# Preservation Typology Plan

> A new dimension of the Formula Dossier: classifying inscriptions by **how much survives and how much is restored**. The same physical fact — extent of damage and editorial supplement — supports two *opposite* significance readings, which the typology makes explicit and quantifies side-by-side.

## 1. The bipolar core axis

The user-named insight is the spine of the typology:

| Pole | Diagnostic | Significance type | Why |
|---|---|---|---|
| **Heavily restored** (large lacunae, many `[…]`, dotted `̣` letters) | Many editorial supplements per surviving character | **Historical significance** | Scholars only bother to restore inscriptions that *matter*. Restoration is itself evidence of inferred historical content — every supplement is a hypothesis about what the lost letters said. The more restoration scholars have proposed, the more historical weight the text has been judged to carry. The dossier's IG II² 1534 + 1535 Asklepieion decrees (case study #6) are heavy in this direction: each preserves an enormous editorial reconstruction. |
| **Near-complete** (no or minimal `[…]`, dotted letters rare, clean continuous text) | Most letters survive on the original stone | **Epigraphic significance** | A complete inscription is a primary witness in its own right: paleographers can read every letterform, philologists can study unbroken syntax, statisticians can count tokens without imputation. The dossier's SC de Bacchanalibus (case study #1, 186 BCE) is the canonical Latin specimen here: bronze tablet, almost zero loss, every letter visible. |

Neither pole is "better" — the two significance types name what the text is *good for*. A maximally-restored inscription is a magnet for prosopography and historical inference; a maximally-complete inscription is a magnet for paleographic and formal study. The dossier's job is to make the position of each record on this axis legible.

## 2. Quantitative metrics

For each record in `inscriptions_index_decrees.jsonl`, compute and store the following derived fields. All metrics operate on the `raw` text field as normalised by the M0 pipeline.

| Metric | Formula | Range | Interpretation |
|---|---|---|---|
| `lacuna_density` | (count of `[…]` / `[— — —]` openings) ÷ (total chars) × 100 | 0–100 | percent-style; >1.0 = visibly lacunose, >5.0 = severe loss |
| `supplement_ratio` | (chars inside `[…]` supplements) ÷ (total chars) | 0.0–1.0 | 0.0 = nothing restored, 1.0 = entirely restored |
| `dotted_density` | (count of combining dot-below `̣` U+0323) ÷ (total chars) × 1000 | 0–1000 | per-thousand-char rate of uncertain-letter dots |
| `complete_run_max` | longest continuous run of un-bracketed, un-dotted text | int chars | larger = stronger uninterrupted-witness value |
| `complete_run_p90` | 90th-percentile complete-text run length | int chars | how typical long undamaged runs are |
| `editorial_apparatus_count` | count of `(?)`, `vacat`, `[?]`, `[—]` markers | int | density of editorial uncertainty signals |

These six metrics feed two **derived scores** that name the two significance types:

| Score | Formula | Bounds |
|---|---|---|
| `historical_significance_score` | normalised composite of `lacuna_density` + `supplement_ratio` + `editorial_apparatus_count` + (log of inscription length, since restoring a long inscription takes more inferential work than restoring a short one) | 0.0–1.0 |
| `epigraphic_significance_score` | normalised composite of `complete_run_max` + `complete_run_p90` × (1 − `lacuna_density` / 100) × (1 − `dotted_density` / 100) | 0.0–1.0 |

Both scores get binned into 5 named tiers for UI legibility:

| Tier | Score range | Label (historical axis) | Label (epigraphic axis) |
|---|---|---|---|
| T5 | 0.80–1.00 | Heavily reconstructed (canonical scholarly investment) | Pristine (a primary paleographic witness) |
| T4 | 0.60–0.79 | Substantially reconstructed | Largely intact |
| T3 | 0.40–0.59 | Moderately reconstructed | Partially intact |
| T2 | 0.20–0.39 | Lightly reconstructed | Fragmentary |
| T1 | 0.00–0.19 | Minimally reconstructed | Heavily damaged |

