Jesse.Markowitz/interactive-story
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

M7: a first-class imported knowledge library

Browse Source 
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
This commit is contained in:
JesseMarkowitz
2026-09-06 15:40:13 -04:00
co-authored by Claude Opus 5
parent a6e9c7a32b
commit 480414efe0
52 changed files with 10894 additions and 52 deletions
Download Patch File Download Diff File Expand all files Collapse all files
planning/README.md
+49 -17
View File
@@ -6,7 +6,11 @@
milestones **M1 through M6 implemented and accepted** (M3 and M4: 2026-09-03;
M5: 2026-09-04; M6: 2026-09-06). M5 and M6 were each accepted only after an
independent review found a real defect and a corrective pass fixed it.
**Next:** **M7 — First-Class Imported Knowledge Library.** Its brief has not
**M7 — First-Class Imported Knowledge Library — is complete** (2026-09-06),
after an independent review, a corrective pass and a closeout verification. Its
report keeps all three in sequence: `reports/M7-IMPLEMENTATION-REPORT.md`.
**Next: M8 — Browser UX Completion for v1 Story Operations.** Its brief has not
been written yet, and writing it is the current action.
**Package version:** see `VERSION.md`, which records what each revision changed
@@ -69,7 +73,7 @@ Two standing qualifications:
| Document | What it is for |
| --- | --- |
| `SPECIFICATION.md` | What the product must do. The top of the authority order. |
| `TECHNICAL-DESIGN.md` | The selected architecture, including what M1-M3 built, recorded as fact. |
| `TECHNICAL-DESIGN.md` | The selected architecture, including what M1-M7 built, recorded as fact. |
| `DATA-MODEL.md` | Entities, the stored head, branch disposition, and the export shape. |
| `STORY-BRANCH-SEMANTICS.md` | Undo/Redo/Retry/branch/take behavior, including the M3 ratifications. |
| `CONTEXT-AND-MEMORY.md` | Prompt assembly, summarization, branch-safe memory. |
@@ -97,8 +101,8 @@ Two standing qualifications:
10. `BROWSER-UX-SPEC.md`
11. `V1-ACCEPTANCE-TESTS.md`
12. `DECISIONS/` — all of them; they are short.
13. `reports/M4-IMPLEMENTATION-REPORT.md`, for what the last milestone actually
left behind. Nothing in `planning/archive/` unless sent there.
13. `reports/M7-IMPLEMENTATION-REPORT.md`, for what the last accepted milestone
actually left behind. Nothing in `planning/archive/` unless sent there.
## Architectural decisions
@@ -117,6 +121,7 @@ Active ADRs, all of which still constrain current or future work:
| `010-explicit-typed-narrative-state-events.md` | Explicit typed events / absolute assignments, not relative deltas. |
| `011-local-inference-endpoint-policy.md` | Address allowlist, deny by default, checked twice, TLS mandatory. |
| `012-active-head-non-destructive-history.md` | The stored active head. **The architecture implementing ADR 005.** |
| `013-authoritative-narrative-state-document.md` | The authoritative state document, its pipeline, and where each part lives. **The architecture implementing ADR 010.** |
`008-phase0-before-build-plan.md` was a process gate — do not begin production
work until Phase 0 closes — which Phase 0 satisfied on 2026-09-01. It is in
@@ -128,12 +133,16 @@ work until Phase 0 closes — which Phase 0 satisfied on 2026-09-01. It is in
`reports/` holds the report for the milestone most recently completed, because
that is the one the next milestone's planning has to consult:
- `reports/M4-IMPLEMENTATION-REPORT.md` — M4's review and its evidence record.
- `reports/M7-IMPLEMENTATION-REPORT.md` — M7's independent review, its
corrective closeout and its closeout verification, in that order and none
overwriting another. It is the longest report in the package because M7 is the
milestone whose first implementation was most wrong, and the sequence is the
point: what was claimed, what was measured, what that forced.
Completed earlier milestones are in `archive/milestone-reports/`, which M3's
report joined when M4's landed: a milestone report is useful during the
immediate next milestone and historical afterwards. M3's is now at
`archive/milestone-reports/M3-IMPLEMENTATION-REPORT.md`, unedited.
Completed earlier milestones are in `archive/milestone-reports/`, which M6's
report joined at M7's closeout: a milestone report is useful during the
immediate next milestone and historical afterwards. M1-M6 are all there,
unedited.
## The decision this package rests on
@@ -236,22 +245,45 @@ Milestone M3 COMPLETE (2026-09-03)
|
v
Milestone M4 COMPLETE (2026-09-03)
named Save Points reports/M4-IMPLEMENTATION-REPORT.md
named Save Points archive/milestone-reports/M4-*.md
| browser verification: PASS (M3 + M4)
v
M5-M11, one at a time see BUILD-MILESTONES.md
Milestone M5 COMPLETE (2026-09-04)
authoritative narrative state archive/milestone-reports/M5-*.md
| and ADR 013
v
Milestone M6 COMPLETE (2026-09-06)
branch-safe context, summaries archive/milestone-reports/M6-*.md
and long-term story memory
|
v
Milestone M7 COMPLETE (2026-09-06)
first-class imported knowledge reports/M7-IMPLEMENTATION-REPORT.md
library review + corrective + closeout, in sequence
|
v
M8-M11, one at a time see BUILD-MILESTONES.md
```
## Stop Rule
**One milestone at a time. Do not begin a milestone before its brief exists.**
**No M7 brief has been prepared.** Writing one is the current action, informed by
the M6 report's §W readiness assessment and by the retrieval debt
`BUILD-MILESTONES.md` records against M6 — in particular that ranking is
similarity plus a pin, so imported material will compete for the same memory
budget as story memory, and that cross-layer duplication
(`CONTEXT-AND-MEMORY.md` §22) is still open.
**No M8 brief has been prepared.** Writing one is the current action, informed by
the M7 report and by the debt `BUILD-MILESTONES.md` records against M7 — in
particular that the knowledge panel and the Insights knowledge rows are
functional rather than designed, that a "nothing was relevant enough" result and
an uncalibrated-embedding-model warning are both surfaced plainly and want a
considered treatment, and that `BROWSER-UX-SPEC.md` §47's import preview and
§52's retrieval-usage count are not built.
The M6 retrieval debt this milestone was warned about is partly addressed and
partly still open. Imported material does **not** compete with story memory for
one budget — M7 gave it a separate bounded budget of its own — and ranking for
imported knowledge is now relevance × class rather than similarity plus a pin.
Story-memory ranking is unchanged, and cross-layer duplication
(`CONTEXT-AND-MEMORY.md` §22) is still open: the same fact can still appear in
state, memory, history and now an imported passage at once.
**No conditions remain open on M1-M6.** The browser smoke condition that M3 and
M4 both carried was satisfied at M4 closeout: a real Firefox exercised the