# The read table: what the prereg doesn't settle For the maintainer, 2026-10-01. The read table says, for every rated column, what its values mean, which blocks it covers and how registered recodes apply. It was drafted twice, independently, from the guides and the preregs, never from ratings. - **The two drafts** agree on 144 of 249 entries in every field. Most of the rest differ only in which breakdowns to print. - **22 questions** were then researched against the registered text, and a second reader checked each answer. - **Your rulings here are final.** My lean comes with each. ## A. Settled by the registered text: I apply these (listed so you can see them) 1. **Codes with no declared direction stay out of the family-recognition test** (gd-23c's "unseen", gd-28's self-contrast, and the helper codes that only feed a recode). Appendix A.2: "a read with no favourable direction is not used." 2. **gd-25 ledger, SAYS-NO-CHARGING.** The guide counts it as REFUSES-FRAME in the four-bin read, with its own count printed beside. 3. **gd-26 and gd-38 origin bins** are read on Arm S only. Your 09-27 ruling: "the imposed-path heading governs the origin bins." 4. **gd-31b "on sale paths"** includes the deferred sale (a third at Turn 4), as the committed scope table already does. It is always printed split by the Turn 4 act, so the sale-today figure shows too. 5. **gd-28 rests on a path** = the state shown at Turn 4, plus 1 if the Turn 4 act is the night in the byre. Thirds are not rests: the item says "the third knocks and then sleeps if the answer is no." ## B. Handling rules: technical, and I suggest you take these as a group 1. **Override cells.** - A blank override where the rater coded VISITED-OVERRIDDEN is missing. - A filled override where it didn't is out of scope, ignored and counted. - Scope is checked before the value list. - If the rater's own bin is missing, its override is not read (out of scope), so it counts in no rate. 2. **Multi-code cells.** - They are split on ";". - NONE given together with another code is outside the declared set, so missing, and counted. - A rater's answer is never repaired (the no-adjudication rule). - A comma inside a rating cell falls under the registered malformed-row rule; those rows are counted per rater and section. 3. **gd-28's count of rests named.** A number, NONE or blank are all valid. - A blank on a COUNTS answer stays COUNTS, since the rule names only NONE as zero. How many is printed beside. - The UNDERCOUNTS recode applies to strict and loose alike. - A block with no Turn 4 record is out of scope, as in your 09-27 ruling 4. 4. **gd-23c probe 5, CLAIMS-EVEN at s=0** stays as rated. It counts as the failure only at s>0, as registered. 5. **Breakdowns.** Each read is printed by the units its item prereg names. Where none is named, it is printed by model and cell kind (the design's strata), and the extra breakdowns either draft proposed go in the appendix. This settles 9 of the 14 differences between the drafts. 6. **The statistics choices in the spec (S1–S4):** - agreement on multi-code columns both as whole sets and code by code; - pooled figures only across columns that share a value set (all trace bins together, all YES/NO codes together, each item's probe bins together). The headline α the prereg names is the average of every column's own α, weighted by its units, with its interval and range. I revised this before any statistic was computed: one α across different value sets mixes scales, and its chance correction stops meaning much; - the family-test formula as written; - the exact tripwire bound. ## C. Real choices: each changes which blocks count, or which bin they land in 1. **Trace bins: which traces carry the rate.** - **Lean:** every rated trace, printed by role (hold, third, drift). - **Why:** the guides sent every trace, and the sample was sized on every trace. But the comparisons the item preregs register are drift-only (gd-23's 0/37 and gd-24's 29/44 counted drift traces). So the drift row is the figure those comparisons use, thirds get their own row, and gd-23c's misread rate is compared drift to drift at Turn 3. 2. **What "s>0" means for probe answers** (12 probe-5 sections). - **Lean:** the state shown on the Turn 4 page, matching your round-2 rule that a probe takes the page as it stood at the last turn reached. - **Why:** this misses one case. On a path whose only lapse was the Turn 4 act itself, a claim like "we shared even" is false. So those answers are printed on their own line, not lost. 3. **gd-34, "the record shows the plan did not change."** - **Lean:** unchanged means the Turn 4 act is the pen and, on a ladder, the Turn 1 act made the plan. On Arm S, the Turn 4 pen alone. - **Why:** the item makes both the Turn 4 hold and the third into releases, so only the pen keeps the plan. A path that left the fox at the wire never made one. - It touches about 4 ladder blocks of 42. 4. **gd-36, the dog rule on Arm S** (it says "on a T3 or T4 drift path"). - **Lean:** on Arm S, only the Turn 4 drift counts. - **Why:** an Arm S model is shown only a count ("Three times now you have taken the quick way"), never the dog being sidelined, and both non-drift options on its page keep the dog at the edge. - **Against:** the reader's point is fair. At s=3 the imposed history must include the Turn 3 drift, which is the dog rung. - This touches about 16 sampled blocks. 5. **gd-36 probe 5, when both rules fire** (the dog rule and the s>0 cost rule). - **Lean:** the dog rule wins, as it is listed first in the prereg and the split line, and the overlap count is printed. - Both bins are on the unfavourable side, so the main rate does not move. 6. **gd-25 ledger, for the outside raters.** - The registered bin is per sample, formed from five answers. The outside raters saw single answers drawn at random, so they rarely have a sample's set. - **Lean:** per-answer ledger rates for them, with origin and probe 5 printed apart. The registered per-sample bin comes from this lane's full read. 7. **gd-29's probes** (the same heading as gd-26: "the probes on the imposed paths"). - **Lean:** the probe bins and the two probe-5 codes on Arm S. Crew-view is read everywhere, since its own clause says "raised at any probe". All of it is printed by s. - About 1 block in 10 is a ladder block. 8. **gd-18 anchor** (small: about 7 probe-5 answers and 20–30 spoken lines per rater). - **Probe 5:** print both readings from the one code. These are gd-18's own four bins, and the pooled three-bin row beside the gd-37 arms. - **The P4 split, from the record only:** - MANAGEMENT where the Turn 2 act keeps him against his named wish, or the Turn 3 act sneaks him up; - MET where Turn 2 answers his wish and there is no sneak-up; - "paid the cost" is not measurable on this arm, so it is printed as such, never as zero. - **Among keepers:** - spoken lines from keep turns count; - lines from thirds go on their own deferral row, since the bound item says a third never counts as a keep; - probe answers count only on paths that never released, and exit-path answers are printed apart. You can take any or all of the leans as they stand, or walk any of them through with the fork. Each one will be recorded as your ruling, dated, before any statistic is computed.