A record's pair `(historical_tier, epigraphic_tier)` is its **preservation type**. The product space gives 25 cells; in practice records cluster on the **diagonal** (more restoration ↔ more damage; less restoration ↔ more completeness) and the dossier's interesting outliers live **off the diagonal**:

- **(T5 historical, T5 epigraphic)** — vanishingly rare: a complete inscription that has also attracted heavy scholarly inferential investment (typically because it contains a name or formula needing identification). Case-study material.
- **(T1 historical, T5 epigraphic)** — a perfectly preserved but routine bureaucratic text. Common; paleographic gold mine, historically dull.
- **(T5 historical, T1 epigraphic)** — a fragmentary but historically explosive text whose every recoverable scrap has been argued over. The SC Cyrene-style controversial fragments.

## 3. Other typology axes worth adding

The user said "there can be other types." Three additional axes are designed to complement (not compete with) the preservation axis:

### 3.1 Genre stability — how formulaic this text is
| Metric | `formula_density` = (count of detected formula instances) ÷ (total chars / 100) |
|---|---|
| Diagnostic | High = genre-typical (e.g. canonical SC, canonical Athenian decree); Low = genre-atypical (e.g. letter, edict, free composition). |
| Why it matters | A record that scores high on this axis confirms the dossier's central thesis (formulaicity is the genre's grammar). A record that scores low is interesting precisely as a *boundary case* — what makes an inscription stop being a decree? |
| Tier labels (T5→T1) | Hyper-canonical · Canonical · Mixed · Genre-loose · Free composition |

### 3.2 Prosopographic weight — how many named persons appear
| Metric | `proper_name_count` = count of capitalised tokens that match a known Athenian/Roman prosopographical onomasticon (`LGPN`, *PIR²*) |
|---|---|
| Diagnostic | High = densely populated with magistrates, proposers, witnesses; Low = anonymous or impersonal text |
| Why it matters | The two case studies #2 (SCpP) and #6 (Asklepieion) are densely prosopographic — SCpP names 20+ senators, the Asklepieion decrees name archons, secretaries, proposers, and ten-man commissions. Prosopographic density is what enables the dossier to be used as an *historical source* and not just a *formal* one. |
| Tier labels | Heavily personalised · Substantially personalised · Mixed · Sparsely personalised · Impersonal |

### 3.3 Paleographic / dialectal peculiarity — how *unusual* the script and language are
| Metric | composite of (a) deviation from the corpus's modal script for the period, (b) presence of dialect markers (archaic Greek ξ-, Doric -ᾱ, Aeolic -οις-, Latin -onss-, etc.), (c) letter-form anomalies flagged in editorial apparatus |
|---|---|
| Diagnostic | High = local-dialect, archaic, or experimental orthography; Low = chancery-standard for the period and place |
| Why it matters | The dialect axis is where formula stability gets a stress-test: when the same formula bundle is rendered in Eleian wratra-prose or Lesbian Aeolic, does the formula skeleton survive the transposition? The Eleans-and-Heraeans treaty (ML 17, in our test OCR output above) is the diagnostic case. |
| Tier labels | Idiosyncratic · Strongly dialectal · Mildly dialectal · Near-standard · Chancery-standard |

### 3.4 Monumental significance — how the inscription was set up
| Metric | join (place, material, find-context type) against a manually-curated venue table — Acropolis stelai, panhellenic sanctuary slabs, agora tablets, etc. score high; quarry walls and graffiti score low |
|---|---|
| Diagnostic | High = monumentally framed, intentionally public; Low = ad-hoc or private |
| Why it matters | The dossier already has a Geography tab and a Restoration plugin that read material/place. This metric distils that into a single named score that explains *why this inscription was meant to be read* — and therefore *why it was formulaic in the first place*. |
| Tier labels | Panhellenic monumental · Polis-monumental · Locally monumental · Routine documentary · Ephemeral |

