Full team
All eight domains; reference contribution schedule.
Earlier fixed workflow · Preserved study records
The preserved earlier fixed-workflow pilot: separate rubric indices, browser coverage, missing evidence and retained Build prototypes. These results do not evaluate the revised dynamic-conversation protocol. No human outcomes were measured.
The study makes expertise assignments, shared understanding, and work allocation inspectable across the LENS capability cycle. The flower organizes contributing domains; human flourishing support, system performance, and learning remain distinct evidence questions.
Five configurations addressed the same fictional challenge: helping new team leads practise a difficult feedback conversation. Three repetitions were planned per configuration. The results describe generated learning solutions and repeated model judgments; no human outcomes were measured.
Descriptive comparison · Fixed condition order
These 0–100 rubric indices summarize 0–4 criterion ratings. They are not percentages of learners, probabilities, or measured human outcomes. Each candidate index requires both reviewers to rate all three criteria on that axis. Unknown values remain unknown.
14 of 15 planned candidates are included in the review set. The batch reports completion for all generation runs, but 1 candidate has no verified terminal record and is excluded from evaluation. Its card below retains the reason; it receives no rating or evaluated preview.
| Condition | Human flourishing support | System performance | Learning |
|---|---|---|---|
| Full team3 planned candidates | 62.5SD 5.9 · 2/3 known1 unknown rating · 0 excluded from review | 58.3SD 5.9 · 2/3 known1 unknown rating · 0 excluded from review | 72.9SD 14.7 · 2/3 known1 unknown rating · 0 excluded from review |
| Domain omitted everywhere3 planned candidates | 70.8SD 5.9 · 2/3 known1 unknown rating · 0 excluded from review | 64.6SD 2.9 · 2/3 known1 unknown rating · 0 excluded from review | 77.1SD 8.8 · 2/3 known1 unknown rating · 0 excluded from review |
| Domain withdrawn from one member3 planned candidates | 70.8SD 5.9 · 2/3 known0 unknown rating · 1 excluded from review | 56.3SD 2.9 · 2/3 known0 unknown rating · 1 excluded from review | 62.5SD 5.9 · 2/3 known0 unknown rating · 1 excluded from review |
| Member removed3 planned candidates | 66.7SD 4.2 · 3/3 known | 59.7SD 4.8 · 3/3 known | 69.4SD 10.5 · 3/3 known |
| One agent with matched slots3 planned candidates | 73.6SD 2.4 · 3/3 known | 59.7SD 2.4 · 3/3 known | 77.8SD 10.5 · 3/3 known |
The table uses candidate means from the frozen aggregate without imputation. Standard deviation describes the spread of known candidate indices; it is unavailable with fewer than two known values. Repeated reviews and criteria are not additional generated candidates. There is no combined score or overall rank.

Each review rates nine criteria, three on each axis. An axis index is twenty-five times the mean of its three 0–4 ratings. The displayed candidate value is the mean of two complete reviewer indices. A missing criterion makes that reviewer’s axis unknown; a missing reviewer index makes the candidate axis unknown.
Two fresh calls to the same model (gpt-6.1-sol, reasoning effort low) produced the reviewer pair. These are repeated model judgments, not independent human raters. Failed review attempts remain in the record as unknown values. Exact evidence quotations support traceability, not the correctness of a judgment.
All eight domains; reference contribution schedule.
Assessment instructions omitted across the team; all contribution slots retained.
Assessment instructions withdrawn from the primary member; a colleague retains the domain.
Primary assessment member removed, reducing the schedule from forty to thirty-six calls.
One identity works through the reference contribution slots and their assigned duties.
Assessment, Measurement, Evaluation is the domain manipulated here. The primary member’s colleague retains an explicit overlap in the reference and single-member-withdrawal conditions. Shared records and duties to expose assumptions and disagreements apply across all conditions. Omission removes instructions, not a model’s latent knowledge; matching contribution slots does not match token volume or human effort.
Observed behavior · Separate from model ratings
The fixed protocol checks initial rendering at three viewport sizes and exercises the same five desktop flows where the candidate provides the relevant controls. Each record is bound to the exact Build source, completed journal head, frozen harness, and executed interaction protocol.
Automated accessibility: 0 violation-rule occurrences and 0 incomplete-rule occurrences. 0 viewports have unknown automated results. 0 runtime events were dropped by capture limits. 0 viewports have unknown overflow status.
| Common check | Recorded states |
|---|---|
| F1 · Initial practice and feedback | observed: 14 |
| F2 · Revision and feedback | observed: 14 |
| F3 · Changed situation | observed: 14 |
| F4 · Skip | observed: 14 |
| F5 · Clear or reset | observed: 14 |
“Observed” means the protocol executed and retained evidence. It is not a correctness score. A literal feedback change does not establish useful feedback; unchanged wording is interpreted with the actual inputs and the resource’s stated purpose. Missing controls, failed actions, and unexecuted steps remain explicit in each candidate’s coverage.
