Add bitemporal, token-efficient recall and reproducible quality evidence - #88
Conversation
![chatgpt-codex-connector[bot]](https://avatars.githubusercontent.com/in/1144995?s=60&v=4){kind=small}
There was a problem hiding this comment.
💡 Codex Review
Here are some automated review suggestions for this pull request.
Reviewed commit: 8f54b156e6
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".
If Codex has suggestions, it will comment; otherwise it will react with 👍.
Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".
There was a problem hiding this comment.
💡 Codex Review
Here are some automated review suggestions for this pull request.
Reviewed commit: bce39522d9
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".
If Codex has suggestions, it will comment; otherwise it will react with 👍.
Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".
There was a problem hiding this comment.
💡 Codex Review
Here are some automated review suggestions for this pull request.
Reviewed commit: fedac81339
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".
If Codex has suggestions, it will comment; otherwise it will react with 👍.
Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".
There was a problem hiding this comment.
💡 Codex Review
Here are some automated review suggestions for this pull request.
Reviewed commit: 9bc908c4fb
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".
If Codex has suggestions, it will comment; otherwise it will react with 👍.
Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".
There was a problem hiding this comment.
💡 Codex Review
Here are some automated review suggestions for this pull request.
Reviewed commit: 501330ddfd
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".
If Codex has suggestions, it will comment; otherwise it will react with 👍.
Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".
There was a problem hiding this comment.
💡 Codex Review
Here are some automated review suggestions for this pull request.
Reviewed commit: be8a9b8cef
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".
If Codex has suggestions, it will comment; otherwise it will react with 👍.
Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".
There was a problem hiding this comment.
💡 Codex Review
Here are some automated review suggestions for this pull request.
Reviewed commit: 043c332722
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".
If Codex has suggestions, it will comment; otherwise it will react with 👍.
Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".
There was a problem hiding this comment.
💡 Codex Review
Here are some automated review suggestions for this pull request.
Reviewed commit: 01787cdb67
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".
If Codex has suggestions, it will comment; otherwise it will react with 👍.
Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".
There was a problem hiding this comment.
💡 Codex Review
Here are some automated review suggestions for this pull request.
Reviewed commit: ac2d99478f
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".
If Codex has suggestions, it will comment; otherwise it will react with 👍.
Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".
There was a problem hiding this comment.
💡 Codex Review
Here are some automated review suggestions for this pull request.
Reviewed commit: 632c1d83e2
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".
If Codex has suggestions, it will comment; otherwise it will react with 👍.
Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".
There was a problem hiding this comment.
💡 Codex Review
Here are some automated review suggestions for this pull request.
Reviewed commit: e3f8858647
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".
If Codex has suggestions, it will comment; otherwise it will react with 👍.
Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".
There was a problem hiding this comment.
💡 Codex Review
Here are some automated review suggestions for this pull request.
Reviewed commit: 5ff733dd68
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".
If Codex has suggestions, it will comment; otherwise it will react with 👍.
Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".
There was a problem hiding this comment.
💡 Codex Review
Here are some automated review suggestions for this pull request.
Reviewed commit: dc776c51bf
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".
If Codex has suggestions, it will comment; otherwise it will react with 👍.
Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".
| identifier, | ||
| limit=per_term, | ||
| requested_historical=historical, |
There was a problem hiding this comment.
When a repository has more than per_term matching symbols (at most 12) whose names alphabetically precede the exact identifier, Store.search_symbols() applies its ORDER BY name LIMIT before this code can assign the exact-match boost. The exact symbol is therefore absent from symbols, so memories bridged only to that symbol disappear from the code recall arm; request exact name/fqname matches first or rank exact matches before applying this cap.
AGENTS.md reference: AGENTS.md:L94-L100
Useful? React with 👍 / 👎.
| sql += " ORDER BY id LIMIT ?" | ||
| params.append(max(0, int(limit))) |
There was a problem hiding this comment.
When a scope has more than 2,048 entities containing a common query term, the broad OR-substring predicate admits all of them and this ORDER BY id LIMIT keeps only the oldest IDs before the exact entity-pattern check runs. A newer entity whose full name actually occurs in the query can consequently never become a seed, causing both the 1-hop and PPR graph arms to miss its attached memories and relations; prioritize exact/full-name matches before applying the cap.
AGENTS.md reference: AGENTS.md:L94-L100
Useful? React with 👍 / 👎.
1 participant
What changed
valid_at/known_atbi-temporal reads across memory, graph, support, code, sync, and historical-link evidence while preserving theas_ofcompatibility aliasengraphis-benchmark/v2, adversarial fixtures, strict artifact validation, fixed-budget curves, bootstrap confidence intervals, LongMemEval-V2 adapter plumbing, canonical export/receipt verification, and release-evidence toolingWhy
Engraphis should improve grounded memory quality per context token without rewriting history, leaking evidence across scopes, or making unsupported competitive claims. This change makes those guarantees measurable, reproducible, and fail-closed.
Three final-audit defects were also resolved:
known_atviewsMeasured offline fixtures
1.000across 26 questions / 44 memories17,172 -> 7,663serialized response tokens, avoiding9,509tokens (55.38%reduction) with unchanged retrieval scores808.8 -> 219.0tokens (72.9%reduction) at Recall@51.000162.2 -> 42.4tokens (73.9%reduction; about3.8xsmaller) with evidence returned for the same 18 fixture questions1,500-token budget with observed mean87.73and maximum106tokens10/10correct (5/5grounded,5/5abstained)The chunking, compact-response, and budget-accounting measures use different counting boundaries and are not added together. They are deterministic repository fixtures, not official external QA results, model-billing savings, or named-system outperformance claims.
Validation
The release-evidence manifest remains a release-time artifact: it requires an actual release tag plus the completed release workflow. Official complete-dataset LoCoMo/LongMemEval/LongMemEval-V2 runs and leaderboard submissions have not been performed.