Repository path: workshop/experiments/E-20260729-drift-window-verify/design/sites-frozen.md · rendered 2026-09-09
Page metadata (front matter)
| type | manifest |
|---|---|
| id | E-20260729-sites-frozen |
| status | frozen |
| created | 2026-07-29 |
| updated | 2026-07-29 |
| senses | style-correspondence, accuracy |
| internal-judgment-only | true |
| links | wiki/base/anchors/A-beowulf-ingeld/A-beowulf-ingeld.md, wiki/findings/claims/CL-20260726-drift-window.md, workshop/experiments/E-20260729-drift-window-verify/design/design.md |
Frozen reflex-site census — Bēowulf ll. 1800–1866 (H&S)
This file was written and committed to git before the passage was translated by anyone in this project and before any published rendering of these lines was opened. That order is the point of the file, and it is the one thing about it that a later reader cannot reconstruct from the content. The commit is the evidence.
A-beowulf-ingeld §4.3 enumerated its fifteen sites after reading all four renderings of its
passage. This census is enumerated from the Old English and the edition's own glossary alone.
1. The inclusion rule, stated because the anchor did not state one
A-beowulf-ingeld §4.3 says only "fifteen sites in 56 lines where modern English holds the Old
English word's descendant form", and separately lists eleven words that "copied free, with no
decision at all" (medu, ealu, bēor, gold, īren, land, word, blōd, fæder, sunu,
dohtor). So an inclusion rule was operating and was never written down. Restated here, and it is a
repair of the anchor, not a new instrument:
A lemma of the passage is a site iff all four hold:
- (a) Survival. A modern English descendant of the Old English word, or a transparent derivative of it, is a current word of English — continuously inherited, not a later re-borrowing.
- (b) Recognisability. The Old English form and the modern form are transparently the same word to a reader who is not a philologist. This is what makes a translator tempted; a reflex nobody would recognise cannot be a false friend.
- (c) Not regional-only. Reflexes surviving only in a regional dialect are excluded (
lēod→ lede,bearn→ bairn, `drēogan → dree*). They are unavailable to these translators as unmarked English and their absence is not a decision. - (d) Using the reflex is a decision. The modern descendant, substituted at this site, does not
give correct current unmarked English. Excluded therefore: reflexes whose form and sense are
both current and right here (
sunne→ sun,hrefn→ raven,sweord→ sword,īren→ iron,frēond→ friend,word,grund,cyning,gold). Those are the anchor's "free copy" set and they are not sites.
Unit = lemma, not occurrence. This follows the anchor, whose table groups 2027 / 2048 under one
wine row and 2021 / 2036 / 2032 under one duguð / dugan row. Fifteen rows there are fifteen
lemmas. Every line number at which a lemma occurs is listed below.
2. Drift classes, carried verbatim from the anchor
Drift is classed total where the modern sense has no overlap with the Old English one, partial where the modern sense is wrong but locally survivable.
One class is added, because the anchor needed it and did not name it: marked — the modern
descendant carries the right sense but is restricted to archaic or historical register, so taking it
is still a decision. This is the class the anchor's þegn row occupies ("none, marked archaic").
The lead's classification below is recorded so that it is on the record before the run. It is not the classification the result will be computed on. Independent raters classify every lemma blind to this file, and the totals are recomputed under each rater's own classification (design §4).
3. CLEAR sites — the primary set
23 lemmas. "CLEAR" means the lead judges (a)–(d) to hold without argument.
| # | line(s) | Old English | H&S gloss | modern reflex | lead's drift class | replicates anchor site |
|---|---|---|---|---|---|---|
| N1 | 1803 | blīðe (in blīð-heort) |
blithe, joyous; gracious | blithe | partial | — |
| N2 | 1803 | bodian (bodode) |
to announce, make known | bode | partial | — |
| N3 | 1804 | sceacan (scacan) |
to move quickly, hasten | shake | partial | — |
| N4 | 1805 | æðeling (æðelingas) |
nobleman, man of noble descent | atheling | marked | — |
| N5 | 1806 | faran (farenne) |
to go, travel | fare | partial | — |
| N6 | 1807 | cēol (cēoles) |
ship | keel | partial | — |
| N7 | 1808 | heard (hearda) |
brave, hardy | hard | partial | — |
| N8 | 1810 | lēoflīc |
dear, precious, beloved | lovely | partial | — |
| N9 | 1811 | cweðan (cwæð) |
to say, speak | quoth | marked | — |
| N10 | 1811 | wine (in gūð-wine) |
friend, lord | wine | total | yes — 2027 / 2048 |
| N11 | 1813 | mōdig |
spirited, brave, high-hearted | moody | partial | sibling of 2045 mōd |
| N12 | 1817 | dēor (in hilde-dēor) |
brave, bold | deer | total | — |
| N13 | 1822, 1840 | dugan (dohtest, dēah) |
to avail, be good, be capable | doughty | partial | yes — 2021 / 2036 / 2032 |
| N14 | 1824, 1845, 1854 | mōd (mōd-lufan, mōde, mōd-sefa) |
mind, spirit, heart, courage | mood | partial | yes — 2045 |
| N15 | 1830 | þegn (þegna) |
thane, retainer | thane | marked | yes — 2060 |
| N16 | 1833, 1850, 1856 | folc (folces, folcum) |
nation, people | folk | partial | yes — 2018 |
| N17 | 1833, 1850 | hyrde |
keeper, guardian, shepherd | herd | partial | yes — 2028 |
| N18 | 1842 | wittig |
wise | witty | partial | — |
| N19 | 1855 | līcian (līcað) |
to please (impersonal) | like | partial | — |
| N20 | 1858 | sacu |
strife, contention, lawsuit | sake | total | — |
| N21 | 1860 | wealdan (wealde) |
to rule, govern, have power over | wield | partial | — |
| N22 | 1865 | fēond |
enemy, foe | fiend | partial | — |
| N23 | 1866 | wīse (wīsan) |
manner, way, custom | wise (n.) | total | — |
Lead's class totals: 4 total · 16 partial · 3 marked.
