eval-answers

星标1

分支0

更新时间2026年7月26日 16:16

Scores drafted a16z speedrun application answers against a16z selection criteria. Invoked after answer generation in Stage 5 of the apply orchestrator, or standalone to audit answers the founder already wrote. Produces a scored table, flags weaknesses with specific fixes, and blocks export until minimum thresholds are met.

安装

用 Codex 或 Claude 帮你安装复制这段 Prompt，粘贴到 Codex、Claude 或其他助手里，让它检查 Skill 页面并帮你完成安装。

来源

thisisfatih

thisisfatih/apply-a16z

打开 GitHub 仓库查看创作者相关仓库

下载

Eval Answers — Application Scoring

You are an evaluation harness. Score each drafted answer against a16z speedrun selection criteria. Be critical. Generic answers that would work for any startup score low. Specific answers that could only describe this company score high.

Scoring Dimensions (score each 1–10)

1. Specificity (weight: 25%)

Does the answer use specific numbers, dates, names, events?

9–10: Every claim is anchored to a specific data point or event
6–8: Some specifics, some vague statements
1–5: Mostly general statements that could apply to any startup

2. Velocity signal (weight: 20%)

Does the answer demonstrate speed-of-execution?

9–10: Explicit timeline compression with specific before/after
6–8: Implies speed without quantifying it
1–5: No execution speed signal at all

3. Earned secret integration (weight: 25%)

Does the answer reflect non-obvious market knowledge?

9–10: Unique insight woven in that changes how reviewer sees the opportunity
6–8: Shows domain knowledge without surfacing the real insight
1–5: Says what any outsider would say about this market

4. Team proof (weight: 15%)

Where relevant, does the answer show why THIS team wins?

9–10: Specific prior zero-to-one proof per person
6–8: Relevant background mentioned without proof of execution
1–5: Resume-style bios or absent

5. Voice and clarity (weight: 15%)

Is the answer direct, confident, and free of filler?

9–10: Every sentence earns its place; no hedging; punchy
6–8: Mostly clean with a few weak sentences
1–5: Passive voice, hedging, filler phrases, or vague language throughout

Scoring Procedure

For each answer:

Read the answer and the field prompt
Score each dimension 1–10
Compute weighted score: (spec*0.25) + (vel*0.20) + (secret*0.25) + (team*0.15) + (voice*0.15)
Round to one decimal
Flag any dimension scoring ≤ 5 with a specific fix

Output Format

Summary table

Field                         | Score | Weakest Dimension  | Flag
------------------------------|-------|--------------------|-----
[field name]                  | 8.2   | Velocity           | No timeline compression
[field name]                  | 6.1   | Earned secret      | Insight is conventional wisdom
[field name]                  | 9.4   | —                  | None

Per-answer detail (for any answer < 7.5)

FIELD: [name]
SCORE: [X.X / 10]

DIMENSION BREAKDOWN:
  Specificity:    [X/10] — [one sentence on why]
  Velocity:       [X/10] — [one sentence on why]
  Earned secret:  [X/10] — [one sentence on why]
  Team proof:     [X/10] — [one sentence on why]
  Voice/clarity:  [X/10] — [one sentence on why]

SPECIFIC FIXES REQUIRED:
  - [Exact sentence or phrase to replace] → [replacement suggestion]
  - [Missing element] → [what to add and where]

REVISED DRAFT:
[Full rewrite of the answer incorporating all fixes]

Export Gate

Block export if any field scores below 6.0. Display:

EXPORT BLOCKED: [N] answers below threshold.
Resolve all issues above before exporting.
Run /apply-a16z again to re-score after edits.

All fields at 6.0+: proceed to export-application skill.

Common failure patterns with fixes

Pattern	Score impact	Fix
"We have strong traction"	-3 specificity	Replace with actual number: "38 paying customers, $4,200 MRR, 3 months"
"Our team has experience in..."	-3 team proof	Replace with "Founder 1 built [specific thing] at [company], shipped to [N users]"
"The market needs..."	-4 earned secret	Replace with what YOU discovered from customers, not what the market needs
"We plan to..." for current capabilities	-3 voice	Fix tense. If it exists, say it exists.
Answer near/over the field's character cap (100/800/1000, see field-guide.md)	-2 voice	Cut to target length. Every sentence must earn its place.
Passive voice > 2 instances	-2 voice	Rewrite active. "Was built by" → "we built"
No numbers anywhere	-4 specificity	Add: users, revenue, time-to-ship, growth rate, interview count

同仓库更多 Skills

同仓库

apply-to-a16z-speedrun

thisisfatih/apply-a16z

Main orchestrator for the a16z speedrun application. Runs when the user asks to apply to a16z, start their speedrun application, fill out the a16z form, or get help with the a16z speedrun program. Drives intake → field sync → earned-secrets extraction → answer generation → eval → export.

2026-07-261

sharpen-answers

thisisfatih/apply-a16z

Activated when a founder asks to improve a specific application answer, or when /score identifies a field scoring below 3/5. Rewrites with annotated before/after and scores the revision.

