Class: Insika::Harvest::Miner

Inherits:
Object
  • Object
show all
Defined in:
lib/insika/harvest.rb

Overview

The model's raw skills, filtered into data the command then filters (negative list, grounding, dedup — C6) and a human then gates. Pure over an injected ask — the Refinement::Proposer shape.

Defined Under Namespace

Classes: Unusable

Constant Summary collapse

MAX_NAME = 
64
MAX_DESCRIPTION = 
300
MAX_BODY = 
6000
MAX_TRIGGERS = 
10
MAX_TRIGGER = 
80
MAX_RATIONALE = 
500
MAX_EVIDENCE_TURNS = 
20
MAX_SESSIONS =

The window cap the RUN applies (C6): a candidate whose origin sessions are a lie of provenance must never be stamped from a window of 50.

5
ALLOWED_KEYS = 
%w[name description body triggers rationale evidence_turns].freeze
DROP_KEYS = 
%w[schema unknown_key oversized bad_turns duplicate capped].freeze

Instance Attribute Summary collapse

Instance Method Summary collapse

Constructor Details

#initialize(ask:, model: "utility_model") ⇒ Miner

Returns a new instance of Miner.



97
98
99
100
 
# File 'lib/insika/harvest.rb', line 97

def initialize(ask:, model: "utility_model")
  @ask = ask
  @model = model.to_s
end

Instance Attribute Details

#model ⇒ Object (readonly)

ask: ->(prompt) { "" } | something answering #content (+ #input_tokens/#output_tokens/#cached_tokens for cost). model: the ref recorded as the candidate's proposer ("utility_model" default).



95
96
97
 
# File 'lib/insika/harvest.rb', line 95

def model
  @model
end

Instance Method Details

#mine(prompt:, message_counts:, max_proposals: 10) ⇒ Object

-> { skills: [ raw candidate hashes ], dropped: { "schema" => N, "unknown_key" => N, "oversized" => N, "bad_turns" => N, "duplicate" => N, "capped" => N }, cost: { "spent" => N, "cached" => N } | nil } prompt: the pack prompt or DEFAULT_PROMPT (the caller resolved it). message_counts: the origin sessions' transcript sizes, in prompt order, so evidence_turns indexes are validated against the sessions they name (an index is valid if it fits at least one session). max_proposals: cap on surviving raw skills. Drops counted, never fixed.



111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
 
# File 'lib/insika/harvest.rb', line 111

def mine(prompt:, message_counts:, max_proposals: 10)
  answer = @ask.call(prompt)
  raw = parse(text_of(answer))
  skills = []
  dropped = DROP_KEYS.to_h { |k| [k, 0] }
  seen = {}
  raw.each do |item|
    verdict, reason = classify(item, message_counts)
    case verdict
    when :keep
      tuple = [item["name"].to_s, item["description"].to_s, item["body"].to_s]
      if seen[tuple]
        dropped["duplicate"] += 1
      elsif skills.size >= max_proposals
        dropped["capped"] += 1
      else
        seen[tuple] = true
        skills << normalize(item)
      end
    when :drop
      dropped[reason] += 1
    end
  end
  { skills: skills, dropped: dropped, cost: cost_of(answer) }
end