The checks do not establish human learning, transfer, five-minute feasibility, a complete assistive-technology audit, or browser-wide deletion. Generic and functional reports are retained separately; their repeated initial checks are not counted as extra candidates.
All planned candidates · Grouped by repetition
Every available preview uses the original retained Build source in an isolated frame with external resources and API requests blocked. Later generated plans are part of the review packet; proposed repairs do not alter these prototypes.
Repetition 1 · Included in review
Original batch reports: Complete.
Build record: fictional feedback-conversation practice candidate
Functional record: bound.
Generic record: bound · original watcher checks.
Evaluation record: Exact byte copy of the completed primary journal.
Exact HTML SHA-256
81e2e913d75f8d190a0beb87f13454d6f2623645777fbea0bb13383ad62da6a6Build content hash
e549f4403aa96735f9c385b49860f6e9b057e50c55e8b69e2c75aab7627fd2a5Evaluation-provenance file SHA-256
d4f51788f50a6b472cf9a737225133d57cc47e12688d6dfeceda0e8acfeb6f2aFrozen evidence packet
fac01e246bdefe499e29b384b61bf9fb76d7544d64bab82916afd19c5eefb4cdReview states: complete; complete
Repetition 1 · Included in review
Original batch reports: Complete.
Build record: fictional feedback-conversation practice candidate
Functional record: bound.
Generic record: bound · original watcher checks.
Evaluation record: Exact byte copy of the completed primary journal.
Exact HTML SHA-256
457cc0411c96af07f27d609e82d696e9bc5a9b6821b952c5e624aeb6bac6c3b7Build content hash
0631c580d8483530ad09b1dfc27bce780e7b71ad594c0601f0be161442d7e980Evaluation-provenance file SHA-256
abcbc9445a5fcac6326e7fc02ee705ed9b7cdd607b7f032cf45228a82677b328Frozen evidence packet
6be3547b6cb6c1046bd5948d8781eba152bbf5b526caf2dac316c5ff691f7b8dReview states: complete; complete
Repetition 1 · Included in review
Original batch reports: Complete.
Build candidate: Supporting difficult feedback conversations
Functional record: bound.
Generic record: bound · original watcher checks.
Evaluation record: Exact byte copy of the completed primary journal.
Exact HTML SHA-256
813aa90fbebc24769f0028a3ad7072da819c8f9494bad1a91d7502c9b234c7b0Build content hash
6ba7dd4f38bfa230c625176e1fd65f64db3f040f5d2f6c68e5e88313b149067bEvaluation-provenance file SHA-256
60b57eaeb0f5f40075c0ed188207756a93de23595c16987bfbb31a8a34df86eaFrozen evidence packet
57a33555be842cfa40b7c0ca2a99fd5caa562fd86211522602e0dddb3d8c28efReview states: complete; complete
Repetition 1 · Included in review
Original batch reports: Complete.
Build record: fictional feedback practice candidate
Functional record: bound.
Generic record: bound · original watcher checks.
Evaluation record: Exact byte copy of the completed primary journal.
Exact HTML SHA-256
9c93f6aeb6601fc726bb6175b288614b2dc1cfcbb0b27ae8e5da782b111b3e34Build content hash
7feaf92295b109185484086a73d358c0ea7ecf8c654ad13bd1285c3811c55372Evaluation-provenance file SHA-256
38c4e79cdee6cb1037070c43a64fbf00b9ad33e6ef117ada6111db0a3e50e3f9Frozen evidence packet
5cd8d86cf4553dd6d86d4c7a54a9b5e2ddbd6ed9b2811435e008b47689675822Review states: complete; complete
Repetition 1 · Included in review
Original batch reports: Complete.
Build record — fictional feedback-conversation practice candidate
Functional record: bound.
Generic record: bound · original watcher checks.
Evaluation record: Exact byte copy of the completed primary journal.
Exact HTML SHA-256
b54f28970ab1434f99e23f87883d41e85e8bdb6f7b0af2d7301b2ccc374baac4Build content hash
4bef6dcf7973c97597a60d414d315d2c4c626cda85139db2fe5437afc4595cd1Evaluation-provenance file SHA-256
e13f5bca32f2a2818717289fa45c725c907ce25bc36b5c722a7b65282692dd45Frozen evidence packet
1a9f6ff6d557f841e912ed144f9fd7cb694cf2e825ff9fa9691a2a4d85203510Review states: complete; complete
Repetition 2 · Included in review
Original batch reports: Complete.
Build record: fictional feedback practice candidate
Functional record: bound.
Generic record: bound · original watcher checks.
Evaluation record: Exact byte copy of the completed primary journal.
Exact HTML SHA-256
964100a9d6b32d6dc334f9902b11961ea1f0ca536fb2c9fa9837dfed1b1f6b81Build content hash
3d6409b456b9b4b5cf1c6049ca4d4342366e16fd366057cdacb6c8ddbea3ad53Evaluation-provenance file SHA-256
32a9eb22139bc5dc2fb4b59a1e185ff68394df78c9b54f7644883e409361a602Frozen evidence packet
0a35edfb6a0e6408fbb7dae182db9bc9fd5f53009280760327ccf1b1967fb189Review states: failed_attempt_unknown; complete
Repetition 2 · Included in review
Original batch reports: Complete.
Build record: fictional feedback conversation practice candidate
Functional record: bound.
Generic record: bound · supplemental evaluation-view checks.
Evaluation record: Separately documented view of the complete recorded event-log extension.
Exact HTML SHA-256
1d62e6b5cff70db4fb856e2b24fd49abac12b2688a7b914f84d58a359583dde7Build content hash
4e97b5b080f6d7c6cd6a8709fdcbe0efcb3005eb37382f54e7188ddf586dd8deEvaluation-provenance file SHA-256
0b00e3a4b3f4d4e1de6de2cdf06c09631f7f824add65453f46609f7f7adbbe6dFrozen evidence packet
77ae10a0ae37878cfe882d35177ad7b2c05a2f53f7c6d21549b793c14b275b87Review states: complete; failed_attempt_unknown
Repetition 2 · Included in review
Original batch reports: Complete.
Build record: fictional feedback-conversation practice candidate
Functional record: bound.
Generic record: bound · original watcher checks.
Evaluation record: Exact byte copy of the completed primary journal.
Exact HTML SHA-256
b84d8571a5b1df96ef720c9b522823979302528e56af4b4d67e0fb167bbb408eBuild content hash
42576b50e7eedde81c30beebfda8a60d1a3ab82755c9470c6030e103fe8fbe83Evaluation-provenance file SHA-256
8fd624741135b60245c2b335d70c24bd34f60e66437e8b8e80d50678f952992aFrozen evidence packet
93ba918303ac23c1bd01a9e532702e11c86341231572bde8eefeb16b7278badeReview states: complete; complete
Repetition 2 · Included in review
Original batch reports: Complete.
Build record: fictional feedback-conversation practice candidate
Functional record: bound.
Generic record: bound · original watcher checks.
Evaluation record: Exact byte copy of the completed primary journal.
Exact HTML SHA-256
6a47a14e4ce3bc5e55e7b22a6a71b45c84c0a886212d91c136947743a276e2afBuild content hash
386b6b38d2977b771be95ce0dadb58fd9abc8239b649fe13b4ca8b6c7cada8e1Evaluation-provenance file SHA-256
a0a2e3e40b514820a1537aa8ef46c1dea3502aa1d2b910e08f371a60228dbfc2Frozen evidence packet
76599dbb5739d4b5f4b599d60eb34fc88fbe0cf8611e6958f6f4d9b63d0da561Review states: complete; complete
Repetition 2 · Included in review
Original batch reports: Complete.
Build record: fictional feedback practice candidate
Functional record: bound.
Generic record: bound · original watcher checks.
Evaluation record: Exact byte copy of the completed primary journal.
Exact HTML SHA-256
84db24df85eb9ea94e75991de23cfb3735eb4b796bdadd39d189c5d4742c4598Build content hash
e897b9d2af950f437dc28f5f37f4ccb12e82520def5047755c1496fe95d09428Evaluation-provenance file SHA-256
705d681e0d4fc4641da7cad6cd4b95e50e11f796a0f0be2f2734c33eee08cf63Frozen evidence packet
f0c858acf132e9aa2312578c904f87a92a4ce2fe90faf30423be95b3cfea5c65Review states: complete; complete
Repetition 3 · Included in review
Original batch reports: Complete.
Build record: fictional feedback-conversation practice candidate
Functional record: bound.
Generic record: bound · original watcher checks.
Evaluation record: Exact byte copy of the completed primary journal.
