Spaces:
Running
Running
Commit History
-
feat(library): settle the last two wording questions; CAMEL logo
Both decisions confirmed with the PI on 2026-09-07, so no construct in the library is left without a verification decision. IPIP keeps its leading "I". Worth recording why the question dissolved: the platform never added it. import_from_xlsx.py only strips whitespace, and the items arrived with the "I" already present in the commit that first created the library, from lib.xlsx. The IPIP source really does print bare stems ("Feel little concern for others") with the subject supplied by its instruction header, so the reviewer's observation was right about the state and wrong about the cause: this is a transcription convention in the lab's own spreadsheet. The five constructs are verified with that provenance recorded on each. K10 is retired rather than rewritten. Its items are fragments under a shared header, and CCR embeds item text directly, so "worthless?" is a poor stand-in for the construct. Archived, not deleted: constructs are append-only and any run that used it must still reso…726d852
-
fix(constructs): resolve the review's open questions from Noor's answers
She replied with the source she used for each disputed item, and every one checks out. Three of the four disagreements were about which PRINTING of a scale is authoritative, not about who read it correctly. MFQ Fairness: I was wrong to revert her. She read the questionnaire printed at the end of Graham et al. (2011), which is the paper this construct cites, and it does say "treated differently from others" (twice). The moralfoundations.org MFQ30 handout says "than others". Her correction is applied and source_url now points at the paper, so items, citation, and source finally agree. RSES and CBI work-related burnout: she used the source_url recorded on each construct, and both mappings are exactly right against it. The Rosenberg questionnaire at socy.umd.edu starts with "I feel that I'm a person of worth" and its wording and reverse keys match ours item for item, so only our ORDER was wrong. Both are reordered to the printed order with item ids renumbered, so item_id means "source it…0af19f8 -
fix(constructs): verify MFQ against the source the review sheet omitted
Checking whether the follow-up questions for the reviewer were fair turned up that two of them were not. The review sheet shipped with an EMPTY source_url for mfq_care and mfq_fairness, so the reviewer had nothing to check those against and correctly flagged it. That was our omission, not a missing document: the MFQ30 is public, all four items in each construct match it verbatim, and both are now verified with the source recorded. This also settles the "differently than/from others" disagreement in favour of the wording already stored. Also confirmed independently, so the reviewer does not need to be asked: the Subjective Happiness Scale PDF at the source_url we gave her does read "Compared to most of my peers", exactly as she corrected it. And for context on the two item-order findings: the APA measures-package printing of the RSES numbers its items exactly as we store them, while the reviewer's renumbering matches another widely-used printing. Same version-mismatch pattern as Edmo…10425ec -
fix(constructs): correct the review pass after a validity review
A research-validity review of the previous commit found four real problems. All four are fixed here. Two of the reviewer's wording corrections moved an item AWAY from the publication its construct cites, and I had applied them under a `verified` badge. Checked against the sources: the published MFQ30 reads "treated differently than others" (not "from"), and Edmondson (1999) reads "It is safe to take a risk on this team" (not "in") - the proposed "in this team" comes from a third-party questionnaire, whose URL I had also promoted over the working Edmondson PDF. Both items keep their published wording, both constructs are held back as unverified with the disagreement recorded, and the source_url override is dropped. The reviewer's item-ORDER findings for rses and cbi_work_related_burnout were silently dropped: my classifier only read notes on rows that failed one of the four checks, and these sat on rows marked OK. Order feeds item_hash and decides what sim_item_N means, so both are no…461b859 -
chore(registry): record measured CPU throughput for all seven models
Every model said expected_time_per_1k_texts_cpu: benchmark_required, not just the four PsyEmbedding ones. Measured with scripts/bench_models.py --threads 2 --quick. Throughput varies about 50x between a 15-word social post and a 250-word document, so the field is three numbers rather than one, and the comment names the host and date: a shared cloud vCPU runs roughly 2-4x slower than the machine these came from, so they need a re-run on the Space before they are shown to users. Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>684411f -
feat(constructs): apply the 2026-08-25 library review [spec 0007]
Noor Skhiri reviewed all 525 items across the 94 constructs (wording, reverse keying, subscale grouping, citations). This applies it. Split by what item_hash covers. Item changes ship as a NEW version: 61 reverse-scoring flags corrected across 14 constructs (35 -> 96 flagged items), 9 verbatim wording fixes, and one SCS-SF item moved to Over-identification per the source coding key. 23 constructs gain a version 2; their version 1 files keep their original items and are marked archived so runs that pinned them still resolve. Everything outside the hash is edited in place: 88 constructs promoted to verified with reviewer and date recorded, plus citation and source_url repairs. Six constructs stay needs_verification pending two PI decisions (the IPIP "I" prefix and the K10 stem). Two loader gaps had to close for this to ship: sync_library was insert-only, so a status change could never reach an already-seeded database, and the picker listed every version, so a v2 would show the same co…4afd259
-
fix(models): drop the PsyEmbedding pooling workaround, re-pin to fixed repos
The four lab PsyEmbedding repos shipped a modules.json referencing a 1_Pooling config the repos did not contain, so SentenceTransformer(id) could not load them and the registry carried a pooling_fallback flag telling the backend to assemble Transformer+mean-Pooling by hand. The missing config was pushed upstream on 2026-08-22, so the workaround is no longer needed. - models.yaml: re-pin all four models to the fixed revision (one commit above the old pin) and remove pooling_fallback. - registry test inverted: psyembedding models must now report pooling_fallback is None, so the workaround cannot creep back in. No code change is required. Both branch points (ccr.py model loading and reproducibility.py script generation) key off pooling_fallback being truthy, so new runs take the plain one-line loader automatically, while older runs keep the explicit module assembly recorded in their own metadata. Verified against the Hub: all four pinned revisions exist and carry 1_Pooling/config.j…597c9b7
-
Pin all registry models to fixed HF commit hashes
Reproducibility: an unpinned model silently changes if the provider re-pushes its revision, shifting scores between runs. Pin all 7 models (the four PsyEmbedding fine-tunes plus MiniLM, E5-large-v2, and multilingual-E5) to their current commit SHAs, clearing the last PIN_ME entries and satisfying the Phase-1 exit criterion. Behavior-preserving: a pin freezes the current revision, it does not change scores.767c15e
-
new upload restrictions
a9d817a
-
Apply PI decisions: full construct library (rights approved), anon cap 200, drop CRS
PI approved including questionnaire items in public repos; library now ships all 94 constructs (CRS removed per PI). Anonymous row cap 200.4ce78b4
-
Sync platform from lab mainline: dev instance ready to deploy
Full feature set: accounts (email/password + feature-flagged Google via Supabase PKCE), anonymous tiers (3 runs/day, delete-after-analysis, 24h purge), saved-run cap, construct upload from CSV/XLSX, structured warnings + language detection, model registry (MiniLM/E5), per-run reproduction scripts, corpus-embedding cache, storage interface (local/ R2), sample-data test kit, MANUAL_TESTING.md. Rights guard: only the 5 original seed constructs ship in this public repo; the imported lab collection stays out pending the redistribution decision. HF remote URL detokenized. 64 tests passing.4b09d2d