The corpus
Counts from the frozen critical releases. Sources, readings and physical objects are different units and are never added together.
- physical supports
- 855
- lead sheets, stelae, vessels, coins and graffiti identified in the frozen critical corpus
- edition observations
- 17,128
- separate scholarly readings, kept apart rather than merged
- analysable word tokens
- 2,089
- after duplicate collapse and cleaning
- of word types occur once
- 89.8%
- the hapax rate that limits every statistical method
- supports with date candidates
- 311
- source-reported intervals; far fewer are securely dated
Progress
Commits per day, 22 Sep – 13 Oct 2026, with each milestone marked. Every commit saves a source check, an analysis or its tests. Independent bilingual lexical anchors remain inconclusive; historical duplicates and disputed restorations supply active testable phonetic and participant-role questions. Neither solution nor exhaustion.
Show as a table
| Day | Commits |
|---|---|
| 22 Sep 2026 | 2 |
| 30 Sep 2026 | 1 |
| 02 Oct 2026 | 1 |
| 03 Oct 2026 | 1 |
| 05 Oct 2026 | 7 |
| 06 Oct 2026 | 106 |
| 07 Oct 2026 | 135 |
| 08 Oct 2026 | 85 |
| 09 Oct 2026 | 48 |
| 10 Oct 2026 | 41 |
| 11 Oct 2026 | 4 |
| 12 Oct 2026 | 66 |
| 13 Oct 2026 | 36 |
Research foundation
Mapped the corpora, bibliography, transcription conventions, scholarly debates and earlier computational work. 65 bibliography records, 25 archived sources, 20 attributed hypotheses.
Critical corpus
Built and froze a verified corpus: 15,428 source records and 17,128 edition observations on 855 supports. Readings, alternatives and interpretations stay in separate layers.
Blind structural study
Analysed the corpus with the scholarship withheld, then compared. Five results independently rediscovered published findings. None was new with high confidence; the corpus size is the binding limit.
Evidence expansion
Added a sign-level corpus on 705 supports, primary-edition reviews of three long inscriptions, and dating candidates for 311 supports.
Measurement audit
Found that the earlier negative morphology tests could not detect the structure they looked for. Repaired the methods; one narrow comparison shows organisation beyond a simple baseline, not established morphology.
Candidate adjudication
Tested whether surviving candidate word families are real. Line breaks explain many; lauŕberton-ar/-te survives as a credible, already published pairing.
Source-critical continuation
Inscription-by-inscription work on specific endings and constructions, rechecking each reading against original editions and photographs before testing a grammatical model.
Open questions
The 8 questions at the front of the work. Open one to see the test that would move it forward. 15 source tasks are queued behind them.
te ekiarWhat exactly does -te contribute in the source-secure NP-te + ekiar construction?
Next test
Keep likine-te and śalaiaŕkis-te as morphology anchors, but seek independent argument-role evidence. Do not use T.07.01 as a secure signer/sender semantic anchor.
erok categoryAre eŕoka nominal/participant forms and eŕok/ŕokan predicate candidates derivationally related, category-flexible, homonymous, or over-grouped?
Next test
Boundary evidence now supports śalaiaŕkis-te-ŕok-an. Test finite predicate versus verbal noun or participle using Tivissa, La Serreta, Pech Maho and Orleyl contexts with independently anchored participants or quantities.
np relationsWhat selects -en versus -ar/-an in relational/property constructions?
Next test
Use property contexts with independently fixed name bases and the new T.00.01 quantitative-ledger constraint; do not promote its attributed -en/-ika segmentation to an exact-base proof.
mi inflectionCan NP-(en/ar/an)-ḿi property formulas and Joncosa ḿi/ḿie distinguish pronoun, copula, deictic and clitic models?
Next test
Compare eŕokarḿi/eŕokaḿi to source-secure property and memorial constructions without assuming eŕoka is globally nominal.
ai concordDoes -ai mark group concord, linkage or phrase-level scope?
Next test
Use exact baŕbinke/baŕbinkeai plus uskeike/uskeikeai as whole-host anchors; test whether -ai scope follows construction rather than lexical identity.
iunstirCan iunstir's epistolary/opening and production contexts distinguish formula, predicate and discourse-marker analyses?
Next test
Source-critical Rébé, de Hoz & Orduña 2012 p221–222 makes Ruscino outer baŕete possibly a cross-line Gaulish compound-name continuation; inspect glyph transition, possible added e, restored [ti]/[u], and inner damaged following -te before treating it as NP-te+predicate. Still pursue Ger/Badalona baŕbin source and independent role-controlled iunstir cases; all prior material exposed.
quantity anchorsWhich numerical values and unit ratios are externally fixed, and can they constrain clause roles?