### 3.5 Scholarly citation density — how often modern scholarship cites it
| Metric | count of citations in `formulae_catalogue_HRO.md` × weight by number of citing works |
|---|---|
| Diagnostic | High = dossier's hot-list (Tod 108, Tod 123, ML 94 — each cited by 3+ of Henry/Rhodes/RO/R&L); Low = corpus's long tail |
| Why it matters | Independent confirmation of historical significance — modern scholarship's choices are themselves data. A record that is heavily restored (T5 historical) AND heavily cited (T5 citation) is the dossier's most defensible canonical specimen. |
| Tier labels | Citation backbone · Frequently cited · Occasionally cited · Rarely cited · Uncited |

### 3.6 Diachronic centrality — how this text positions in the formula's evolution
| Metric | for each formula UID the record instantiates, distance (in years) from the formula's median attested date; minimum across UIDs |
|---|---|
| Diagnostic | High = the inscription sits at the chronological *centre* of multiple formula traditions (and is therefore a low-risk diagnostic specimen); Low = the inscription sits at a formula's *edge* (and is therefore a high-information boundary specimen). |
| Why it matters | This explicitly couples preservation typology to the chronology dimension — a "boundary specimen" near a formula's birth or extinction date carries different evidentiary weight than a specimen from its mid-life. |
| Tier labels | Genre core · Mainstream · Periphery · Transitional · Outlier-edge |

## 4. The seven-axis dashboard

The full typology is therefore **2 named axes + 5 supplementary axes = 7-axis classification**. Each record gets a 7-tuple of tiers (T1–T5 each), and the UI surfaces both:

- a **radar chart per record** (7 spokes, tier value on each — instantly readable record-character profile)
- a **2-axis scatter** (historical × epigraphic) as the primary projection — that's the user-named bipolar core
- a **filterable corpus view** where any axis can be a facet, and the cross-products generate the "preservation type" labels

This composes naturally with the dossier's existing structure: each axis becomes a *sortable column* in the existing browse tab and a *filterable facet* in the geography / chronology / institutions tabs, while the radar chart and 2-axis scatter become elements of the **detail modal** when a record is opened.

## 5. UI integration plan

| Surface | Change | Effort |
|---|---|---|
| `pipeline/25_preservation_metrics.py` (new) | Compute the 6 metrics + 7 tier scores per record, write back to `inscriptions_index_decrees.jsonl` | 2 h |
| `pipeline/40_bundle_html.py` (extend) | Include the new fields in the inlined data JSON; bundle a CSS file with tier-colour swatches | 0.5 h |
| `dossier.html` Browse tab | Add new sortable columns: Hist-tier, Epig-tier, Formula-density, Proso-weight (one column per axis), plus a "sort by composite" mode | 1 h |
| `dossier.html` Overview tab | Add a **Preservation panel** with the 2-axis scatter + tier-distribution histogram | 1 h |
| `dossier.html` detail modal | Add a small radar chart showing the 7-axis profile of the selected record | 1.5 h |
| `dossier.html` selection bar | Adjacent to the existing institution/era badges, show two new badges: `Hist:Tx / Epig:Ty` | 0.5 h |
| `plugins/restoration.html` | Cross-link: hovering a candidate restoration in the workbench shows the host record's `historical_significance_score` and explains "this restoration is high-stakes" or "low-stakes" | 1 h |
| `case_studies/*.html` | Each of the six case studies adds an inline "preservation type" sidebar — case #1 is `(T1 historical, T5 epigraphic)`, case #6 is `(T5 historical, T2 epigraphic)`, etc. — making the typology's argument concrete on the readers' most visible documents | 1.5 h |
| `INSTRUCTOR.md` (Week 13) | Append a §4.5 "Preservation as significance" rubric — a class-discussion prompt asking students to defend a case's *historical* vs *epigraphic* significance using the dossier's tier scores | 0.5 h |
| `html_dossier_plan.md` | Append §15 "Preservation-typology dimension" describing the bipolar axis + the five supplementary axes | 0.5 h |
| `formulae_catalogue_HRO.md` | Cross-tag each catalogued inscription with its preservation type, so the catalogue can be re-sorted by significance class | 0.5 h |

**Total implementation effort: ~10.5 hours**, deployable across two sessions. Phase 1 (pipeline + Browse columns + tier badges) is the smallest viable cut and ships in ~4 h.

