The current projection builds reproducibly and permits no canonical writes.
Method, quality and authority
Know what is ready.
Know what is not.
Countercase keeps historical taxonomy, evidence classes, crosswalks, estimates and confidence assessments explicit. Automation can move research; it cannot inherit judgment over the CivStudy Evidence Corpus.
A successful build proves reproducibility, not historical or quantitative readiness. Material integrity and semantic decisions remain open.
1 duplicate card ID and 5 duplicate reference IDs remain visible for review.
Measurement contract
Extraction is not publication.
Recognition, extraction and governed publication are distinct measurement layers.
Inspect the four layers →- V0.1 recognized
- 0 series
- Corrected legacy
- 593 series
- Typed migration
- 12 observations
- Governed
- 1 series
Corpus extension program
References first.
Estimates second. Scale last.
The integration sequence is now machine-readable and visible in the product. It prevents a bulk metadata backfill from manufacturing confidence the evidence does not support.
Epistemic contract
Taxonomies and fail-closed acceptance rules defined
contract testsDossier assessment overlay
8 of 8 focus dossiers classified
overlay checksumReference enrichment
8 mapped excerpt receipts across 2 normalized dossiers plus 2 Qing execution receipts; 0 human-reviewed
named evidence and semantic review before promotionEstimate migration
2 legacy series and 12 observations typed; 1 derived result reproducible and withheld
upstream row pointers and source-native entity lineageExtension objects
12 schema-extension candidates await typed object contracts
human schema decisionDeterministic backfill
Full-corpus confidence backfill prohibited until CE-2 and CE-3 mature
reviewed release compiler- Preserve the stable reference ID while adding creators, title, publisher, date, identifiers, locator and version.
- Assign a claim-scoped role; a work can play different roles for different claims.
- Do not treat the presence of a URL as source verification.
- Use typed relations for cites, references, versions, parts and supplements.
- Retain the source-native value, subject, unit and vintage unchanged.
- Declare truth class and every applicable estimate tag.
- Store bounds or ranges without silently averaging them.
- Bind each transformation to a method version and inputs.
- Attach a multidimensional confidence assessment and review receipt before publication.
- Legacy evidence_quality remains visible as an author label only.
- A confidence band requires a rationale and all six dimensions.
- Dimension values aid inspection; they do not mechanically average into the band.
- Likelihood language is used only when a probability has a defensible statistical meaning.
Corpus migration ledger
Content parity,
with the debt exposed.
All 742 unique bodies now carry pinned source paths, Git blob identities and body checksums. Migration is complete at the byte-preservation layer; interpretation and content quality remain governed work.
Every registry row is represented; one duplicate stable ID remains explicit.
The unique-ID total maps one-to-one to 742 verified source paths at a pinned repository commit.
Stable ID, title, type, summary, citations, relationships and CAD facets are projected.
All complete Markdown bodies are preserved verbatim across 30 lazy-loaded shards (4,720,066 bytes; 454,681 words).
Every unique card is bound to a verified source path, exact branch commit, Git blob SHA and body SHA-256.
Thirty-three current-schema and twelve schema-extension candidates remain proposals. They are not counted as migrated canonical cards.
CM-2 complete · repair queue
54 thin records,
none silently rewritten.
The full-body audit found no missing required frontmatter, absent section structure or placeholder markers. It did identify 54 bodies under 80 words. Each now has a typed, priority-ranked research proposal with source receipts and review requirements.
Open the thin-record lens →- P0
- 26 comparative or high-connectivity
- P1
- 24 connected concepts
- P2
- 4 bounded backlog
- Auto-applied
- 0 canonical writes prohibited
CM-3 + CM-4A complete
One proposition.
One finish contract.
Represent every priority historical proposition with one inspectable pattern linking inherited text, scope, truth class, evidence, estimates, confidence and review posture without rewriting the canonical card. Every proposition now also carries a typed execution packet; the queue is complete even though the evidence is not.
Every core proposition across the eight priority dossiers now has a stable identity and source-body receipt.
Every existing atomic evidence claim maps into the shared pattern without inheriting support into broader interpretation.
All six dimensions are present at claim level; every assessment remains staged and unreviewed.
Normalization improves inspectability. It does not manufacture source maturity or reviewer authority.
4 P0 packets · 19 source records
3 newly ready for named review · 1 owner receipt · 0 specialist receipts
Inspect the research receipts →- CM-4B.2B.2 · Named specialist evidence receipts · active
- CM-4B.2C · Regional, semantic and crosswalk co-review · active
- CM-4B.2D · Publication and Git promotion authorization · blocked
Pattern alignment
All 55 inherited rows now include their complete body.
Body extraction recovers typed case mentions for 32 records. None can publish until claim structure, counterevidence, falsifiers and human review are present.
Browse all pattern records →- Structure aligned
- 0
- Partial
- 38
- Reclassify
- 15
- Broken edges
- 9
Product contract
Four layers, four authorities.
The design permits fast research and disposable interfaces without making temporary output indistinguishable from committed historical judgment.
Committed Markdown
Canonical historical corpus
Read onlyGenerated read model
Disposable product projection
ReplaceableResearch state
Runs, evidence, reviews and proposals
Append onlyGit proposal
Reviewed promotion mechanism
Separate actionNon-negotiable invariants
The product fails closed.
- No interface action edits a committed historical card.
- No source-bound proposal advances without an eligible snapshot or separately accepted, proposal-specific exception.
- No dataset region silently becomes a civilization, empire or cultural sphere.
- No range, missing value or contested result is collapsed merely to complete a chart.
- No workflow approves its own evidence or semantic decision.
Current read model
Broad coverage, visible debt.
Scale is not readiness. The source manifest proves 742 unique identities and complete body preservation; the duplicate registry row, thin records and ungoverned quantitative observations remain explicit.