Round 2 found the functional side true and six record/discipline defects.
- D9: the write-path line said "comparison uses secrets.compare_digest" as if
it covered both tokens. It does not: the operator's plaintext is compared with
compare_digest (api_auth.py:54, :110), a holder token is SQL equality on its
sha256 digest (holder_store.verify_token) because no plaintext is stored. The
boundary is now stated in verify_token's docstring as well as the record.
- D10/D13: three historical numbers were written as if they were current. The
four-file 59 is pinned to 2ea48d8 (test_trust_quarantine.py was 25 tests there,
26 from d56996b on), and both 401 gate lines are pinned to the commit they were
measured on (83dfc14, 89c880f) with the current 406 named next to them.
- D12: the compiler cache-key line overstated one direction. The old key had no
identity part at all, so whoever compiled first got served to the next caller;
an empty pack reaching the operator is one direction, not the mechanism.
- D11: the p2-5 acceptance cell's suite count is pinned the same way.
- D14: bench/video_audio_check.py wrote straight into bench/reports/, so a
reviewer running it in its own worktree would have overwritten frozen
evidence. It now takes --report/--payload and refuses a destination that
already holds a frozen report; refuse_frozen_overwrite is tested, along with
the guards that stop the harness from measuring nothing.
Gates: pytest -q 410 passed; ruff clean; bench.filesize_check OK;
bench.audit --product 33 harness files / 2 reports / 6 set files, clean;
pyright 0 errors on the three touched files; mypy cold cache Success on the same
three.
LEGS.md said "schema v9" while every arm ran on v10. The harness read
PRAGMA user_version straight after copying the store, before the product
had opened it, so it printed the version the copy arrived with. The copy is
migrated on first open, and that is the schema the retrieval service ran on.
The version is now read after the arms, and the report line says both: the
copy arrives at v9, the product migrates it when it opens, every arm ran on
v10. A test pins the sentence so the two versions cannot collapse back into
one number.
Gates: bench/tests 68 passed; ruff clean; mypy clean on the three files.
A memory used to have one state: present. A session could not be reloaded
without searching, nothing separated what a person said today from what the
store ingested a year ago, and two statements that disagree about a number
were both handed to the reader as if the store believed both.
Schema v10. Atoms gain session_id, importance, recall_count, last_recalled_at
and archived_at; two tables arrive beside them — consolidations (one watermark
per holder+scenario lane) and conflicts (a pair, the term they disagree on,
both values, and whether anyone has settled it). v_atoms_current now also
hides archived rows, and apply() drops the view before rebuilding it because
CREATE VIEW IF NOT EXISTS would keep the old body and ignore the archive.
Working memory is a read, not a store: `working <holder> <session>` returns
what that session recorded, in order, with no search involved. Consolidation
rolls a lane's new memories into one standing summary and moves the watermark
to the newest memory actually rolled up — not to the clock of the pass, which
made a second pass count the same memories twice. The archive rule is stated
in the command that runs it (recorded before N days, never returned by a
search, importance at or under a floor) rather than hidden in a decay formula
nobody measured; --dry-run names the victims first.
A conflict is recorded when a new memory repeats a label with a different
number in the same holder's same scenario. Retrieval drops the older side
while the conflict is open and both sides answer again once it is settled:
settling is not superseding, and a word-only disagreement is left alone
because a reworded sentence would fire on every paraphrase. Numbers are keyed
to both the memory's subject and the nearest preceding label, measured: the
two thermostat sentences share only their subject, so a nearest-word-only rule
found nothing.
Search now counts the memories it handed back, which is the only use estimate
the store has, and the archive rule reads it. The dense index applies the same
currency rules as the view, so an archived memory cannot stay findable through
one leg while the other hides it.
Two reads over HTTP (working, conflicts) and two tools over MCP
(working_set, memory_conflicts); curation stays with the operator because an
agent should not decide what its own past is worth. remember and supersede
carry a session id; importance does not cross MCP, because recall_count is
observed and importance is stated.
Gates: 366 passed; ruff clean; mypy 24 errors in 9 files, the recorded 9cb46c1
baseline, net 0; pyright 0 on the touched files; go build/vet/test and gofmt
clean; mcp_write_smoke PASS.
