How this works
A visual map of how historians date events and documents — and how strong each justification is. Not a verdict machine. Not “one true timeline.”
History dates are rarely a single stamp
Imagine you want to know when a book was written. Nobody wrote the year on the cover. Instead people say things like:
- “It mentions a war, so it was written after the war.”
- “Someone in the year 180 quoted it, so it was written before that.”
- “Another book copied from it, so that other book is later.”
Each of those is a link with a reason. Some reasons are rock-solid (coins, official records). Some are shakier (how we read a poem or a prophecy).
This tool draws those links as a graph: circles = events or books, lines = “dated because…”. Colors show strength. You can raise a confidence floor and watch what still holds — and what falls apart.
House metaphor: Nodes are rooms. Edges are beams. A thick green beam is strong timber. A thin red dashed beam is duct tape. Raising the floor means: “only count beams this strong or stronger.” Rooms that were only held up by duct tape become isolated — that’s the point.
What you’re looking at
Nodes (circles)
Events, documents, people, physical finds, or absolute markers (like “Roman year-list”).
Edges (lines)
A dating claim: after / before / during / depends on / cites / same calendar scale…
Focus layer + rails
Expand Documents (or another kind). Other swimlanes collapse to connection rails with stubs — you still see that hinges exist without full foreign nodes.
Zoom / pan
Scroll to zoom, drag the background to pan. Reset returns to 100%.
Warrant Desk
Named triage queue for continual improvement: clarify a hinge, demand citations, re-score R, request a new seed/case, or request a bridge. Export JSON when you want it outside the browser.
Two axes (this is the heart of the method)
We never smash “famous” and “solid” into one score.
Reliability (R) — 1 to 5
How good is this reason? Method, independence, circularity, sources. High R = trustworthy warrant. Low R = fragile or speculative.
Dependency (D)
How much hangs on it? If cutting this link wrecks many dates, D is high — even if R is high (a load-bearing beam) or low (a juicy weak link).
Evidence is unequal on purpose
A carbon date is not the same kind of thing as “this paragraph sounds like it knows the Temple fell.” Every edge must declare an evidence class. Classes cap how high R is allowed to go unless stronger corroboration is cited.
| Class | Plain English | Typical R |
|---|---|---|
| A Science | Radiocarbon, tree rings, well-fixed eclipses | up to 5 |
| B Material | Destruction layers, coins in context | 4–5 |
| C Admin docs | Tax receipts, consular years, dated papyri | ~4 |
| D Historiography | Tacitus, Josephus (with criticism) | 3–4 |
| E Literary relative | Text B used text A (e.g. Markan priority) | 2–4 |
| F Literary→history | “Written after the event” from wording alone | 1–3 |
| G Paleography | Handwriting date ranges for manuscripts | 2–3 |
| H Reception | Later authors mention a work | 2–4 |
| K Assumption | Shared calendar spine made explicit | 3–5 |
Example: Temple fall under Roman chronology can be fairly robust as civil history. “Mark is after 70 because of Mark 13” is class F — a literary inference — and is allowed to stay fragile even if many books repeat it.
Confidence floor — “only show me solid stuff”
The dropdown asks: keep edges at or above this reliability. Everything weaker is ghosted or hidden. Then we report:
- What constraints remain (lower bounds, upper bounds, orderings)
- What fell apart (excluded edges, isolated nodes, split groups)
- Whether a focus event still has a trusted floor/ceiling
That’s how you feel the difference between “the whole textbook story” and “what survives if we only trust working-or-better warrants.”
Cases (stories) — optional tours
A case (authoring term: seed) is a guided story for one dating question. Event and document records are autonomous — stories can be rewritten without deleting the underlying nodes. Prefer Find a record search when you know the thing (e.g. Great Pyramid) and don’t care which story pack listed it first.
- Mark × Temple — how Mark’s date hangs on 70 CE and later witnesses
- Nero — reign, fire 64, death 68, and the path toward Temple 70
- Magna Carta 1215 — regnal spine, dating clause, engrossments, later reissues (separate nodes)
- Giza / Khufu — Great Pyramid via king-list, archaeology/graffiti, 14C, vs late reception (Herodotus) and fringe sky claims
Why cases? Pedagogy and curation scope — not ownership of truth.
Shared anchors (Roman spine, Temple 70, Egyptian king-list) use one node ID when the referent is the same.
See also project notes: nodes-and-stories.md, anchor-backlog.md.
Comparing dates across two cases
You can’t always subtract two “CE” labels and call it science.
-
Same node ID?
If both cases include
temple_destruction_70, that’s a claim they mean the same event — a join key. - Shared clock? Absolute comparison needs a bridge onto the same chronology spine (calendar / CE scale), stated as edges with their own R.
- Union + floor Merge graphs on shared IDs, keep only edges above the floor, propagate before/after constraints.
- Report windows & worlds Comparable under spine? Only via soft reception path? Or disconnected for dating purposes?
Deep design lives in the project note bridge-probability-method.md
(hierarchical times + typed bridges + “possible worlds”). The Explore UI will grow a
“combine cases” mode; methodology first so the feature can’t lie.
How a chain gets into the graph
- Question — e.g. “What fixes Mark’s earliest possible date?”
- Draft edges — type, class, R, one-sentence warrant, sources
- Status —
ai_proposeduntil a human markscurated - Score dependency — what breaks if this edge moves?
- Explore — floor slider, weak-link list, cut scenarios
Full detail (for the curious / future curators)
Edge types (constraint language)
| Type | Meaning for time |
|---|---|
tpq | Terminus post quem — target no earlier than source (+ optional δ) |
taq | Terminus ante quem — target no later than bound |
before / after | Strict order |
during | Occurs inside another span |
depends_on | Literary/method dependence (often relative order) |
cites | Textual witness relationship |
synchronism | Tied to the same absolute scale / concurrent markers |
same_as | Identity (use sparingly; must be argued) |
Display vs attack scores
- Display strength ∝ R × class weight (green / amber / red in the lab)
- Weak-link / adjust score ∝ (1 − R) × f(D) — and later, sensitivity of date windows
Population statuses
seed— starter illustrative pack (lab demos; citations may still be placeholders)ai_proposed— machine draft; not mainstream until acceptedcurated— human-accepted for analysisrejected— kept for history, out of live graph
Required fields on a serious edge
from, to, type, evidence_class,
reliability, warrant, sources[] (for curated),
independence, scholarly_status, population_status.
Independence
independent · partially_dependent · circular_risk —
circular pairs should slash R and surface as a cycle risk in weak-link tools.
Revisionist / adversarial mode (optional, secondary)
Not the product identity. Any named alternate chronology can be a labeled overlay pack
to ask: “which load-bearing links would this have to answer?” Fomenko is one example among many.
See project notes revisionist-lens.md.
Cross-case bridges (research track)
Recommended stack: temporal constraint network + hierarchical model over latent years +
typed bridge mixture (calendar vs reception vs spurious) + average over possible worlds.
Never use embedding similarity as the truth layer. Implementation target:
seed-ce-spine-bridge + bridge posterior script (Phase A).
What this is not
- A court that “proves” or “disproves” a religious or revisionist school
- A replacement for Pleiades, Trismegistos, radiocarbon labs, or critical editions
- Auto-polemic content farm
- Permission to treat literary inference as equal to material anchors
It is a microscope for justification structure — so strong and weak reasons stop looking the same color in your head.