[AdaWorldAPI]

What

#24 read property history by scanning the Lance log on every call. Measured, that took 336 ms, 320 ms of it the scan, against P-HIST-1's < 20 ms budget.

The rule this applies (lance-graph, 2026-09-24): one write costs about a million folds, and a Lance scan is recovery cost. So history should be a fold over records already in memory, not a read of Lance.

• Log keeps its records resident. Every record it commits or replays stays in memory, at 29 B each.
• replay is the one Lance read. On any error nothing becomes resident; the log keeps what it held before.
• commit appends to the resident records after the Lance append. It refuses with Diverged when the records are not at the table's version, because otherwise history would silently miss earlier batches.
• history is synchronous now. It runs the same Quack program over memory and never opens the dataset.
• No separate corruption check. A corrupt record cannot become resident, because replay refuses it. So the required-lane check from D-STO-5 (read): property history from the Lance log, with its measured cost #24's review holds structurally, and the per-match check is gone.

No index, no column projection, no Lance filter expression, and no compaction. ISS-HIST-SCAN-COST is resolved. Plan §5.2 records the decision: derived reads write nothing, and only learning adds persistent state, recorded after as one merged, auditable outcome.

Evidence

Timing. Release, examples/probe_history.rs, 100k records, 1,000 versions × 1,000 updates, 1.2M log records:

read	median
history, resident (101 reads)	0.72 ms (0.64–1.99)
replay from Lance (5 cold runs)	532 ms, once at startup
history read from Lance (#24, for reference)	336 ms

Tests (cargo test -p hubspo-store --features lance; the log suite dropped from 16 s to 0.44 s):

• Histories equal an independently written oracle for every record and both lanes.
• A second log replayed from disk gives identical histories.
• A log without resident records refuses to commit.
• A corrupt record is refused by replay and never reaches history: the resident records are unchanged.

Disable runs, all red:

disabled	red
resident-version check	yes
records installed before replay validates	yes
commit not appending to the resident records	yes
row = r term	yes
op = DELETE arm	yes
op = SET term	yes
I32 sign bit	yes

Local: fmt, clippy (-D warnings, default and lance) and the workspace tests pass. No other HubSPO-rs PR is open; the branch is current with main.

🤖 Generated with Claude Code

https://claude.ai/code/session_01PPcakjk7ioQsnasmNDtc5z

---

Generated by Claude Code

[chatgpt-codex-connector]

Codex Review Summary

This comment shows the latest Codex review activity on this pull request.

Review	Status	Commit	Review trigger
📝 Code Review	✅ Completed 2026-10-10T19:24:44.727307Z	ffc56bf	PR opened

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

• Open a pull request for review
• Mark a draft as ready
• Comment "@codex review" or "@codex security review".

Codex reacts with 👀 while any review is running, comments if it has suggestions, and reacts with 👍 once all reviews finish with no findings.