How Timeline works — routed extraction, deadlines, and calendar export

Architecture

One extraction flow. Three transparent stages.

Timeline first gathers deterministic date hints, sends the document through one routed structured extraction flow, then PHP filters, scores, computes deadlines, and prepares exports.

Step 1 · PHP / regex

Detect date hints before the model

Before any model call, PHP scans the full input for dates matching 12+ Norwegian formats and normalises them to ISO 8601 where possible:

dd.mm.yyyy → YYYY-MM-DD
d. månedsnavn yyyy → resolved calendar date
Diary-format lines (starting with a date + colon) → auto-tagged as events
Two-digit years → always interpreted as 20YY

Normalised anchors are added to the extraction prompt to reduce hallucinated or misread dates. They are hints, not a separate AI result.

Step 2 · routed LLM extraction

Route, extract & classify

The selected route reads the document alongside the date hints. For every temporal reference it returns a structured JSON event object with deadline fields:

date — resolved ISO date, or verbatim string if unresolvable
date_type — absolute | relative | recurring | conditional | period
confidence — high | medium | low
actor — attributed entity (from source text, not inferred)
description — one-sentence event summary
source_excerpt — verbatim text fragment (max 200 chars)
is_deadline / deadline_kind — explicit legal deadline markers when present

The prompt explicitly instructs the model not to invent dates or actors not present in the source. Longer inputs are chunked server-side and recombined into the same schema.

Step 3 · PHP post-processor

Filter, compute deadlines & assemble

PHP applies all active filters and derived deadline logic before returning the result:

Focus filter — strips events not matching the requested focus mode (deadlines / hearings / CPS)
Confidence filter — removes LOW-confidence events if requested
Background filter — strips background/narrative events if unchecked
Date-type filter — strips relative/recurring events if unchecked

The post-processor then builds deadlines[], the what_remains_uncertain list, and the next_practical_step recommendation.

Date recognition

12+ Norwegian date formats, all recognised.

Norwegian legal documents use a wide variety of date notations. Timeline recognises these deterministically before the model sees the text, then asks the model to resolve the remaining contextual references.

Format	Example	Notes
`dd.mm.yyyy`	30.07.2015	Standard Norwegian numeric
`dd.mm.yy`	09.04.25	Two-digit year → always 20YY
`d. månedsnavn yyyy`	3. mars 2024	Written month in bokmål/nynorsk
`d. månedsnavn`	15. januar	Year inferred by proximity scanning
`yyyy-mm-dd`	2024-03-12	ISO 8601
`månedsnavn yyyy`	mars 2024	Month + year only
`yyyy`	2024	Year-only reference
Season + year	høsten 2023	Seasonal reference → Q3/Q4
Diary-format line	18.09.2025: Møte avholdt	Date + colon → auto-tagged as event
Relative reference	tre uker etter vedtaket	Anchored to nearest resolved event
Recurring pattern	hver mandag	Classified as recurring
Period / range	fra mars til juni 2024	Yields start_date + end_date

Classification schema

Five event types. Three confidence levels.

date_type values

date_type	Definition	Example
`absolute`	A specific, resolvable calendar date	30.07.2015 → 2015-07-30
`relative`	A date expressed relative to another event	tre uker etter vedtaket
`recurring`	A pattern that repeats on a schedule	each Monday, every 6 months
`conditional`	A date contingent on a condition being met	if no response within 14 days
`period`	A date range or duration with start and end	fra mars til juni 2024

confidence levels

confidence	Meaning	Visual in timeline
`high`	Date is explicitly and unambiguously stated in the source text	Green badge
`medium`	Date is inferred, approximate, or stated with slight ambiguity	Amber badge
`low`	Date is implied, undated, or extracted from a degraded/ambiguous passage	Grey badge

Actor attribution rules

Rule	Example
Named entity in the same sentence	“Trude [saksbehandler] ringte 14. mars” → actor: Trude
Role label without a name	“Barnevernet fattet vedtak” → actor: Barnevernet
No clear attribution in sentence	actor: `[unattributed]`
Document-level default	If no per-event actor, defaults to the document sender/issuing body

Engines

Three routes, one structured output.

Quick, Standard, and Deep return the same JSON schema, so the post-processor handles them identically. Route choice affects speed, quality, credit cost, and fallback reporting only.

Engine	Model	Latency	Best for
Quick	`nova-lite` (self-hosted GPU via LiteLLM)	~10-25 s	Fast drafts and shorter documents where speed matters most.
Standard ★	`Claude Haiku 4.5` (Amazon Bedrock EU; `gpt-4o-mini` only if Bedrock is disabled)	~20-45 s	Default route for most legal documents; balanced speed, quality, and cost.
Deep	`Claude Sonnet 4.6` (Amazon Bedrock EU)	~45-90 s	Dense or complex cases with many actors, overlapping events, or weak source quality.

Live updates & export

See progress as it happens. Export to Word or calendar.

SSE + DOCX + ICS

SSE streaming + DOCX/ICS export

Timeline uses Server-Sent Events (SSE) to stream live status messages to the browser as extraction runs, including preparation, extraction, parsing, and final assembly.

Once extraction completes, export a formatted .docx with labelled events and source excerpts, or download an .ics calendar file for dated events and deadlines. DOCX files are assembled directly from OOXML via PHP ZipArchive.

Server-Sent Events OOXML / .docx iCalendar / .ics ZipArchive live progress Save to My Docs

Privacy & security

Your documents never leave your session.

Privacy by design

All uploaded files are extracted to text in memory using PHP's in-process file handlers. The raw binary is never written to disk on the server.
Session context (pasted text, uploaded content, extracted timeline events) is scoped to your authenticated session and discarded when the session ends.
Model routing is explicit: Quick uses self-hosted nova-lite through LiteLLM; Standard and Deep use Amazon Bedrock EU when enabled, with the existing routed cloud fallback if Bedrock is disabled.
External model calls happen only for the extraction flow. Uploaded binaries are never stored by the tools app, and extracted text is processed in-session.
Telemetry logged: tool name, requested/effective engine, model/deployment, focus mode, event count, deadline count, credits, and latency. No document text, case references, actor names, or extracted events are logged.

How Timeline knows when things happened.