## 6. Diagnostic specimens — calibrating the tier thresholds

The thresholds for T1–T5 binning need empirical calibration. The dossier's six shipped case studies serve as the calibration set:

| Case | Inscription | Expected historical tier | Expected epigraphic tier | Why |
|---|---|---|---|---|
| #1 | SC de Bacchanalibus | T2 | T5 | Bronze tablet, near-complete; routine senatorial business; well-known and edited many times but supplements minimal |
| #2 | SC de Cn. Pisone Patre | T4 | T4 | Six copies survive; restorations needed to reconcile readings; historically explosive (Germanicus's death) |
| #3 | SC de Beguensis | T2 | T5 | Bronze tablet, almost zero loss; routine market-permission |
| #4 | SC de Plarasensibus | T5 | T2 | Latin original lost entirely; only the Greek translation survives; every Latin reconstruction is hypothesis |
| #5 | SC de Asclepiade | T3 | T3 | Bilingual stone; both faces partly damaged; both restored substantially |
| #6 | IG II² 1534 + 1535 Asklepieion | T5 | T2 | Heavily lacunose Greek; massive editorial reconstruction; each decree's slot skeleton is restored more than read |

This calibration set produces clear separations across the 2-axis grid: cases #1 and #3 cluster in the (T2-historical, T5-epigraphic) cell; case #4 sits at the extreme (T5-historical, T2-epigraphic); case #6 at (T5, T2). The diagonal hypothesis — that historical and epigraphic significance trade off — is testable directly against the calibration set, and the dossier ships with a default tier-bin tuning that places the six cases at their expected coordinates.

## 7. Phasing

**Phase 1 — Core bipolar axis (4 h, this session if budget allows; otherwise next session)**
1. Write `pipeline/25_preservation_metrics.py` computing the 6 raw metrics + 2 derived scores (historical, epigraphic).
2. Add the fields to the inlined data JSON via `40_bundle_html.py`.
3. Add two sortable columns to the Browse tab + two new badges to the selection bar.
4. Calibrate tier thresholds against the six case studies.

**Phase 2 — Five supplementary axes + Overview panel (3 h)**
5. Extend `25_preservation_metrics.py` to compute the 5 supplementary axes.
6. Add the Preservation panel to the Overview tab with the 2-axis scatter + tier histogram.
7. Add the 7-spoke radar chart to the detail modal.
8. Cross-link with the Restoration workbench plugin.

**Phase 3 — Case-study integration + catalogue cross-tag + docs (3.5 h)**
9. Add the preservation-type sidebar to each of the six case studies.
10. Cross-tag every inscription in `formulae_catalogue_HRO.md` with its preservation type.
11. Update `INSTRUCTOR.md` and `html_dossier_plan.md`.

Cumulatively this becomes **§15 of the dossier plan** — a new top-level dimension alongside the existing Chronology, Geography, Institutional, and Cross-influence dimensions. The preservation typology is the dossier's first explicitly **two-pole evaluative** dimension (the others are descriptive); it is where the dossier moves from cataloguing to argumentation about *why* particular inscriptions matter.

---

## §M15 revision — HT, ET, RT as three independent dimensions

> **The user's correction** (turn-15): "epigraphically preserved does not necessarily mean historically dull; assess need to be assessed in a third dimension, the number of significant terms (persons, institutions, dates, important concepts and references of socio-political, cultural, economic and other)."

