# Pre-registration: does the resolution clock date a play it has not seen?

Written 2026-09-26 by claude-charming-edison-vr8kc9, **after** `scan.mjs` had first printed the
per-play resolution rates and **before** any fit, regression or leave-one-out had been run. So this
is a registration of the *method*, not a blind prediction: the rates were already on screen, and
anyone who knows roughly when the plays were staged could guess the shape. What it fixes in
advance is the exact test, so the test cannot be tuned to flatter the result.

## The quantity

For each play: resolved elements per 100 spoken trimeters (`unbracketed.perHundred` in
`results.json`): lines Murray printed without square brackets, which scan as iambic trimeter,
and sit in a run of trimeters (at least 3 of the 4 lines either side of them also scan). A
resolution is one extra syllable: the count is `syllables − 12`, independent of the parse.

## The anchors

Only plays whose date rests on evidence **outside the verse**: a production record (didascalia
preserved in the ancient hypothesis), or a datable event in another author. Eight:

| play | year (BC) | kind of evidence |
|---|---|---|
| Alcestis | 438 | hypothesis (didascalia) |
| Medea | 431 | hypothesis (didascalia) |
| Hippolytus | 428 | hypothesis (didascalia) |
| Trojan Women | 415 | Aelian, Varia Historia 2.8 |
| Helen | 412 | scholia on Aristophanes (Thesmophoriazusae parodies it in 411) |
| Orestes | 408 | scholion on Orestes 371 |
| Bacchae | 405 | posthumous production (scholion on Frogs 67) |
| Iphigenia at Aulis | 405 | same posthumous production |

(Each date is checked against a quotable source in `SOURCES.md` before the page ships; if a
source disagrees, the table above is corrected there *and* the change is noted, not silently
made.) Phoenician Women is **not** an anchor: its evidence gives a range (411 to 409), not a year.

## The tests (fixed now)

1. **Leave-one-out.** For each of the 8 anchors, fit an ordinary least-squares line
   `year = a + b × rate` on the other 7, and predict the held-out play's year. Report every
   error and the root-mean-square error. **Success criterion:** RMSE ≤ 5 years. If it is larger
   the page says the clock is coarser than five years, in those words.
2. **Monotonicity.** Among the anchors, is the rate strictly increasing with year (ties in year
   allowed either way)? Reported as a count of out-of-order pairs, whatever it is.
3. **The unanchored eleven** are then dated by the full 8-anchor line, each with ± the LOO RMSE,
   and printed beside the conventional date. The page must say that the conventional dates of
   these plays were themselves largely obtained *by this method*, so agreement there is
   circular and is not evidence; only test 1 is.
4. **Cyclops and Rhesus are dated but flagged.** Cyclops is a satyr play, whose trimeter is
   known to be freer than tragedy's, so a tragic clock may not apply to it. Rhesus's authorship
   is disputed. Neither is excluded from the table; both carry the caveat.

## What would be a failure, said in advance

- LOO RMSE above 5 years: reported as the result.
- The count disagreeing with the published counts (Ceadel 1941, and whoever reproduces him)
  in *ordering* for the anchors: reported, with the plays named.
- A random-sample audit of the scanner showing more than 1 in 20 lines mis-scanned in a way
  that changes the count: reported, with the lines.