M7: a first-class imported knowledge library

A campaign can import local .txt and .md files as Canon, Reference or
Inspiration, and the class is load-bearing rather than a label: it decides the
words a passage is framed with in the prompt, the weight it carries when
passages are ranked, and which budget it competes in when the context is tight.

This is a separate subsystem, which is the Phase 0B decision
(IMPORTED-KNOWLEDGE-DESIGN.md §73). Story Cards do not carry classification,
provenance, content identity, chunking, an index or a lifecycle, and they were
not promoted into something that does. Nothing here reads or writes one.

The subsystem, in backend/app/knowledge/:

  classes      the three classes, their weights, and the prompt framing
  chunking     deterministic, heading-aware, 60-800 tokens, no overlap
  fts          SQLite FTS5 with porter stemming; scoped and bounded in SQL
  importer     validate, hash, store, chunk, index — in one transaction
  embeddings   local Ollama vectors through the shared provider
  retrieval    query construction, hybrid merge, rerank
  inject       the budgeted cut and the rendered prompt sections

Relevance admission is a separate stage from ranking, and that separation is
the milestone's most expensive lesson. An independent review found the first
implementation deciding relevance with a floor expressed as a share of the best
candidate — which the best clears by construction — so a passage was admitted on
every turn regardless of the scene. A query about tide tables and container
tonnage retrieved all five sources of a fantasy campaign, narrator-only hidden
Canon among them.

So the pipeline is now:

  candidate generation -> admission -> ranking -> class weighting -> budget

Admission reads raw, candidate-set-independent signals: the cosine the model
returned, and how many distinct meaningful query terms a passage contains.
Ranking reads normalized ones, because bm25 has no fixed range and cosine's zero
is not zero. Normalization decides order among things that matched; it can never
decide whether anything matched. Authority is applied after admission, so a
class orders what matched and never rescues what did not.

Retrieval may therefore return nothing, and on a scene unrelated to the library
it does.

The other decisions that each replaced an obvious wrong one:

- The class multiplies relevance rather than adding to it. An additive bonus
  satisfies "Canon outranks Reference" and makes "do not include irrelevant
  Canon" impossible, because a large enough constant wins on its own.
- The semantic floor is measured, not guessed: 113 production-path pairs against
  nomic-embed-text put targeted matches at 0.55-0.85 and off-topic pairs at
  0.36-0.56, and 0.58 sits between them. Because it is a property of that model
  and not of cosine similarity, it is keyed to the model rather than applied to
  whatever is configured: an embedding model with no measured calibration in
  this build does not borrow the number. Semantic admission is skipped, the
  campaign retrieves lexically, and the reason is stated in the knowledge status
  and in the turn's provenance. Degrading to lexical keeps the library usable;
  lending the threshold to an unmeasured model is how the admitted-everything
  defect would return.
- One lexical term is not evidence. Two distinct meaningful terms, or one that
  is neither a standing campaign entity nor a negligible share of the query.
  The stop list grew from 42 words to 261, all function words — no subject
  matter, because a stop list that removes subject matter stops finding "The
  Silver Key".
- Lexical retrieval is a production path, not a fallback. It finds the proper
  nouns and invented terms a setting bible is made of, and the library is fully
  usable with no embedding model configured.

Safety is structural rather than filtered. Imported text reaches the prompt
whole, inside a section that says what it is, under a rule stating the authority
order in words and refusing every instruction inside it. No endpoint accepts a
filesystem path, so H08 has no mechanism to escape from. Nothing renders
imported content as HTML, so a script tag is five visible characters and a
remote image is never fetched. Import, chunking, indexing, retrieval and a turn
open no socket at all; only embeddings do, through the endpoint allowlist the
memory bank already uses.

Provenance is the rendered text, not a foreign key: deleting a source cannot
turn a historical turn's evidence into dangling ids.

Schema: knowledge_sources, knowledge_chunks, knowledge_embeddings, and an FTS5
virtual table attached to knowledge_chunks as a DDL hook so it is created and
dropped with the table it indexes. Migration 92. A pre-M7 database opens
unchanged and needs no sources to play.

Bundle: the source content and the reader's judgements about it travel; the
passages, index rows and vectors are rebuilt on import, so a restored campaign
is searchable immediately without a reindex step.

One runtime dependency: python-multipart, Starlette's multipart parser. It is
what makes the upload surface possible, and the upload surface is why no
pathname is ever accepted.