The original model treated historical significance and epigraphic significance as two ends of a bipolar axis with restoration density as the diagnostic for historical. The user identified the flaw: **restoration density measures scholarly inferential investment, not historical significance**. An inscription can be epigraphically perfect AND historically rich — the SC de Bacchanalibus is precisely this case (bronze tablet, zero lacunae, 26 named senators + magistrates + 14 institutional terms + a sanction-concept vocabulary cluster).

### The HT / ET / RT triad

| Tier | Stands for | Measure | What high score means |
|---|---|---|---|
| **HT** | Historical Tier | significant-terms density per 1000 chars | text is content-rich: named persons + institutions + dates + concepts |
| **ET** | Epigraphic Tier | complete-run lengths × (1 − damage) | text is preserved: long unbroken runs, low lacuna density, low dotted-letter density |
| **RT** | Restoration Tier | restoration density × apparatus density × length | text has attracted heavy editorial inferential investment: many `[…]` brackets, much hypothesis built in |

The three are **independent**. Any product of the three tiers is admissible:

- `HT5/ET5/RT1` — rare ideal. Content-rich, pristine, minimal restoration. SC de Bacchanalibus + SC de Beguensis.
- `HT4/ET3/RT4` — content-dense, partly intact, substantially restored. SCpP (the Tiberian senatorial decree, six-copy reconciliation).
- `HT2/ET1/RT4` — content-thin, fragmentary, heavily restored. IG II² 1534+1535 Asklepieion. The formal-skeleton class.
- `HT5/ET2/RT5` — controversy magnet: content-dense, fragmentary, maximally restored. SC Cyrene-style.
- `HT2/ET5/RT1` — paleographic gold mine but historically limited: complete + minimal restoration but content-thin. Standard private dedications, brief lapidary records.

### Significant-terms metric

`significant_terms_density` (per 1000 chars) combines four categories:

| Category | Greek markers | Latin markers |
|---|---|---|
| **Persons** | capitalised tokens (Α-Ω), ethnonyms, demotics | capitalised tokens (A-Z), gens names, patronymics |
| **Institutional** | βουλή, δῆμος, πρυτάνεις, ἐκκλησία, ἄρχων, γραμματεύς, πρόεδροι, στρατηγός, ἱερεύς, πρόξενος | senatus, consul, praetor, quaestor, tribunus, curia, ordo, decurion, augur, pontifex, civitas, sodalitas |
| **Dates** | ἐπὶ X ἄρχοντος, ἐπὶ τῆς N πρυτανείας, Attic month names | consulibus, conss., Kalendis, Idibus, Nonis, Latin month names |
| **Concepts** | honorific (χρυσῷ στεφάνῳ, εὐεργεσίας), dedication (ἀνάθημα), privilege (πολιτείαν, ἀτέλειαν, ἔγκτησιν), money (τάλαντον, δραχμή), diplomatic (συμμαχίαν), religious (ἱερά), decree types (νόμος, ψήφισμα, δόγμα) | sanction (censuerunt, placere), SC machinery (scribendo adfuerunt, in senatu fuerunt), market (nundinae), sacred (sacrosanctus, aerarium), polity (populus Romanus, res publica), money (sestertios), legal (testamentum, heres) |

A record's HT score is `0.75 × min(1, density / 25.0) + 0.25 × normalise_log(length)`. The 25-per-1000-chars saturation point is calibrated to the densest SCs (SC de Beguensis at 73 per 1000, SC de Bacchanalibus at 20 per 1000).