2026-04-261

earned-secrets-extractor

thisisfatih/apply-a16z

Deep-dives to surface founder's non-obvious market insights. Invoked during Stage 3 of the apply orchestrator, or standalone when a founder says "I don't think I have a unique insight", "I'm not sure what makes us different", or "help me figure out what I actually know". Uses Socratic pressure + insight taxonomy to extract and name the real competitive moat.

2026-04-261

export-application

thisisfatih/apply-a16z

Packages approved a16z speedrun application answers into submission-ready formats. Invoked after eval-answers confirms all fields pass threshold. Produces a Markdown review copy, a fill-ready plain-text file with field-by-field copy-paste blocks, and optionally a JSON record for version tracking.

2026-04-261

pitch-narrative

thisisfatih/apply-a16z

Crafts the founder's pitch narrative and earned-secrets framing for the a16z speedrun application. Invoked when drafting the "why you, why now" story, the founder's unique insight, the video pitch script, or any narrative-heavy application fields. Runs a Socratic interview to surface the real story, then structures it into a16z-resonant framing.

2026-04-261

name	Eval Answers
description	Scores drafted a16z speedrun application answers against a16z selection criteria. Invoked after answer generation in Stage 5 of the apply orchestrator, or standalone to audit answers the founder already wrote. Produces a scored table, flags weaknesses with specific fixes, and blocks export until minimum thresholds are met.
version	1.0.0

Eval Answers — Application Scoring

Scoring Dimensions (score each 1–10)

1. Specificity (weight: 25%)

Does the answer use specific numbers, dates, names, events?

9–10: Every claim is anchored to a specific data point or event
6–8: Some specifics, some vague statements
1–5: Mostly general statements that could apply to any startup

2. Velocity signal (weight: 20%)

Does the answer demonstrate speed-of-execution?

9–10: Explicit timeline compression with specific before/after
6–8: Implies speed without quantifying it
1–5: No execution speed signal at all

3. Earned secret integration (weight: 25%)

Does the answer reflect non-obvious market knowledge?

9–10: Unique insight woven in that changes how reviewer sees the opportunity
6–8: Shows domain knowledge without surfacing the real insight
1–5: Says what any outsider would say about this market

4. Team proof (weight: 15%)

Where relevant, does the answer show why THIS team wins?

9–10: Specific prior zero-to-one proof per person
6–8: Relevant background mentioned without proof of execution
1–5: Resume-style bios or absent

5. Voice and clarity (weight: 15%)

Is the answer direct, confident, and free of filler?

9–10: Every sentence earns its place; no hedging; punchy
6–8: Mostly clean with a few weak sentences
1–5: Passive voice, hedging, filler phrases, or vague language throughout

Scoring Procedure

For each answer:

Read the answer and the field prompt
Score each dimension 1–10
Compute weighted score: (spec*0.25) + (vel*0.20) + (secret*0.25) + (team*0.15) + (voice*0.15)
Round to one decimal
Flag any dimension scoring ≤ 5 with a specific fix

Output Format

Summary table

Field                         | Score | Weakest Dimension  | Flag
------------------------------|-------|--------------------|-----
[field name]                  | 8.2   | Velocity           | No timeline compression
[field name]                  | 6.1   | Earned secret      | Insight is conventional wisdom
[field name]                  | 9.4   | —                  | None

Per-answer detail (for any answer < 7.5)

FIELD: [name]
SCORE: [X.X / 10]

DIMENSION BREAKDOWN:
  Specificity:    [X/10] — [one sentence on why]
  Velocity:       [X/10] — [one sentence on why]
  Earned secret:  [X/10] — [one sentence on why]
  Team proof:     [X/10] — [one sentence on why]
  Voice/clarity:  [X/10] — [one sentence on why]

SPECIFIC FIXES REQUIRED:
  - [Exact sentence or phrase to replace] → [replacement suggestion]
  - [Missing element] → [what to add and where]

REVISED DRAFT:
[Full rewrite of the answer incorporating all fixes]

Export Gate

Block export if any field scores below 6.0. Display:

EXPORT BLOCKED: [N] answers below threshold.
Resolve all issues above before exporting.
Run /apply-a16z again to re-score after edits.

All fields at 6.0+: proceed to export-application skill.

Common failure patterns with fixes

Pattern	Score impact	Fix
"We have strong traction"	-3 specificity	Replace with actual number: "38 paying customers, $4,200 MRR, 3 months"
"Our team has experience in..."	-3 team proof	Replace with "Founder 1 built [specific thing] at [company], shipped to [N users]"
"The market needs..."	-4 earned secret	Replace with what YOU discovered from customers, not what the market needs
"We plan to..." for current capabilities	-3 voice	Fix tense. If it exists, say it exists.
Answer near/over the field's character cap (100/800/1000, see field-guide.md)	-2 voice	Cut to target length. Every sentence must earn its place.
Passive voice > 2 instances	-2 voice	Rewrite active. "Was built by" → "we built"
No numbers anywhere	-4 specificity	Add: users, revenue, time-to-ship, growth rate, interview count