The test doubles were the reason the defect shipped, so they were corrected too.
The retrieval stub scored unrelated text at 0.06-0.20 where the real model
scores it at 0.43-0.44, and its docstring said it had deliberately removed the
constant component that "would put a similarity floor under every pair" — which
is exactly the property real models have. The stub now has that floor, one test
fails if it is ever removed, and another reproduces the superseded rule and
asserts it is still fooled by the same fixture. Run against the pre-corrective
implementation, the new suite fails 13 of 18.

Tests: 939 passed, 14 skipped (836/7 at M6). 110 new across seven files, one of
which mocks nothing between itself and Ollama and re-measures the similarity
separation on every run. 43/43 checks in a real Firefox, reproduced.
Docker build clean.

Four other defects found by review or by the browser run were fixed here rather
than carried: an unreachable relevance constant that appeared to enforce
something and did not; acceptance tests using the wrong fixture files, so G07's
trap was never exercised; a bidirectional override surviving into displayed
filenames; and, from the implementation pass, the Insights panel showing M5's
two state sections as raw keys and the source inspector refetching on every
keystroke.

M7 was independently reviewed, which returned PASS WITH CORRECTIVE WORK
REQUIRED. Both blocking findings are closed, and closeout resolved the
embedding-model calibration boundary the corrective pass had left as debt.
planning/reports/M7-IMPLEMENTATION-REPORT.md carries the review, the corrective
closeout and the closeout verification in sequence, none overwriting another.

Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_017HdaXiFbscatQaLS7dJk6b

planning/CONTEXT-AND-MEMORY.md

@@ -709,6 +709,33 @@ Output generation reserve      protected
 
 Exact percentages should be configurable or derived from model context size.
 
+### As implemented (M7, for imported knowledge)
+
+The knowledge budget is a share of what is left after everything protected and
+the reply reserve are subtracted, and it is spent in authority order:
+
+```text
+always-included Canon   protected. Counted with the system block, before any
+                        history is chosen, and capped at 20% of the whole
+                        context budget. If it cannot fit alongside the other
+                        protected sections and the reply reserve, the turn fails
+                        with `ContextOverflow` rather than sending a prompt
+                        known to overflow. What does not fit is reported as
+                        dropped, with its token cost.
+retrieved knowledge     33% of what is left, filled Canon first, then Reference
+                        (capped at half the knowledge budget), then Inspiration
+                        (capped at a quarter). Whatever is not spent returns to
+                        the story history rather than being lost.
+```
+
+So Reference and Inspiration cannot crowd out retrieved Canon, and none of the
+three can reach the current authoritative state, the reader's input, the narrator
+rules, critical Canon or the output reserve — all of which are priced before the
+knowledge budget exists.
+
+Every included passage's token cost is in the context report, and so is every
+passage there was no budget for.
+
 ## 30. Protected vs Elastic Context
 
 ### Protected
@@ -921,6 +948,24 @@ FTL does not exist.
 
 This should not disappear just because the current user input does not semantically resemble "FTL".
 
+### As implemented (M7)
+
+A Canon source may be marked `always_include`. Its passages are supplied on every
+turn whatever the scene is, in their own protected section framed as standing
+rules of the world. The flag is **Canon's alone** — it bypasses relevance
+entirely, and asserting unranked Reference on every turn would spend a protected
+budget on material that establishes nothing — and it is enforced both ways: a
+source reclassified away from Canon loses the flag.
+
+Always-included Canon does not set the relevance floor for the passages that had
+to earn their place, because it did not earn its own; letting it do so would let
+one standing rule silence everything the scene actually turned up.
+
+Entity-linked and tag-based retrieval are **not** implemented. Entity names do
+reach the query — it is built partly from the authoritative state, so the
+characters and places in play are among the search terms — but there is no
+explicit link from a source to an entity, and no tags. Deferred.
+
 ## 42. Global Canon
 
 Some canon should always be active.
@@ -991,6 +1036,22 @@ The narrator may receive hidden information while being instructed not to reveal
 
 This is a prompt discipline requirement.
 
+### As implemented (M7)
+
+Source-level, and treated as prompt discipline exactly as this section says. A
+source marked `hidden` is retrieved and supplied to the narrator like any other,
+and two things mark it: the passage itself carries `[narrator only]` on its
+provenance line, and the knowledge rule in the system block says what that means
+— the protagonist does not know it, must not be told it, must not act on it, and
+a direct question about it is answered from what the protagonist actually knows.
+
+The marker travels on the passage rather than only in the preamble because a
+passage is read where it sits. Per-chunk visibility is deferred (§69 of
+`IMPORTED-KNOWLEDGE-DESIGN.md` asks only for source level in v1).
+
+"Hidden" is about the protagonist, not about the person running the campaign:
+the source is fully readable in the knowledge panel.
+
 ## 47. Story Style Memory
 
 Some user preferences may be durable within a campaign: