 Jesse Vincent
|
35464d67c0
E27 stack: conditional impl tier + final-review tier pin + narration recipe + terse reviewer contract
|
2 месяцев назад |
|
Jesse Vincent
|
90b5433f59
E03: cheapest-tier implementers when plan carries complete code (transcription hypothesis)
|
2 месяцев назад |
 Jesse Vincent
|
420c234a2c
Bump evals submodule: E29-E34 quality investigation + L2 gate results (af05326)
|
2 месяцев назад |
|
Jesse Vincent
|
d1fcc9889a
Strict-cost spec: L2 final — died at gates; explicit escalation holds at sonnet, implicit adjudication does not
|
2 месяцев назад |
|
Jesse Vincent
|
ac11700642
Bump evals submodule: L1 elicitation + autoresearch scenarios and logs (649b1f8)
|
2 месяцев назад |
|
Jesse Vincent
|
710f031ad0
writing-plans: task right-sizing, Global Constraints header, per-task Interfaces blocks
|
2 месяцев назад |
|
Jesse Vincent
|
72cb21b82c
Constraints block is the reviewer's attention lens: copy spec verbatim, never improvise process rules
|
2 месяцев назад |
|
Jesse Vincent
|
de1d35e5e7
Strict-cost spec: L1 final — cost win re-attributed to complete-code plans; guidance owns fidelity/variance
|
2 месяцев назад |
|
Jesse Vincent
|
ec014e7a7f
Bump evals submodule to merged superpowers-evals main (ac264b1)
|
2 месяцев назад |
|
Jesse Vincent
|
eba16f6b91
Strict-cost spec: L2 recon n=2 (sonnet controller $6.68/$8.05, judgment clean, escalation points unstressed)
|
2 месяцев назад |
|
Jesse Vincent
|
27788fdef9
Strict-cost spec: record batch A-E rung verdicts (L1 validated, L2 recon positive, L3 dead)
|
2 месяцев назад |
|
Jesse Vincent
|
9a25a75bac
Spec: strict-cost SDD experiment ladder — judgment as co-invariant, plan-side crispness first
|
2 месяцев назад |
|
Jesse Vincent
|
60fa4f6fc4
Record writing-plans micro-test result: resolved, no change needed
|
2 месяцев назад |
|
Jesse Vincent
|
43a6ee23f7
Spec: record iterations 4-5 (variance honesty, structural fixes, final validated ranges)
|
2 месяцев назад |
|
Jesse Vincent
|
fe90d6c469
Adopt audited positive phrasings: evidence rule leads positive; fix-report completeness as checklist
|
2 месяцев назад |
|
Jesse Vincent
|
b81f35bb1e
Land eval-tuned combo: file handoffs, progress ledger, final-review package, REQUIRED model lines, reviewer risk budget
|
2 месяцев назад |
|
Jesse Vincent
|
926096a1d7
Spec: positive-instruction redesign — audit results, micro-test method, writing-plans variants
|
2 месяцев назад |
|
Jesse Vincent
|
a995af2e24
Shared: unique review-package collateral names
|
2 месяцев назад |
|
Jesse Vincent
|
d4dbf44162
Add review-package script; close fix-dispatch test gap
|
2 месяцев назад |
|
Jesse Vincent
|
2434ef7f35
Describe the review design as current state, not as a delta
|
2 месяцев назад |
|
Jesse Vincent
|
7cf78437e2
Spec: record iterations 2-3 results and final frozen-config matrix
|
2 месяцев назад |
|
Jesse Vincent
|
e355795625
Hand reviewers the diff as a file, not a paste
|
2 месяцев назад |
|
Jesse Vincent
|
29ee4e8e44
Reviewer skepticism covers the implementer's design rationales
|
2 месяцев назад |
|
Jesse Vincent
|
28498a5cde
Make diff-pasting non-optional for task reviewer dispatch
|
2 месяцев назад |
|
Jesse Vincent
|
5e2907fc4f
Close the Minor-severity escape hatch
|
2 месяцев назад |
|
Jesse Vincent
|
e532f24df7
Spec: document cost iterations and the per-task review consolidation
|
2 месяцев назад |
|
Jesse Vincent
|
e3c74fc1c9
Merge per-task reviews into one task reviewer (iteration 2)
|
2 месяцев назад |
|
Jesse Vincent
|
3e3e1e701e
Cut review-cost drivers: turn-aware models, inline diffs, scoped evidence
|
2 месяцев назад |
|
Jesse Vincent
|
853396e3ae
Add phrase-level pre-judging triggers to reviewer prompt rule
|
2 месяцев назад |
|
Jesse Vincent
|
83d54f7ddd
Red Flags: never tell a reviewer what not to flag or pre-rate severity
|
2 месяцев назад |
