estimate-actual

name	estimate-actual
description	[Planning] Use when calibrating estimates from actual code, diff, PR scope, and developer time.
disable-model-invocation	false

Codex compatibility note:

• Invoke repository skills with $skill-name in Codex; this mirrored copy rewrites legacy Claude /skill-name references.
• Task tracker mandate: BEFORE executing any workflow or skill step, create/update task tracking for all steps and keep it synchronized as progress changes.
• User-question prompts mean to ask the user directly in Codex.
• Ignore Claude-specific mode-switch instructions when they appear.
• Strict execution contract: when a user explicitly invokes a skill, execute that skill protocol as written.
• Subagent authorization: when a skill is user-invoked or AI-detected and its protocol requires subagents, that skill activation authorizes use of the required spawn_agent subagent(s) for that task.
• Do not skip, reorder, or merge protocol steps unless the user explicitly approves the deviation first.
• For workflow skills, execute each listed child-skill step explicitly and report step-by-step evidence.
• If a required step/tool cannot run in this environment, stop and ask the user before adapting.

Codex Project-Reference Loading (No Hooks)

Codex uses static project-reference loading instead of runtime-injected project docs. When coding, planning, debugging, testing, or reviewing, open project docs explicitly using this routing.

Always read:

• docs/project-config.json (project-specific paths, commands, modules, and workflow/test settings)
• docs/project-reference/docs-index-reference.md (routes to the full docs/project-reference/* catalog)
• docs/project-reference/lessons.md (always-on guardrails and anti-patterns)

Missing/stale context route: If docs/project-config.json, the docs index, lessons.md, CLAUDE.md, AGENTS.md, or any task-required reference doc is missing or stale, auto-run $project-init or the narrow setup route ($project-config, $docs-init, $scan-all, $scan --target=<key>, $claude-md-init) before ordinary project-specific work. If Codex mirrors or AGENTS.md are missing/stale, ask the user to run $sync-codex; do not auto-run it.

Situation-based docs:

• Backend/CQRS/API/domain/entity changes: backend-patterns-reference.md, domain-entities-reference.md, project-structure-reference.md
• Frontend/UI/styling/design-system: frontend-patterns-reference.md, scss-styling-guide.md, design-system/README.md
• Spec authoring, docs/specs/ pathing, or TC format: feature-spec-reference.md, spec-system-reference.md, spec-principles.md
• Behavior/public-contract changes or spec-test-code sync: workflow-spec-test-code-cycle-reference.md plus the spec docs above
• Derived spec indexes/ERDs/reimplementation guides: spec-system-reference.md and source Feature Specs under docs/specs/
• Integration test implementation/review: integration-test-reference.md
• E2E test implementation/review: e2e-test-reference.md
• Code review/audit work: code-review-rules.md plus domain docs above based on changed files

Do not read all docs blindly. Start from docs-index-reference.md, then open only relevant files for the task.

Quick Summary

Goal: Produce a 3-way estimation calibration report — pre_impl_estimate (from plan) vs true_estimate (from observed scope) vs actual_time (from git/user) — yielding two INDEPENDENT signals: developer execution variance and estimation model calibration variance.

Why two signals matter: They are confounded if not separated. If actual >> pre-impl, the bug could be (a) developer was slow, OR (b) the model under-estimated scope. Without computing TRUE from observed scope, you cannot tell which. Single-sample calibration has near-zero statistical power — the skill always reports this.

Workflow:

1. Detect input mode — plan-file path / --changes / --pr <n>
2. Read pre-impl estimate from plan frontmatter (if present)
3. Observe actual scope — git diff, file list, line counts, blast radius from the diff
4. Compute TRUE estimate — apply the canonical estimation framework (inlined below) to the OBSERVED scope (post-hoc, with full visibility)
5. Get actual time — derive from git timestamps if available; ask user to confirm/override (timestamps ≠ coding time)
6. Compute variances — scope variance (TRUE vs pre-impl) and execution variance (actual vs TRUE)
7. Per-layer breakdown — UI tier, backend tier, test count, blast radius — predicted vs observed
8. Report — calibration verdict + suggested model adjustments (only if pattern across ≥3 samples)

