Proposals
A proposal is a draft change to the profile. It is inert until accepted, which is the whole reason the queue exists rather than the skill writing rules directly.
A proposal either ADDS a rule, AMENDS one that exists, or DISABLES one. An amend that only describes a change in prose cannot be applied, so an amend carries the changed fields themselves and they are printed below against the current values.
Decided
7 already settled. Read only. The before and after shown against a decided proposal is measured against the profile as it stands TODAY, not as it stood when the verdict was given.
Verification probe: raise the ten-slide guideline from soft to strong.
End-to-end check of the amend path after the accept-only-add gap was closed.
| Field | Now | After accepting |
|---|---|---|
| hardness | soft |
strong |
- Created by an agent to verify the amend path, not a real finding.
Cut hedges, intensifiers and trailing explanatory clauses. Say the thing once and stop.
He repeatedly deletes qualifying words that add no claim, and amputates trailing clauses that explain what the sentence has already implied.
- 841 fully qualified -> 841 qualified (fully deleted)
- Molecule.One and XtalPi are already tracked, scored and inside market synthesis -> Molecule.One and XtalPi are tracked and scored
- let the result set the outreach strategy for the rest of the book -> let results set the strategy
- transfers the saving from dead-end syntheses to you -> transfers the savings to you
- more than any company we track, and nobody automates the other two -> more than any peer; two remain open
Open a line telegraphically. Drop the leading article and any scene-setting words, and start on the noun that carries the claim.
Visible across his edits as a consistent opening move, independent of the compression finding, because he does it even where the line barely gets shorter.
- The lead book and the investor map are built -> Lead book and investor map are built
- Test on one small cohort -> Test one cohort
- Two gaps sit between the book and the first send -> Two gaps sit between the book and first send
- Six materials give the front line what we know, the so what -> Six materials give the front line the what, the so-what
Narrow no-em-dash and no-semicolon from hard universal rules to Chemify-deck rules. In a deck the author says best carries his voice, he ADDED both while editing, repeatedly and deliberately.
Attribution is now settled, so the earlier hedge is gone. In each case he took a model line that already complied with the rule and rewrote it to use the forbidden mark, which is the opposite of an oversight. Two readings remain and both point the same way on scope: either these marks are fine outside a CEO deck, or his taste has moved. The rules are currently hard and universal, so the voice skill enforces them on email, memo and web copy too, and that is very likely over-scoped. This does NOT touch the separate en-dash range fix in MIRA, which is about a different character doing a different job.
- Source: sample 5, vSHARE manual-edits, attested edited, author confirmed he made a lot of the changes himself.
- He introduced the mark where the draft had none, 4 semicolons and 1 em dash added, 0 removed.
- Draft: The lead book and the investor map are built. The next moves are a CRM integration, a European contact source, and an agreed filter
- His edit: Lead book and investor map are built; CRM integration, a European source, and a filter come next
- Draft: Chemify automates five of the seven reaction modes, more than any company we track, and nobody automates the other two
- His edit: Chemify automates five of seven reaction modes - more than any peer; two remain open (an em dash AND a semicolon, replacing two commas)
- Draft: 6,351 screened companies narrow to 841 fully qualified, and 1,150 of the small-molecule base can be emailed today
- His edit: 6,351 screened companies narrow to 841 qualified; 1,150 can be emailed today
- Countervailing: the CI deck v13 to v21 pass removed 18 semicolons, but that pass was agent-led deck building, so it may be an agent applying the declared rule rather than independent evidence.
Decide whether no-semicolon is a Chemify-deck rule or a universal one. Two edit pairs disagree, and the disagreement is systematic rather than noise.
Surfaced, not resolved, because the two readings imply opposite changes and the evidence cannot separate them. Reading one: the rule is genuinely universal and the CEO-facing deck shows it working. Reading two: it is a Chemify-deck rule and the internal share deck is a register where you are comfortable with a semicolon. There is also an attribution problem I cannot settle from the files, and it matters: I do not know whether you typed the manual-edits pass or whether I produced it from your instructions. If the latter, the four added semicolons are my prose, not yours, and this proposal should be rejected outright.
