Evidence-First Claude Session MiningLesson 8/8
ENPT-BR
Proof and continuation

Proof at the Real Boundary

Tests establish behavior; only the compiled live path plus an independent probe closes the capability claim.

U1 PROVEN · U2 PROVENSYNTHETIC EXAMPLE · /demo-claude/
Proof ladder from fixture tests to real boundary and independent probefixture testsnecessaryCLI testsnecessarycompiled U1PROVENcompiled U2PROVENindependent reviewACCEPTED
Proof ladder from fixture tests to real boundary and independent probe.
You will be able to
  • Match each claim to the receipt that actually proves it.
  • Read U1 and U2 status from their distinct receipts.
  • Name the evidence that closed the U2 gate.
1

Core model

A proof ladder has different rungs. Unit tests prove algorithms on controlled inputs. CLI tests prove parsing and exit semantics. A compiled live run proves the operator path. An independent metadata or byte probe cross-checks the result. The course validator proves the course—not the archive capability.

Mental model: A bridge model, material test, and inspection report all matter, but traffic across the finished bridge is the boundary proof. None of the earlier receipts can impersonate the last one.

The loop log records both live boundaries. U1 counted 661 eligible JSONL and 752,116,382 bytes. U2 later completed on 661 JSONL and 752,603,828 bytes with 136,770 physical records, 658 windows, and zero faults. An independent walker matched 661/661 file hashes and the exact byte total; a stable clone also produced byte-identical manifests. Therefore U1 and U2 are PROVEN.

2

Boundary table

QuestionObserved boundaryOwnerVerdict
Focused testsAlgorithm and fixture contractsBoth unitsNecessary
Compiled live CLIOperator boundaryU1PROVEN
Independent cross-checkCounts and bytesU1PROVEN
Live manifest + byte/hash probeU2 real archiveU2PROVEN
Primary local sourcedocs/fable-session-mining/LOOP-LOG.md
# U1 receipt (sanitized)
compiled inventory: 661 JSONL · 752116382 bytes
independent metadata probe: MATCH

# U2 receipt (sanitized)
compiled manifest: 661 JSONL · 752603828 bytes
136770 records · 658 windows · 0 faults
independent file/hash/byte probe: MATCH 661/661
sentinel content: ABSENT · validator: ACCEPTED
Fixture proof: Controlled evidence for a focused behavior.Fixture proof
Controlled evidence for a focused behavior.
CLI proof: Argument parsing, output shape, and exit semantics.CLI proof
Argument parsing, output shape, and exit semantics.
Boundary proof: The real compiled consumer path completes honestly.Boundary proof
The real compiled consumer path completes honestly.
Independent probe: A second mechanism cross-checks the same invariant.Independent probe
A second mechanism cross-checks the same invariant.
3

Synthetic lab

Choose a receipt. What exactly does it prove?

4

Retrieval and feedback

Predict before revealing

Which receipts, beyond green tests, permit the course to label U2's live boundary proven?

A compiled live run, an independent 661/661 file/hash/byte match, sentinel exclusion, stable-clone determinism, and an ACCEPTED independent review.

Retrieval cards

Retrieval cardsFixture proofActivate to flip
Fixture proofControlled evidence for a focused behavior.
Retrieval cardsCLI proofActivate to flip
CLI proofArgument parsing, output shape, and exit semantics.
Retrieval cardsBoundary proofActivate to flip
Boundary proofThe real compiled consumer path completes honestly.
Retrieval cardsIndependent probeActivate to flip
Independent probeA second mechanism cross-checks the same invariant.

Check your model

What is the current honest status?

The authoritative loop log contains compiled live receipts and independent cross-checks for both boundaries.
5

Evidence ledger

Focused contracts · PROVENU1 + U2 tests
U1 live boundary · PROVENcompiled receipt recorded
U2 live boundary · PROVENcompiled receipt recorded
Independent review · PROVENACCEPTED · no P0/P1

Authority: docs/fable-session-mining/LOOP-LOG.md. Tests never upgrade a missing live receipt.