Key Rules:

• MUST ATTENTION compute TRUE estimate using the SAME canonical framework that was used for pre-impl — fair comparison requires identical methodology
• MUST ATTENTION separate developer execution signal from model calibration signal — never collapse to a single "good/bad estimate" verdict
• NEVER claim a model adjustment from a single sample — explicitly state "needs ≥3 samples for signal"
• NEVER trust git timestamps as actual coding time — they include sleep, meetings, context switches; ALWAYS ask user to validate
• MUST ATTENTION list per-layer deltas (UI/backend/tests/blast) — aggregate variance hides where the model went wrong
• Use min-max range for both pre-impl and TRUE — comparing single points is dishonest about uncertainty

Input Modes

Mode	Trigger	What's read
Plan-file	<path/to/plan.md>	Plan frontmatter (pre-impl estimate) + git diff scoped to plan branch
Changes	--changes	git diff working tree + last commit timestamps
PR	--pr <n>	gh pr view <n> + gh pr diff <n> + PR open/merge times

If multiple modes detected (e.g., plan file AND --changes), prefer plan-file (carries the original estimate); use changes for the diff source.

Workflow Detail

Step 1: Detect mode and gather pre-impl estimate

• Plan-file mode: read frontmatter, extract man_days_traditional, story_points, risk_margin_pct, risk_factors, blast_radius, estimate_reasoning if present
• Changes/PR mode without plan: skip pre-impl; report only TRUE vs actual
• If plan exists but uses old single-point format (no range, no risk_factors), flag in report — comparison is approximate

Step 2: Observe actual scope from diff

Run (PowerShell or Bash via tool):

• git diff --stat <base>..<head> — file count, lines added/removed
• git diff --name-only <base>..<head> — file list
• git log --format='%H %ai %s' <base>..<head> — commit timeline

Classify changed files:

• UI files (component/template/style) — count, group by screen
• Backend handlers/entities/repos — count, classify per backend tier table
• Tests (unit/integration/e2e) — count test files, count test cases (grep describe|it|Fact|Test\b)
• Migrations / contracts / shared code — flag separately

Step 3: Run Blast Radius pass on observed scope

• Touched files / components — count
• Of those, complex (>500 LOC area, multi-handler, central) — count
• Downstream consumers — use code graph trace if available: python .claude/scripts/code_graph trace <file> --direction both --json for changed entry-point files
• Shared/common code touched — yes/no
• Regression scope — list affected areas

Step 4: Apply canonical estimation framework to observed scope

Apply each tier table (UI / backend / test / risk margin / risk factors) from the inline framework below to the OBSERVED scope. Output:

• true_likely_days (single midpoint)
• true_min_days = likely × 0.9
• true_max_days = likely × (1 + risk_margin)
• true_estimate = '<min>-<max>d' range

Step 5: Get actual time

Try in order:

1. Git: timestamp of first commit on feature branch → timestamp of last commit (or merge commit). Convert to working days (8h business days, exclude weekends).
2. PR: open time → merge time. Same conversion.
3. Ask user via a direct user question: "Git suggests N working days from first commit to merge. How much was actual coding time? (excludes meetings, code-review wait, context switches, vacations)"

ALWAYS surface the gap between elapsed time and reported coding time — they are different signals.

Step 6: Compute variances

scope_variance_pct  = (true_likely - preimpl_likely) / preimpl_likely × 100
exec_variance_pct   = (actual_time - true_likely) / true_likely × 100

Interpretation matrix:

scope_var	exec_var	Verdict
~0% (±15%)	~0% (±15%)	Estimate matched scope; developer matched estimate. Healthy.
~0%	>+25%	Model OK; developer slower than expected. Performance signal.
~0%	<-25%	Model OK; developer faster than expected. Either skilled or scope simpler than apparent.
>+25%	~0%	Model UNDER-estimated scope; developer matched the harder-than-predicted reality. Model signal — too optimistic.
<-25%	~0%	Model OVER-estimated scope; actual work was simpler. Model signal — too pessimistic.
>+25%	>+25%	Both — scope was harder AND developer slower. Disambiguate over multiple samples.
<-25%	<-25%	Original estimate was way over; developer also fast. Likely simple task padded heavily.