- Against the rule, pair A (vSHARE to manual-edits): 4 prose semicolons ADDED, 0 removed.
- Span: Lead book and investor map are built; CRM integration, a European source, and a filter come next
- Span: 6,351 screened companies narrow to 841 qualified; 1,150 can be emailed today
- Span: Molecule.One and XtalPi are tracked and scored; the gap is signal depth, not coverage
- Span: Chemify automates five of seven reaction modes - more than any peer; two remain open (this one adds an em dash as well)
- For the rule, pair B (CI deck v13 to v21): 18 semicolons REMOVED, 2 added, and 1 em dash removed.
- Confound on pair B: those versions were produced during agent-led deck building, so the removals may be an agent applying your declared rule rather than you revealing a preference. That would make pair B circular and not independent support.
On revision, cut about a fifth of the words. Across two independent draft-to-final pairs the edit made the line shorter far more often than longer, and the net word count fell every time.
This is the only finding that replicated across both usable edit pairs, and it is the one least sensitive to who held the pen, because the decision to compress is an editorial judgement either way. It is not currently a rule. The nearest seed rules govern deck length and parenthetical detail, neither of which says what to do when revising an existing line.
- Pair A, vSHARE generated to manual-edits, 18 rewordings: 14 shorter, 3 longer, 1 same. 348 words in, 274 out, down 21 percent.
- Pair B, CI deck v13 to v21, 51 rewordings: 25 shorter, 16 longer, 10 same. 600 words in, 557 out, down 7 percent.
- Span: The lead book and the investor map are built. The next moves are a CRM integration, a European contact source, and an agreed filter -> Lead book and investor map are built; CRM integration, a European source, and a filter come next
- Span: Test on one small cohort, change one lever per wave, and let the result set the outreach strategy for the rest of the book -> Test one cohort, change one lever per wave, and let results set the strategy
- Span: Fortified the investor map 702 firms across 1,702 cite-verified financing edges, led by RA Capital at 121 companies -> Fortified the investor map with 702 firms across 1,702 financing rounds
- Caveat: a third pair (prior to shared) produced only 3 rewordings against 402 additions, so it was a wholesale rebuild and carries no style signal. It is excluded rather than counted.
Reconsider the no-colon-separator rule, or scope it to Chemify decks only. In the one deck you have attested as your own writing it is your most frequent sentence pattern, not an occasional slip.
This is a declared-versus-revealed conflict and it is surfaced rather than resolved. The rule says avoid the X colon Y separator. Your attested MediSure deck uses it 81 times in 2619 words, and the spans are genuine prose rather than field labels. Two readings fit: the declared rule is aspirational, or it is genuinely a Chemify-deck rule and coursework is a different register. Only you can say which. Nothing changes until you accept or reject this.
| Field | Now | After accepting |
|---|---|---|
| hardness | soft |
soft |
- Attested sample 1, MediSure Regulatory Intelligence Workflow v3, authorship authored, 2619 words.
- Measured: 81 colon-separator matches, against 0 em dashes and 0 prose semicolons in the same text, so the other two declared mechanics rules ARE corroborated here.
- Span: Regulatory monitoring is manual and fragmented: analysts track the FDA, EMA, EU Commission, CDSCO, and WHO
- Span: The process is slow and inconsistent: analysts read lengthy unstructured notices
- Span: Manual monitoring raises compliance risk: missed changes to approvals, labeling, or post-market obligations
- Span: A lightweight no-code solution is required: search sources, extract and synthesize updates
- Counter-consideration: some matches are field labels such as Goal, Outcome and Approach, which are a document convention rather than the prose pattern the rule targets.
- Confound: MediSure is an AI-course submission, not a CEO-facing Chemify deck. The declared rule came from Chemify deck corrections, so genre may explain the gap.