### Calibration set under the new model

| Case | Type | HT | ET | RT | Sig terms | Notes |
|---|---|---|---|---|---|---|
| 1 — SC Bacchanalibus | **HT5/ET5/RT1** | 0.83 | 1.00 | 0.14 | 44 (19.9/1000) | rare ideal — anchor for the top-right grid corner |
| 2 — SCpP | **HT4/ET3/RT4** | 0.68 | 0.42 | 0.65 | 216 (14.5/1000) | densely prosopographic + reconciliation-restored |
| 3 — SC Beguensis | **HT5/ET5/RT1** | 1.00 | 1.00 | 0.15 | 261 (73.2/1000) | densest content-per-character — anchor for HT5 ceiling |
| 4 — SC Plarasensibus | HT3/ET2/RT5 (calibrated) | — | — | — | — | Greek-only survival, calibrated |
| 5 — SC Asclepiade | HT3/ET3/RT3 (calibrated) | — | — | — | — | symmetric bilingual, calibrated |
| 6 — Asklepieion | **HT2/ET1/RT4** | 0.33 | 0.16 | 0.70 | 66 (2.6/1000) | content-thin + fragmentary + heavily restored — formal-skeleton class |

The diagonal of the original HT × ET grid is gone: under the new model **cases 1 + 3 occupy the top-right corner together with the diagonal** rather than the anti-diagonal predicted by the old restoration-density-as-HT model. RT is the dimension that captures the restoration-versus-completeness opposition the original model was trying to express.

### UI changes (M15.2 / M15.3)

- Distribution table on the Preservation tab now has three columns: HT, ET, RT (each binned into 5 tiers).
- 2-axis scatter is HT (Y) × ET (X). Tier-corner labels are `HTn/ETm`.
- Type label format: `HT4/ET3/RT4` (was `T4/T3`).
- Detail modal: preservation badge shows the new 3-tier type + HT, ET, RT scores + the significant-terms breakdown (persons / institutional / dates / concepts).
- Radar chart now has 9 axes (HT, ET, RT + the 5 supplementary + diachronic centrality).


---

<a id="evidentiary-profile"></a>
## §16 — Evidentiary Profile (the umbrella term, M19.1)

After the M15 redesign decoupled the bipolar "preservation typology" into three independent tiers — HT (Historical), ET (Epigraphic), RT (Restoration) — the dossier needed a single shared name for the family of three. The user asked for one (turn 2026-05-27, "consider a separate term to capture what the three aspects are measuring").

**Adopted term: Evidentiary Profile.**

Each tier names a *kind of evidence* the inscription offers:

| Tier | Kind of evidence | What high score means |
|---|---|---|
| **HT** | **Content as evidence** | the text is dense with named persons, institutions, dates, concepts |
| **ET** | **Survival as evidence** | the text is materially preserved — long unbroken runs of intact letters |
| **RT** | **Inferential mediation as evidence** | the text has attracted heavy editorial restoration |

The *Evidentiary Profile* of a record is its tuple `(HT, ET, RT)`. The dossier's case-study calibration set lays out the corners: HT5/ET5/RT1 (SC Bacchanalibus — content-rich, pristine, minimal inference), HT4/ET3/RT4 (SCpP — content-dense, partial, reconstructed), HT2/ET1/RT4 (Asklepieion — content-thin, fragmentary, heavily restored).

Why this name:

- **Accurate**: each tier is in fact a kind of evidence-quality measurement. The shared noun is *evidence*.
- **Avoids overloading "significance"**: the M15 redesign was precisely about separating "significance" into three. Keeping that word inside individual axis names (`historical_significance_score`) and using *Evidentiary Profile* as the umbrella keeps the vocabulary disciplined.
- **Generalisable**: if a 4th axis is promoted later (e.g. diachronic centrality), the profile becomes a 4-tuple; the name still works.
- **Reads cleanly**: "this case's Evidentiary Profile is HT5/ET5/RT1" is parseable as a single phrase.

---

## §17 — How to read an Evidentiary Profile (M19.3)

This is the dossier reader's guide to making operational sense of HT/ET/RT scores.