One stride over file order, truncated to the probe count, dropped the
tail of the suite: a 300-probe sample held 128 code, 86 pdf, 36
screenshot, 26 image and 24 video queries and no conversation query at
all, so every report built from it measured five of six modalities.
The budget is now split by suite weight with a floor of one probe per
modality, and a test pins that the frozen sample covers all six.
The set's question type rides along on each query, and a second metric
asks the question recall@10 cannot: when the answer lives in several
dated memories, did the newest come back first? A knowledge-update
question scores 1.000 on recall while the ranking still leads with the
value the speaker has since replaced, and that is the failure the next
task is meant to fix.
Sixteen mutating routes now refuse a caller who cannot present the token the
server started with; queries stay open, because a question is not a write. The
environment variable is ELY_EYE_AUTH_TOKEN, following the repo's env_prefix
convention — the shorter name I reached for first returned 403 where the smoke
expected 401, and the server printed the reason.
Words from a stranger are quarantined on the way in: an HTTP upload is
third_party by definition, and a sentence carrying an instruction-shaped phrase
is quarantined whoever wrote it. A quarantined atom stays findable and stays out
of the compiled context until a second, live, unquarantined source agrees with
its rarest words. Corroboration is computed at read time because it is a
property of the corpus now, not of the moment a memory was written; it is
term-level, and the docstring says so.
forget now erases the source row and the compiled contexts beside the atoms, so
a forgotten memory leaves nothing that can still speak. The semantic cache key
carries a store revision, which fixes a real staleness: a context compiled
before new evidence arrived kept its old quarantine decision forever.
bench.audit --product still pointed at evals.py, which the p1-6 split moved into
backend/src/ely_eye/proofs/. It failed fast, but the anti-self-justification
rule had quietly stopped auditing the product; it now walks the package, and a
test pins that the package exists and holds no re-roll.
Gates: pytest 306 passed; ruff clean; mypy cold cache 24 errors in 9 files, the
recorded baseline, net zero; pyright 0 errors on the 23 backend files and the 2
bench files; go build/vet/test and gofmt clean; npm run build 221.39 kB with
eslint exit 0; bench/filesize_check clean; bench.audit --product clean over 31
harness files; scripts/mcp_write_smoke.sh PASS with the gate live across two
servers and the MCP child.
A review of the p2-1 record found seven places where the prose had drifted
away from the reports it cites. None were product defects; all were record
keeping, and record keeping is what a memory system is judged on.
The quantization row mis-transcribed nine latencies, and the two worst made
dimension truncation look twice as fast as the full scan when the report
measures six percent. The extrapolation table scaled latency by atoms and
bytes by vectors under one heading that named neither switch. The picture
index stats existed only in prose, so the harness now reads the index shape
off the artifact and times the encoder on a sample, which reproduces the
hand-timed build within seven percent. The dense arms named a store that
holds no vectors, so the report names the store the vectors came from and
refuses to run where a dense leg would rank an empty index.
The frozen baseline was migrated in place by the product CLI while the
picture index was built, so its pinned mtime, digest and schema version are
re-pinned and the step that moved it is named. The sqlite_vec import added
two mypy errors; the fix is the missing stub declaration, not an inline
ignore, and the count is back to the recorded baseline of twenty-four.
The product fuses a sparse page, a dense vector leg, and a picture leg that
only speaks when the words come back thin. Two of the three are off by
default, and until now no committed artifact said why.
bench/leg_check.py asks the frozen suite's 300 answerable queries as their
own holder, once per configuration, and prints recall, first-hit quality,
latency and per-modality recall in one table. The picture index and the
CLIP snapshot are 48 minutes of encoding, so they are carried in rather
than rebuilt per row; the harness refuses to run a leg whose snapshot is
missing instead of quietly scoring nothing.
The numbers that came back:
- words only, 10 atoms: recall@10 0.832, MRR 0.683, p50 64ms
- words only, the configured 24: 0.848, 0.688, 69ms
- words only, page 30: 0.855, 0.690, 67ms
- dense leg on: 0.845 / 0.858, MRR down, p50 469ms / 411ms
- picture leg available: 0.815 / 0.825, MRR 0.553 / 0.551, p50 148ms / 164ms
So both defaults stay off, and the acceptance bar (recall@10 >= 0.90) is
not met by the product path yet: 0.858 is the best measured row, the
in-repo counterpart at holder scope is 0.866. The loss concentrates in
picture golds, and the current CLIP leg pays more in dilution than it
returns there.
CHECKLIST p2-1 and p0-3.7 now quote the committed reports (DENSE.md,
QUANT.md, SCALE.md, TERMS.md) instead of scratch scripts.
The 500-line rule and the one-line-comment rule were enforced only from
memory, so they drifted. bench/filesize_check.py now checks the budget on
every tracked source file, and it passes.
Split, with behaviour proven unchanged by comparing each moved function body
against HEAD byte for byte (53 training bodies, 21 benchmark bodies, 0
differences):
db.py 503 -> 396 store_search.py 105
bench/run.py 609 -> 351 query_set.py 80, report.py 211
benchmarks.py 619 -> 424 benchmark_scoring.py 262
training.py 1630 -> 94 training_{common,encode,metrics,datasets,lora,retrieval}
TrainingService stays as a thin facade for the CLI; the facade's parameter
lists are pinned against the module functions by inspect, which is what
catches a leaked self.
Deleted the eight top-level functions no file called, after fixing the scan
that had reported 201 false positives by skipping the defining file. The
v1/v2/v3 sweep found nothing to delete and says so: every remaining version
tag is an on-disk format marker, a migration step, a fixture name, or a
third-party model name.
Collapsed 39 multi-line comment blocks (86 lines) to one line each, keeping
the measured facts and error strings that justify the code.
269 tests pass; ruff clean; mypy 24 errors in 9 files against a 24-in-6
baseline, net zero; pyright 4, the same four HEAD measures.
The reviewer's round 2 verified all nine round-1 corrections by re-measuring
them, then failed the commit on the one thing it added that the repo's own bar
forbids: a two-line docstring on _TinyDense. p1-6.4 still lists "comments to one
line" as open work, so a new commit should not widen it. The 470MB rationale
already lives in CHECKLIST.md.
Also cites the CI run for the commit under review (37744547992 @ 20c0416, also
success) instead of stopping at the previous one.
CI failed four tests with "SystemExit: bench/models/paraphrase-multilingual-
MiniLM-L12-v2: model not downloaded". The claim those tests make is that the
joint ablation pass ranks exactly like the per-arm pass and that three arms
needing BM25 score the corpus once. That is a claim about the harness, not
about embedding quality, so it should not require a 470MB download to check.
The tests now inject a deterministic 384-dimensional stand-in dense model and
a stand-in cross-encoder at the seam both paths share. The real models still
carry every measurement in bench/reports; nothing measured changed.
Proven by the condition that broke: with both model directories moved out of
the tree, the whole suite passes (253 passed).
The frozen baseline claimed the product's top-10 shared 0.0% with the fast
BM25 path. That was a harness bug: gold names a corpus source
(mmlongbench:PH_2016.06.08_Economy-Final.pdf) while the store names a content
hash (PH_2016_06_08_Economy_Final_10e57152f44d1a2b), so the two id spaces never
met and the right document, ranked first, scored as a miss.
- bench/systems.py: `_corpus_ids` joins store sources through the
`source_map()` the counterpart already uses, and refuses a store whose
sources are not corpus files instead of silently scoring zero.
- bench/systems.py: `live_product_rankings(..., as_holders=...)` asks the
product as the query's own holder, which is how the counterpart is scored.
- bench/run.py: the live section now reports recall@10 / MRR / hit@10 with a
95% lower bound, and probes 300 queries instead of 40 — the SQL path answers
in well under a second, so the old cap was a cost estimate that expired.
- bench/run.py: `--name CURRENT` writes a post-P0 report beside the frozen
BASELINE without touching it, stamped with the commit and the store schema
version the pass started on (the live pass migrates the store mid-run, so a
plain stamp would be a lie).
- bench/stamp_holders.py: the pre-P0 store holds 165,973 memories nobody owns.
The corpus index states the owner, so holders, scenarios and event times are
stamped in place — 165,973 atoms across 50,278 sources, 0 left unowned.
Numbers, same machine, load 1.0-1.7:
- CURRENT (post-P0, schema v5): no_rerank recall@10 0.565 p50 0.643s;
fusion+rerank 0.453 p50 7.000s; graph 0.001.
- Product as its holder: recall@10 0.870, MRR 0.740, n=173.
- Counterpart: mem0_flat holder scope 0.866, whole corpus 0.412; the holder
boundary is worth +0.454 with an identical ranker and identical items.
Verification: pytest backend/tests bench/tests -q -> 173 passed; ruff and mypy
(fresh cache) clean on every touched file.
The product path pulled every atom into Python and rebuilt a BM25Okapi
index per query: p50 20.16s over 165,973 atoms. It now runs one SQLite
statement with LIMIT/OFFSET.
- text_index.py: one normalizer for both sides of the index. CJK runs are
split per character (unicode61 cannot segment them), camelCase and
separators are split so "search atoms" finds SearchAtoms, latin tokens
get a prefix match. Rows written before this existed hold raw text in
atom_fts, so schema v3 rebuilds the index on upgrade.
- retrieval.py: filters (holder / scenario / source / modality / since /
until / min_trust), a base64 cursor, and reciprocal rank fusion in place
of the 0.55/0.35/0.10 blend whose graph term was always zero.
- An empty result is {status: "no_match"}, not a bare array.
- Go: search_atoms and compile_context rank through POST /api/search. The
second Go tokenizer is deleted; two implementations could drift, which
is what the MCP/HTTP parity requirement forbids.
- db.py passed 500 lines, so the row codec moved to row_codec.py.
- Tooling: installed the repo's own pyright. pyrightconfig.json had no
exclude and swept the downloaded corpus (1171 errors); with excludes the
count is 62, all pre-existing (evals 48, profiles 4, training 4,
ingestion 3, runtime 2, cartridge_assets 1). backend/pyproject.toml
carried a second [tool.pyright] that shadowed the root config; deleted.
- bench/scale_check.py measures the product search at corpus size. The
100k-atom p50 number is not written until it is measured on an idle
machine.
Verification: 112 pytest passed; go build/vet/test clean, gofmt clean;
ruff clean on every touched file; mypy clean on the 14 touched Python
files; repo pyright 62 errors, unchanged per-file ledger.
rankings.jsonl stores the seconds each query took, so the report inherited whatever
the machine was doing at the time: the frozen baseline's dense latency included a
per-query model load, and the 50,000-source row moved from 1.315s to 0.809s between
two runs of identical code.
- bench/run.py: --remeasure-latency N re-times a stride sample with warm models and
replaces each arm's latency section, recording the sample size and a note; recall
stays a full-pass number. BASELINE.md prints that note under the table.
- bench/tests/test_joint.py: 62 tests total.
Verification: pytest bench/tests -q -> 62 passed; ruff check bench/ -> all checks
passed; mypy (fresh cache) -> no issues in 25 source files.
Three defects made the external baseline numbers and labels wrong.
- bench/systems.py: _sentence_model rebuilt the model on every query, so dense arm
latency was mostly model loading and the latency-vs-size curve was flat because a
constant dominated it. Models are now cached per process.
- bench/counterpart.py: visible() and search() walked 404k Python objects per query.
Per-item holder/source/time columns are built once and the query path is numpy;
12.5 queries/s instead of 0.7.
- bench/metrics.py: the aggregate key was named wrong_holder_confidence_unanswerable
while it counts top hits that ARE from the asking holder. BASELINE.md read that
backwards. Renamed to top_hit_from_holder_unanswerable and the sentence rewritten.
- bench/run_external.py: the by-modality table read row['answerable'] but aggregate
emits n_answerable, so the writer died after scoring finished. Each store is now
scored twice, [holder scope] and [whole corpus], so the gain from the holder
boundary is separable from the gain from ranking; progress lines every 500 queries.
Verification: pytest bench/tests -q -> 61 passed; ruff check bench/ -> all checks
passed; mypy (fresh cache) -> no issues in 25 source files.
Latency measured while another job hogs the cores is not comparable: the scale curve
moved from 1.315s to 0.809s at 50,000 sources between two runs of the same code.
- bench/systems.py: machine_facts() reports cpus plus the 1/5/15 minute load average.