Step 7: Per-layer breakdown (where the model went wrong)

Layer	Pre-impl tier (from plan)	Observed tier (from diff)	Delta
UI	e.g. "Compose components into NEW screen"	e.g. "Add control to existing screen"	-1 tier (~0.7d over)
Backend	e.g. "NEW command on existing aggregate"	e.g. "Small update existing handler"	-1 tier (~0.5d over)
Tests	e.g. "13 cases"	e.g. "5 cases"	-8 cases (~0.5d over)
Blast	e.g. "4 areas, 1 complex"	e.g. "2 areas, 0 complex"	lower regression risk
Risk factors	predicted list	applicable in retrospect	call out missing/unused

Step 8: Report

Produce a markdown report with sections:

1. Summary table — three numbers (pre-impl range, TRUE range, actual single)
2. Variance verdict — interpretation matrix row + plain-English explanation
3. Per-layer breakdown — table above
4. Risk factors — predicted vs applicable; note any new factors that surfaced (e.g., regression-fan-out not flagged but should have been)
5. Calibration suggestion — ONLY if user has run this skill ≥3 times with consistent direction. Single-sample → state "no statistical power, log this sample for future calibration"
6. Confidence — state confidence level for each verdict; uncertainty about actual time goes here

Step 9: Persist sample (optional)

If user wants longitudinal tracking, append the calibration row to plans/_estimation-samples.csv:

date,plan,preimpl_min,preimpl_max,true_min,true_max,actual,scope_var_pct,exec_var_pct,risk_factors_predicted,risk_factors_applicable

After ≥5 rows, run pattern detection on the CSV: if scope_var_pct is consistently negative (model over-estimates), suggest tier adjustment; if consistently positive (under-estimates), suggest adding risk factors or widening tier.

Estimation Framework (canonical — applied in Step 4)

The canonical framework lives in the Estimation Framework sync block at the end of this skill; Step 4 applies it verbatim to the observed (post-hoc) scope.

Output Report Template

# Estimation Calibration Report — <plan or branch name>

## Summary

| Metric            | Range / Value              | Source                                   |
| ----------------- | -------------------------- | ---------------------------------------- |
| Pre-impl estimate | <min>-<max>d (likely <m>d) | <plan path frontmatter>                  |
| TRUE estimate     | <min>-<max>d (likely <m>d) | observed scope (post-hoc)                |
| Actual time       | <n>d                       | git <first commit→merge>, user-confirmed |

**Scope variance** (TRUE vs pre-impl): <±n>% — <under/over/matched>
**Execution variance** (actual vs TRUE likely): <±n>% — <fast/slow/matched>

## Verdict

| Signal              | Direction                                       | Magnitude | Confidence        |
| ------------------- | ----------------------------------------------- | --------- | ----------------- |
| Estimation model    | <too optimistic / too pessimistic / calibrated> | <±n>%     | <low/medium/high> |
| Developer execution |                          | <±n>%     |  |



| Layer        | Predicted tier     | Observed tier      | Delta           |
| ------------ | ------------------ | ------------------ | --------------- |
| UI           | …                  | …                  | …               |
| Backend      | …                  | …                  | …               |
| Tests        | … cases            | … cases            | …               |
| Blast radius | … areas, … complex | … areas, … complex | …               |
| Risk factors |    |   |  |



  No model adjustment from one data point. Logged to  (row N). Re-run $estimate-actual on future plans to build calibration corpus. Suggested adjustment after ≥3-5 samples with consistent direction.
  e.g. "UI tier 'Compose components into NEW screen' overshoots in 4/5 samples by ~0.5d → suggest splitting into two tiers OR widening band to 1-2.5d"



 Actual time derived from ; 
 Pre-impl estimate format  — comparison 
 Confidence in TRUE estimate:  — observed scope 

Anti-Rationalization Anchors

Evasion	Rebuttal
"Single sample is enough — clearly the dev was slow"	NO. Without separating scope from execution variance, you confound model error and performance. State signal + caveat.
"Use git timestamps as actual time"	Wrong. Includes weekends, meetings, code-review wait, sleep. Always confirm with user.
"Skip TRUE estimate — just compare pre-impl vs actual"	That's the data point that's MISSING and exactly why estimates don't improve over time. Never skip Step 4.
"Apply hindsight to pump up TRUE estimate"	Use the SAME framework that was used for pre-impl. Hindsight bias inflates TRUE and falsely vindicates the original estimate.
"One signal is fine, no need to split"	Two signals is the entire point. Performance review needs execution variance; model tuning needs scope variance. Confounded data is unactionable.

