A decision system, not a role-play prompt.

The fundraise frame controls the room.

The same deck should not be read identically at pre-seed and Series A. Before the simulation starts, the founder supplies stage, round size, geography, business model, traction context, and lead status. That frame changes which evidence is expected and how pass triggers are interpreted.

• Stage and round size are checked deterministically against the disclosed private-offering record corpus before any language-model analysis.
• Business-model and geography fields select the relevant comparison cell; thin samples are widened and disclosed instead of silently overstated.
• The current public corpus is sourced from disclosed US private-offering filings (operating companies only) and is therefore a US reference. European coverage is kept separate until equivalent sources are licensed or integrated.

The deck becomes structured evidence.

PDF, PowerPoint, OpenDocument, and pasted narratives are parsed at slide level. Google Slides decks can be exported to a supported presentation format first. Text, ordering, missing canonical sections, and—where needed—charts and visual slides become a structured deck representation. This lets later stages distinguish a missing claim from a claim that exists but is weakly supported.

• Slide order and absence detection are deterministic product surfaces, not invented pass reasons.
• Vision processing reads charts, axes, screenshots, and image-heavy slides when text extraction is insufficient.
• Untrusted deck text is fenced as data before it reaches the simulation prompts.

Sixteen decision lenses become a structured cohort.

Archetypes define decision psychologies, not named people. Profiles are generated within those families with different mandates, fund types, focuses, evidence thresholds, and known biases. Equivalent objections are consolidated after evaluation, so repeated support strengthens one concern instead of manufacturing several discoveries.

• Profile-level reactions are retained; equivalent concerns are counted as support for one cluster.
• Archetype weights adapt to the fundraise frame while maintaining coverage across materially different decision styles.
• On multi-model runs, evaluator assignments are balanced inside archetype and cohort-fit strata and kept stable by default. If an assigned family cannot evaluate a batch, those profiles are re-evaluated on families that did answer instead of being dropped.
• Named real investors and firms are prohibited in prompts and scrubbed again at output.

Positions evolve through a conditional decision sequence.

Every profile records a cold read. Investors with a movable position can then react to partner signals before final positions are synthesized. A hard pass is not forced to move for theatrical variety, and later rounds are skipped when the cold read leaves no credible movable group.

• Round 0: each profile records a cold read against the deck and fundraise frame before peer signals are introduced.
• Round 1: movable profiles encounter the signals and objections raised across the room.
• Round 2: final positions, questions, and honest-versus-polite pass language are recorded.
• The report carries rounds-run and collapsed-cohort metadata so a shortened simulation cannot masquerade as a full three-round process.

Synthesis preserves recurrence and dissent.

The deliverable is not a transcript dump. SimVC groups near-identical objections, ranks them by frequency and severity, preserves meaningful differences in how a theme was probed, and links questions—and any recorded slide-level friction—back to the profiles that raised them.

• Ranked pass reasons pair the candid objection with the polite email version.
• The diligence dossier deduplicates repeats without collapsing different questions about the same theme.
• Recurring objections and questions expose their supporting profile count. Model-family breadth is reported from the families that delivered the cold read, never inferred from the configured roster alone.
• The targeting profile shows which investor psychologies moved furthest—never a list of real names.
• The shortest path to yes is selected from grounded cohort evidence, not inserted as generic coaching copy.

Calibration measures objection usefulness—not fundraising destiny.

After real meetings, founders can label whether a simulated objection or question actually appeared and log what the system missed. Optional expert blind screens add another comparison layer. These signals sharpen future runs inside comparable fundraise cells.

• Confirmed, refuted, and missed objections remain separate raw signals.
• Precision and recall are computed by archetype and fundraise cell once enough observations exist.
• The market benchmarks behind every threshold claim are concrete and dated—the share of down rounds this cycle, net revenue retention against the private-SaaS median, ARR-growth medians by stage, Rule of 40, and burn multiple—and each is retired the moment it goes stale rather than cited forever, so a claim is only ever checked against current market data.
• No calibration signal is converted into a probability of successfully raising.

The boundary is part of the product.

SimVC pressure-tests what the deck and its framing can influence. It cannot observe warm introductions, founder reputation, partner chemistry, fund timing, or facts that never enter the materials. Every report states how many model families delivered its cold read. If fewer families answer than the configured method expects, affected profiles are re-evaluated on families that did answer and the report is marked as having reduced cold-read breadth. If no model family delivers the cold read, the run fails and the credit is returned. A strong run is preparation evidence, not an investment decision or a prediction.