- bench/run.py, bench/run_external.py: both reports carry it, and BASELINE.md labels
the latency table a ceiling when the machine was busy.
Verification: pytest bench/tests -q -> 60 passed; ruff check bench/ -> all checks
passed; mypy (fresh cache) -> no issues in 25 source files.
The frozen baseline showed the unmodified RetrievalService agreeing with an equivalent
BM25 index on 0% of its top ten results. Reproduced by hand: its top ten atoms came
from one document's pages, so a source-level ranking cannot overlap it.
- bench/systems.py: live_product_rankings also counts distinct sources behind the
product's top-k atoms.
- bench/run.py: scale probes no longer filtered by gold (a source prefix keeps almost
no gold, so the 500/2,000/10,000 rows were silently skipped); scale_arm_names reads
only latency dicts, which is what crashed write_markdown on the note field; the
unanswerable share is computed instead of hard-coded.
- bench/tests/test_joint.py: 60 tests total, including the writer regression.
Verification: pytest bench/tests -q -> 60 passed; ruff check bench/ -> all checks
passed; mypy (fresh cache) -> no issues in 25 source files.
The per-arm runner recomputed BM25 and the dense matmul for every arm, which put the
full ablation at roughly 16 hours and left no way to survive a kill. JointRunner
computes each signal once per query, attributes latency to the components each arm
uses, and checkpoints every query to runs/baseline/rankings.jsonl.
- bench/systems.py: JointRunner + ARM_COMPONENTS signal attribution.
- bench/run.py: joint pass, rankings cache with a run fingerprint, per-query resume,
latency summary, resident-memory progress lines.
- bench/tests/test_joint.py: joint rankings equal the per-arm rankings for all five
arms; the sparse corpus is scored once even with three arms that need it; cache
fingerprint mismatch and malformed rows fail fast.
- bench/reports/EXTERNAL_APPENDIX.md: what a real Mem0/Zep run would cost, and which
published numbers are comparable at all.
Verification: pytest bench/tests -q -> 57 passed; ruff check bench/ -> all checks
passed; mypy (fresh cache) -> no issues in 25 source files.
The merged database names a source by content hash (`11089_4f2a...`) while the
frozen sets name the same file by its corpus id, so every arm scored 0.000 until
load_atoms joined atoms through sources.path. It now returns corpus ids and
refuses an atom it cannot trace to a corpus file.
Also fixes found by running the harness:
- merge rebuilt atom_fts from the deduped atoms table; two attempts on one file
left 288,266 fts rows for 165,973 atoms.
- merge accepts the two honest reasons a corpus file has no product source
(500 byte-identical LongMemEval sessions collapse under content hashing, and
12 photos that ONNX Runtime cannot decode) and reports them.
- --arms used action=extend against a non-empty default, so naming two arms ran
seven.
- the dense cache is keyed by the atom set, so a scale-curve subset cannot be
scored with full-corpus vectors.
- write_markdown crashed on a run that covered only some modalities.
- bench/ingest_baseline.py drives the unmodified product IngestionService over
all 50,789 corpus files in sharded SQLite DBs, then merges them. Thread and
memory ceilings are set explicitly: ONNX Runtime defaulted to every core
(95 threads/worker, load 490) and its CPU arena kept freed pages mapped,
which OOM-killed workers.
- bench/metrics.py: recall@k, MRR, nDCG@10, Wilson lower bound, chance
baseline. No model, no product import, so the scorer cannot grade itself.
- bench/systems.py + bench/run.py: the five ablation arms over the merged DB,
latency percentiles, scale curve, and a top-10 equivalence check against the
live RetrievalService.
- bench/counterpart.py + bench/run_external.py: in-repo Mem0-style flat store
and Zep-style temporal store as the external reference point.
- bench/audit.py: enforces single-attempt scoring, frozen n, a chance row
beside every score, and producer != verifier. It already flags the product's
retry-until-target loop at backend/src/ely_eye/evals.py:2351.
- backend/pyproject.toml: rapidocr-onnxruntime>=1.4.4 was unsatisfiable
(PyPI maxes at 1.2.3); the successor is the renamed rapidocr package.
ZacharyZhang-NY