### §17.1 — What each score range means operationally

| HT score | Operational reading |
|---|---|
| 0.0 – 0.20 (T1) | fewer than 1 significant term per 100 chars; the text is largely connective tissue (verbs, articles, prepositions) without named persons or institutions |
| 0.20 – 0.40 (T2) | sparse named entities; useful only as a genre-shape witness, not a prosopographical source |
| 0.40 – 0.60 (T3) | 3–5 significant terms per 100 chars; reasonable historical-source value with caveats |
| 0.60 – 0.80 (T4) | 5–10 significant terms per 100 chars; rich enough to use as a prosopographical or institutional source |
| 0.80 – 1.00 (T5) | more than 10 significant terms per 100 chars; the text IS a named-personnel + dated-event list as much as a narrative |

| ET score | Operational reading |
|---|---|
| 0.0 – 0.20 (T1) | mostly lacunae; what remains is short scattered runs; primary-source value is limited |
| 0.20 – 0.40 (T2) | fragmentary; long restored stretches |
| 0.40 – 0.60 (T3) | partly intact; one or more long preserved runs but significant gaps |
| 0.60 – 0.80 (T4) | largely intact; most of the text is preserved with minor lacunae |
| 0.80 – 1.00 (T5) | pristine; the inscription is a primary witness for paleography, dialect, and unbroken syntax |

| RT score | Operational reading |
|---|---|
| 0.0 – 0.20 (T1) | minimal editorial restoration; what you read is essentially what survives |
| 0.20 – 0.40 (T2) | lightly supplemented; a few short restorations from clear context |
| 0.40 – 0.60 (T3) | moderately reconstructed; some passages depend on conjecture |
| 0.60 – 0.80 (T4) | substantially reconstructed; significant portions are inference |
| 0.80 – 1.00 (T5) | maximally restored; the surviving text is more inference than reading — treat readings critically |

### §17.2 — What cluster positions on each 2D scatter signal

The three pairwise scatters at the bottom of the Evidentiary Profile tab each answer a different question:

**HT × ET — "Where do content-rich records cluster?"**

- *Top-right* (HT5/ET5) = pristine content-rich records. The rarest and most valuable analytical class. SC Bacchanalibus, SC Beguensis live here.
- *Top-left* (HT5/ET1) = content-rich fragments where the named-person + institutional density survives in scattered runs. High historical value via what's reconstructed, but expect controversy.
- *Bottom-right* (HT1/ET5) = pristine but content-thin records. Routine dedications, brief lapidary records, short receipts. Paleographic gold mine, historically limited.
- *Bottom-left* (HT1/ET1) = damaged, low-content. Often short fragments or graffiti; main value is filling out distributional statistics.

**ET × RT — "Does well-preserved ↔ less-restored hold?"**

- *Top-right* (ET5/RT5) = the unusual class: well-preserved AND heavily restored. Typically means an editor restored letter-traces or partly-erased text rather than full lacunae. The restoration is grounded but the surface text is mixed.
- *Top-left* (ET5/RT1) = the routine well-preserved class: most of the dossier's HT-bottom records.
- *Bottom-right* (ET1/RT5) = the expected class for fragments: heavily damaged AND heavily restored. The slot the dossier's Asklepieion decrees (case #6) occupy.
- *Bottom-left* (ET1/RT1) = damaged AND minimally restored — fragments deemed not worth reconstructing.

**RT × HT — "Are content-rich records restored more?"**

- *Top-right* (HT5/RT5) = the dossier's most argued-over class. Content is rich enough that every restored letter is a piece of a known argument. Treat as a *secondary* witness to scholarly inference and a *primary* witness to the formula bundle the editor restored from.
- *Top-left* (HT1/RT5) = "the editor argued more than the inscription deserved." Heavy restoration without much content density — usually because the inscription is short and editors stretched.
- *Bottom-right* (HT5/RT1) = the M15-corrected "rare ideal" cell: content-rich AND minimal restoration. SC Bacchanalibus, SC Beguensis.
- *Bottom-left* (HT1/RT1) = thin AND unrestored — what most short fragments look like.

