozen predicate, exclusions, sources, validator, and limitations are published for readers who never visit this page.
The bell does not ship 224,023 rows to your browser. It ships 256 deterministic probability-sample rows from the adjusted XML census plus 12 named cabinet specimens. The sampler retains the smallest SHA-256 ranks of a fixed seed and each row's source identity. Cabinet predicates and their first-match rule are written in sample-page-payload.mjs. Odd and even bell pulls never disguise one pool as the other.