Exact HTML SHA-256
00b4f310b4b5312be7b62405a56df11d9a94b5ca89c1f15171c8985f125c222aBuild content hash
fa0a240e16dcf93744ce576d370a93e171973d9d834ae6cd7643a9814d0e04d8Evaluation-provenance file SHA-256
2340c515613e9925afd1831c13fe88f7a0ef94d74d9cee929e5b8f6fb108ac39Frozen evidence packet
9496c5e9a69fc70473f919bdf3162f466147ea5c5c335bedb32537215bb219beReview states: complete; complete
Repetition 3 · Included in review
Original batch reports: Complete.
Build record: fictional handoff practice candidate
Functional record: bound.
Generic record: bound · original watcher checks.
Evaluation record: Exact byte copy of the completed primary journal.
Exact HTML SHA-256
805950764ca674aec82c9ec39662fb1f067aa98f00c40fb51890cc793b8244c5Build content hash
f298e5a95aebdb95e4446b62aff9457576db246c5edfcae54a0a05385cf35b80Evaluation-provenance file SHA-256
3761b5c10b0cfd0042ca98483ff4d5f06c744f8ca39391943424dc17e73f786eFrozen evidence packet
ab2df8f71bb9724b7760a1250bb6222c5ff192524808b3f10992989fe55669fdReview states: complete; complete
Repetition 3 · Excluded from review
Original batch reports: Complete.
No evaluated Build preview
Review unavailable: Raw JSONL extension is not completed at the exact batch head
Functional record: excluded_from_evaluation.
Generic record: excluded_from_evaluation.
Repetition 3 · Included in review
Original batch reports: Complete.
Build record: fictional feedback conversation practice candidate
Functional record: bound.
Generic record: bound · original watcher checks.
Evaluation record: Exact byte copy of the completed primary journal.
Exact HTML SHA-256
54fe55995ac854a9240a4b529170d48956ca16d246618676feba7196f432ff8eBuild content hash
f7e3fe4b3b01195b78422f5d2c14a085a2a9a93e8384757b543cd9057324eac4Evaluation-provenance file SHA-256
03e8c24e1381ae2db9ab207bd8e84e53c2e1d54ad936b3e1f7b36f1ab7be4a07Frozen evidence packet
338b7e28283a46eec258fdee423ded200cde34ff919a349ff84b327e33c612a5Review states: complete; complete
Repetition 3 · Included in review
Original batch reports: Complete.
Build record: fictional feedback-conversation practice candidate
Functional record: bound.
Generic record: bound · original watcher checks.
Evaluation record: Exact byte copy of the completed primary journal.
Exact HTML SHA-256
5638eb684eb07cfab133e289fb97e5c1c85aeb090b389d3af740d9fccd98a545Build content hash
dadd23aaa08faef02dda0045eb420dedb33de20a80227075a78f803544729519Evaluation-provenance file SHA-256
815a90c167b1f2ffefa69c33ad7e8a5996143aa0989031153aab08a5a5443978Frozen evidence packet
e5aa758feb443a0ac7a0d32a131e916ad07a9cd7438ebaf3d0d61008a3d735e5Review states: complete; complete
The evidence concerns one fictional scenario, the configured model and instructions, and a small set of generated candidates. Descriptive differences can inform further investigation; they do not establish causal effects in human teams or a population-level advantage.
The flower’s eight domains organize starting responsibilities around shared understanding. They are not an exhaustive taxonomy or a validated measurement of anyone’s expertise. Human flourishing support, system performance, and learning remain distinct views of the available design evidence.
The earlier paper is preserved in the offline historical appendix. Read the newer paper about dynamic conversations and scenario tests.
Primary journal hashes, instruction fidelity, receipt coverage, and raw-log consistency are separate checks. A verified source fingerprint does not resolve missing provider receipts or a disagreement between record copies.
Evaluation views derived from a strictly verified complete event-log extension: 1. These separately labeled views preserve the incomplete primary snapshot’s metadata and the full recorded event chain, match the batch head, and introduce no event or response. Original files and generated HTML remain unchanged.
Read the final audit counts and retained gaps
Aggregate summary
b968b0227c75cbe8bcde0b686ffb0295731b33f4aa49270f7f7c06888fb94c24Review freeze
6ee9d57e50b636d0c3a63795a4f214ec6880bf949a5d78e4fab9657126a7c5b9Generation audit source
b476955ad4c5300cd5bee415e215600a798577bbd13cd6783b53b17cd714b4e5Functional protocol
3c48309a11fad67377e3540ffc913b19e476f4a7c8af6b000c9c184cb10f956eEvaluation-view manifest
9787faf44df95e95ded7b8a27a55fc7bf9388768918fc5f64d23751ee6f28f16