Six of the twenty-three are the same lemma the anchor scored (N10, N13, N14, N15, N16, N17), which is not a design choice but a property of the poem's vocabulary — and it means part of this test is a within-lemma replication in a different passage, with different translators' sentences around it.
One classification the lead expects to be argued with, recorded in advance because it is exactly the
kind of call the raters exist to make. N22 fēond → fiend is classed partial and not total:
a fiend and a foe overlap in hostility, so "both with fiend and with friend" is locally survivable
in the anchor's sense. Under the opposite call it becomes a fifth total-drift lemma and the primary
statistic changes. N23 wīse is classed total although the noun survives inside otherwise,
likewise and in no wise; the free noun does not.
4. ARGUABLE sites — the sensitivity set, excluded from the primary statistic
14 lemmas where the lead judges (a)–(d) to hold but would not defend it against a reader who disagreed. They are enumerated here, before the run, precisely so that excluding them cannot later look like a choice made after seeing numbers. They are scored by the lead only and reported separately.
| # | line(s) | Old English | H&S gloss | modern reflex | lead's drift class | why arguable |
|---|---|---|---|---|---|---|
| A1 | 1800 | rūm (in rūm-heort) |
roomy, spacious | room | partial | drift sits in a compound modifier |
| A2 | 1802 | wynn (wynne) |
joy, delight | winsome | partial | reflex survives only as a derivative adjective |
| A3 | 1804 | sceaða (scaðan) |
injurer, foe; here: warriors | scathe | total | agent noun scather is not current |
| A4 | 1811 | tellan (tealde) |
to count, reckon, account | tell | partial | modern tell retains a reckoning sense (all told) |
| A5 | 1812 | cræft (in wīg-cræftigne) |
strength, skill, power | craft | partial | modern craft retains skill; drift is slight |
| A6 | 1824 | tilian |
to strive for, obtain | till | total | recognisability marginal |
| A7 | 1827 | flōd (flōda) |
flood, sea, current | flood | partial | flood for sea is archaic-poetic, not dead |
| A8 | 1839 | cȳððu (in feor-cȳððe) |
kinship, native land, acquaintance | kith | partial | survives only in kith and kin |
| A9 | 1846 | wēn |
expectation, hope | ween | marked | very obscure |
| A10 | 1853 | hord (in hord-weard) |
treasure, hoard | hoard | partial | modern hoard retains store of treasure |
| A11 | 1854, 1860 | rīce |
kingdom, realm | rich; -ric | total | noun reflex survives only in bishopric |
| A12 | 1855 | lēof (lēofa) |
dear, beloved | lief | marked | obsolete outside had liefer |
| A13 | 1858, 1861 | gemǣne |
common, shared | mean (adj.) | partial | mean = common survives in restricted use |
| A14 | 1858 | sib |
peace, kinship, relationship | sib; sibling | partial | sib alone is technical |
5. What this census does not claim
- It is one reader's enumeration, which is the defect it exists to start repairing, not to have fixed. Note (bcl): a criterion is not shown to define a set until a reader who is not its author applies it. Design §3 spends one API call on exactly that — an independent census over the same 67 lines, blind to this file — and whatever that call returns is reported, including sites this file missed.
- The CLEAR / ARGUABLE split is the lead's, and the primary statistic runs on CLEAR only. Both sets are frozen here together so that neither can be moved afterwards.
- The drift classes above are the lead's and are not what the result is computed on. They are recorded to be compared against the raters', not to be used.
- Nothing here is a quality claim about any rendering, and no rendering had been read when it was written.