Next test
Retain erder only as a restricted half-value anchor. Verify undikesken CNH 17/18/24 at plate/accession level and seek a non-coin erder amount with independently known half relation to discriminate ordinary lexical HALF from a half-denomination label. Continue independent a:o:ki and coin/weight anchors without using disputed lexical numeral values as calibration constants.
bilingual tebanenDo F.11.8 and independently controlled funerary eban/teban constructions fix a morphological role without assuming line-for-line Latin/Iberian translation?
Next test
Obtain sign-level F.11.8 text and exact participant identity; locate V.04.08 outside extracted search snapshot and test kinship versus production/commission with independent role anchors.
Latest findings
Newest first, from the research state of 13 October 2026. Most narrow what can be claimed. None is a translation.
A 2026-10-13 source-critical baika crosswalk verifies that Ferrer et al. 2022 wrongly reuse valid unrelated GI.17.01/02 and GI.15.09 catalogue IDs for Vilademuls GI.07.01/02 and Ullastret GI.15.31.
Detail
A new original-edition comparison shows GI.07.02 a/b zones are inverse in historical Hesperia versus Velaza 2019, 2022 internally disputes Ullastret cup Attic versus common-Iberian fabric, and 2017/2018 Sant Julià skyphos GI.07.06 disagrees with 2022/current GI.13.01. These are exposed source-identification constraints, not extra ancient attestations or language predictions; whorl PN-ar vs PN-tigiar remains open.
Sabaté 2023 pp122-123 already reports four A lines and explicitly derives B reading from Ferrer 2021 n53, but prints beduginede versus 2021 betuginede.
Detail
This is a derivative edition difference, not ancient spelling evidence. Ferrer 2021 external face must not be conflated with fold-visible address; de Hoz 2011 p423 fold-hidden objection is only secondhand through Sabaté; SEND/MESSAGE semantic proposals are not translations.
V.07.02 full-face A four-line exposed source re-audit: 18 colon-delimited units, terminal gidibabiŕbede; universal orthographic -de immediately before iuśdir fails.
Detail
The same one lead has B(a) betuginede:iuśdir:kutuŕde and separate uncertain B(b), not two independent witnesses.
12 Oct 2026 palaeographic source reinspection: GI.16.02/Can Fontgrau previously normalized as kai is actually +i with first graph l/tu/ka unresolved in original 2019 edition and 2022 Fig11, so six Hesperia exact kai strings are not six independently physically secure ka-i tokens.
Detail
Original 2018 B.44.06 autopsy supports clear two-sign kai, while Ensérune HER.02.327 kaie is a separate amphora with no independently matched name/role; Caius vocative, Iberian short mark, kai-e morphology, whole-name adaptation and unrelated labels remain competing models. The 1613-row source census and local five-case fixture checks test editorial source integrity, not grammar.
Primary Rébé–de Hoz–Orduña 2012 Ruscino autopsy pp.216,221–222,226 explicitly entertains PYO.01.21 baŕete as continuation of the preceding written line and second element of a Gaulish personal name, queries whether its e was later added, and permits iuns[ti]ŕ onomastically rather than as a complete greeting/predicate.
Detail
PYO.01.22 i[u]nstir requires restored [u] and has damaged putative follower -te. Two ancient leads are nested in one find packet, and neither printed full iunstir root is unbroken on the edited preserved glyphs. Thus Ruscino does not independently anchor a complete bare PN-te agent plus finite iunstir clause, although both formula and predicate hypotheses remain live. All source material is EXPOSED, not blind.
Recent commits
- Carry source-critical tagiar model alternatives and next autopsy tasks
- Add locally tested adversarial source-family census audit
- Pin source readings, alternate editions and stamp-dependency groups for r14
- Compare source-autopsied Ensérune en/tagiar stamps and disputed Vilademuls tigiar model
- Record pending Ensérune tagiar source reinspection recovery patch
- Source-recheck 2017 Vilademuls double-ring and test alternative tigiar parses
- Carry baika source-critical findings and next autopsy tasks into continuation state
- Add executable tested baika catalogue crosswalk and pinned evidence
Ground rules
- Published readings, alternative readings and interpretations are recorded separately.
- Damage, separators, line breaks and script variants are preserved. Unknown is never treated as absent.
- Comparisons with Basque and Aquitanian are kept out of the initial Iberian-internal analysis.
- A failed route is reported as a result. Nothing is called solved or exhausted on the strength of one test.