Strategy Focus

Why

Every source in reach-edu belongs to the client. When someone sits down to draft a strategy document or a presentation, they want all of it available — and a pointer that says mainly look here.

That is what the domains: frontmatter field is:

domains:
  - "strategy:adult-literacy-numeracy"

Measured across all 845 sources on 2026-08-22:

Read that table as a pointer, not as a half-finished taxonomy. A source with no domains: is not unclassified; it is simply not the first place to look for any particular strategy. Backfilling the 604 would destroy the signal by making everything equally emphasised — which is why this spec does not.

The zero was read wrong (corrected 2026-08-23)

multi-valued: 0 was taken to mean multi-domain sources do not exist here, and two decisions were built on that reading. It means the opposite of what it was read to mean.

Multi-domain membership in this corpus is expressed as one pointer file per folder, not as a longer list in one file. Measured the same day:

Each pointer is ~1.1KB: the canonical source_uuid, status: metadata-only, the lede, and its own empty # Extracts skeleton — carrying exactly the one domains: value that names the folder it sits in. The body is not there. The canonical content lives in SurrealDB per [[../../../augment-it/context-v/specs/Source-Content-Storage-SurrealDB-Primary-Local-As-Toggle]], which makes corpus_path optional and puts Extracts per usage — and a filing into a second strategy is a second usage.

That is the design, not drift. A corpus that will get very large has to hand an agent a scoped set of sources when it drafts, because accuracy and attribution are the whole product; a source that genuinely bears on two strategies belongs in both scopes, with its own extracts in each. The pointer is what makes that cheap.

So domains: is single-valued by construction. The number to reason from is 79, not 0.

The one rule everything follows from

Focus narrows, and compounds with search.

apprenticeship → 56. apprenticeship + Workforce Development → 12. apprenticeship + Rural Income Boosts → 4.

This is the second version of this rule. The first read “mainly look here” as emphasis: focus returned everything, ordered, with the tagged sources first. That is a defensible reading of the words and it was wrong in practice. Driven in a real browser it was indistinguishable from nothing happening — 200 rows on screen, 82 matches, so only the top of the list moved and rows 83–200 were unrelated. The operator’s report was “I searched a strategy, a tag cloud came up, I clicked a tag, I didn’t see it further filter.”

What preserves access to the rest of the corpus is not that the list stays whole. It is that the toggle is a toggle — one click back — and that corpus_total rides alongside total, so a narrowed list always says what it is a subset of: “12 in Workforce Development · 832 in the corpus”. A single number is what makes a filter look like the whole world.

Behaviours

1. The tag is searchable text

Typing literacy reaches a source carrying strategy:adult-literacy-numeracy even when neither its title nor its excerpt contains the word. The tag is part of what a source says about itself, so search treats it that way.

2. Focus narrows, and reports both numbers

focus=strategy:workforce-development returns only that strategy’s sources, with corpus_total beside total.

3. The type vocabulary is read from the corpus, never assumed

This is the load-bearing decision. The types are open: reach-edu declares strategy and topic; another client uses thesis. There is no rule that maps a tag to a folder across that. strategy/strategies tempts a +s — thesis/theses breaks it on the first try, and whatever replaces the rule breaks on the client after that.

Each domain already states the answer, in the index.md sitting in its folder:

type: "strategy"
slug: "adult-literacy-numeracy"
title: "Adult Literacy & Numeracy"

So the join is read: nine reads in reach-edu, one per declaration, not one per source. That also supplies the human label for the toggle — Adult Literacy & Numeracy, not adult-literacy-numeracy — and makes nesting free, because the folder is wherever the index.md is, at any depth.

A folder with no index.md is not a focus. Nothing is invented for it.

4. Narrowing happens on the row, once a manifest exists

Revised 2026-08-23. This section used to describe a gap and name its fix. The fix shipped: [[Search-Index]].

A plain page load used to narrow on the key, because reading 845 files to check a tag was the only alternative. That left exactly one case wrong — a source tagged into a focus whose folder it does not live under was found by a search and missed by a page load. FOCUS-07 asserted both halves precisely so that the day it changed, a test would say so.

With a manifest, every source’s domains: is legible for the cost of one object. Narrowing now consults the row on every path, the gap is closed, and FOCUS-07 is inverted rather than deleted — the ID stays, the promise moves.

A corpus with no manifest still narrows on the key for a page load and on the row under search. That is the older behaviour, kept because a corpus that has never been indexed has to keep working.

5. Chips are grouped by their declared type

A row of unlabelled chips is legible only to someone who already knows the corpus — the operator’s own reaction was “oh, the tags are strategies.” Each group carries its declared type as a heading (STRATEGY, TOPIC, and whatever a client’s corpus declares), read from the data like everything else here.

6. Multiple focuses are the eventual shape, one is the current one

One focus at a time is what ships, and the API takes a single focus.

The reason given here was wrong — “zero multi-valued data” — and is corrected above: 79 sources are filed across two or more folders. The honest reason is narrower. A focus selects a folder, and a source in six folders is reached by any of the six; what a second simultaneous focus would mean — union or intersection — has never been asked for by anyone drafting against this corpus. That is a question to put to an operator, not one to answer by guessing.

7. A domain reference is not a tag

This spec called them tags until 2026-08-23, and they are not. A source carries both fields and they answer different questions:

Measured across reach-edu, 2026-08-23:

Workforce-Development exists as a tag and strategy:workforce-development as a domain reference. Same idea, two fields — and only one of them is a scope an agent can be pointed at. Calling a domain reference a tag is how a scope quietly becomes a keyword.

8. A domain reference cascades; a tag does not

Per [[../../../context-v/specs/Flexible-Entity-Relationships-to-Mirror-Messy-IRL-Collaboration]] — the cascade is invoked, not stored. A reference names a chain, and naming a shorter chain reaches everything beneath it:

focus                              matches
strategy                        →  strategy:workforce-development
domain:strategy                 →  domain:strategy:workforce-development
strategy:workforce-development  →  itself only

The path already behaves this way — _in_domain is d == domain or d.startswith(domain + "/"), which is why funders/ is a legal filter and the combobox’s Backspace walks up it. This gives the reference the same cascade the path has.

Segment-aware, or it is wrong. strategy must not match strategy-two:something. The test is on the separator, not the string — exactly the trap _in_domain already avoids with /.

A tags: value never cascades. It has no separator to cascade on, and it is a label rather than a scope.

9. A declaration is not a source

index.md, AGENTS.md and README.md describe the corpus rather than being captured material, and are excluded from the listing. Thirteen such files in reach-edu were being rendered with no URL and status: candidate — indistinguishable from a source we found and never fetched — and inflating the count. 845 sources was 832 sources plus 13 documents about them.

This stopped being cosmetic the moment index.md became the file that supplies the type vocabulary. The document that names a domain cannot also be listed as one of its unprocessed leftovers.

Tests

Related

• context-v/specs/Browse-Corpus.md — the listing this extends, and BROWSE-17, the stale-response guard search needed once a search cost 5.8s
• context-v/specs/Domain-Navigation.md — the exclusive verb, deliberately kept separate
• context-v/specs/Corpus-Tree.md — the same keys-not-bodies discipline
• [[Search-Index]] — the manifest that closed the gap §4 used to name