### §17.3 — Triangulating a record's full 3-tuple (worked walkthroughs)

**Case 1 — SC de Bacchanalibus = HT5/ET5/RT1.**

Read: "rare ideal — use this as a primary source for everything." The 26 named persons + 14 institutional markers + 2 dates + 2 sanction-concept terms (Sherk-detected) are entirely on the original bronze; no restoration needed; full surface integrity. Workflow: cite directly for prosopography (the consuls, the senatus), for the Republican SC genre (the four-part scheme is fully visible), for the dialect of mid-Republican chancery Latin. Nothing in this record is editorial inference.

**Case 2 — SCpP = HT4/ET3/RT4.**

Read: "content-dense but reconstructed — use as primary for prosopography, secondary for the reconstruction methodology." 170 senatorial names + 38 institutional markers survive across six provincial copies, but no single copy is intact; the modern text is a six-copy reconciliation. Workflow: cite for prosopography (the named senators are secure across copies); use the reconstruction itself as a study object (Eck 1996 + Cooley 2023 walk through the editorial method); treat individual restored phrases with the apparatus visible.

**Case 6 — IG II² Asklepieion decrees = HT2/ET1/RT4.**

Read: "content-thin and heavily restored — use to study the *formula bundle*, not the inscription's content." Only 66 surviving significant terms (and 0 named persons in the IPHI extraction — the proposer, secretary, ten-man commission names are all bracketed). Workflow: the case study's whole argument is that the formula bundle's slots survive even when the content doesn't; use this case to study how editors restore Athenian decree prescripts; do not cite it as a prosopographical source. The decrees are evidence for *editorial-method* and *formula-stability*, not for the content of the decrees themselves.

### §17.4 — What analytic questions each axis is good for answering

| Axis | Use for | Don't use for |
|---|---|---|
| **HT** | prosopography, institutional history, dated-event reconstruction, conceptual-vocabulary studies, social-network analysis | judging the inscription's political *importance* (HT measures density, not weight) |
| **ET** | paleography, dialect studies, layout / typography / stoichedon analysis, syntactic studies, unbroken-token frequency analysis | judging the inscription's historical interest (a perfectly preserved routine receipt is HT-low / ET-high) |
| **RT** | studies of editorial method, historiography of decipherment, source-criticism of secondary scholarship, identifying which inscriptions have been worked over by how many scholars | adjudicating restoration *quality* (RT measures density only — a well-reasoned restoration and a wild one both contribute the same score) |

### §17.5 — What the Evidentiary Profile CAN'T tell you

Three explicit limits, so the typology isn't pushed past its design:

1. **It does not measure historical importance.** A routine bronze tablet may score HT5/ET5/RT1 (calibration anchor) while being historically minor (e.g. a single private-property dedication). Conversely a fragmentary tax decree may be historically momentous while scoring HT2/ET1/RT5. *Importance* is a separate axis the dossier does not currently quantify.
2. **It does not adjudicate scholarship quality.** Heavy restoration with letter-trace evidence and heavy restoration with mere conjecture both contribute the same RT. To assess restoration *quality*, read the apparatus.
3. **It does not measure the formula bundle's coverage of any given record directly.** That is the supplementary "genre stability" axis (M13.1, §3.1 above). HT and genre-stability are correlated but distinct: a record can be HT-high (content-rich) and genre-low (the rich content is non-formulaic narrative).

### §17.6 — Recommended workflow for using the Evidentiary Profile to find a source

1. **Start with the HT × ET scatter** in the Evidentiary Profile tab. The top-right cells (HT4-5 / ET4-5) are the dossier's best-witnessed records — your default starting set.
2. **Filter by RT in the 3D cube.** Low RT (RT1-2) = primary witness with little editorial inference; high RT (RT4-5) = inferential. Pick based on your question: prosopography wants low RT; editorial-method study wants high RT.
3. **Open the detail modal** to confirm the full 9-axis radar. The five supplementary axes — genre stability, prosopographic weight, paleographic peculiarity, monumental significance, citation density — and diachronic centrality refine the picture beyond the core triad.
4. **Cross-check against the case-study calibration set.** Which of the 11 calibration anchors does your query record sit closest to? Read the corresponding case-study sidebar to see how that profile reads in practice.