---

Estimation Framework — Bottom-up first; SP DERIVED; output min-max range when likely ≥3d. Stack-agnostic. Baseline: 3-5yr dev, 6 productive hrs/day. AI estimate assumes Claude Code + project context.

Method:

1. Blast Radius pass (below) — drives code AND test cost
2. Decompose phases → hours/phase → bottom_up_hours = Σ phase_hours
3. likely_days = ceil(bottom_up_hours / 6) × productivity_factor
4. Sum Risk Margin (base + add-ons) → max_days = likely_days × (1 + margin)
5. min_days = likely_days × 0.9
6. Output as range when likely_days ≥3; single point allowed <3 (still record margin)
7. man_days_ai = same range × AI speedup
8. story_points DERIVED from likely_days via SP-Days — NEVER driver. Disagreement >50% → trust bottom-up

Productivity factor: 0.8 strong scaffolding+codegen+AI hooks · 1.0 mature default · 1.2 weak patterns · 1.5 greenfield

Cost Driver Heuristic (apply BEFORE work-type row):

• UI dominates in CRUD/business apps — 1.5-3x backend (states, validation, responsive, a11y, polish)
• Backend dominates ONLY: multi-aggregate invariants, cross-service contracts, schema migrations, heavy query/perf, new event flows

Reuse-vs-Create axis (PRIMARY lever, per layer):

UI tier	Cost
Reuse component on existing screen	0.1-0.3d
Add control/column to existing screen	0.3-0.8d
Compose components into NEW screen	1-2d
NEW screen, custom layout/states/validation	2-4d
NEW shared/common component (themed, tested)	3-6d+

Backend tier	Cost
Reuse query/handler from new place	0.1-0.3d
Small update existing handler/entity	0.3-0.8d
NEW query on existing repo/model

AI Mistake Prevention — Failure modes to avoid on every task:

Re-read files after context changes. Context compaction, resume, or long-running work can make memory stale; verify current files before acting. Verify generated content against source evidence. AI hallucinates APIs, names, claims, and document facts. Check the relevant source before documenting or referencing. Check downstream references before deleting or renaming. Removing an artifact can stale docs, generated mirrors, configs, and callers; map references first. Trace the full impact chain after edits. Changing a definition can miss derived outputs and consumers. Follow the affected chain before declaring done. Verify ALL affected outputs, not just the first. One green check is not all green checks; validate every output surface the change can affect. Assume existing values are intentional — ask WHY before changing. Before changing a constant, limit, flag, wording, or pattern, read nearby context and history. Surface ambiguity before acting — don't pick silently. Multiple valid interpretations require an explicit question or stated assumption with risk. Keep shared guidance role-relevant. Universal guidance must help every receiving skill or agent; code-specific obligations belong only in code-specific protocols.

Closing Reminders

Protocols in force (concise digest of the SYNC/shared blocks this skill carries):

• Estimation Framework: Bottom-up first, blast-radius pass, min-max range, tier + risk-margin tables.
• AI Mistake Prevention: verify generated content against evidence, trace downstream references, verify all affected outputs, re-read after context loss, surface ambiguity.
• Critical Thinking: Traced file:line proof per claim, confidence >80% to act, no guess-as-fact.

IMPORTANT MUST ATTENTION compute TRUE estimate using the SAME canonical framework — fair comparison requires identical methodology IMPORTANT MUST ATTENTION separate developer execution signal from model calibration signal — never collapse to single verdict IMPORTANT MUST ATTENTION never claim model adjustment from a single sample — explicitly state "needs ≥3 samples for signal" IMPORTANT MUST ATTENTION never trust git timestamps as coding time — always ask user to confirm/override IMPORTANT MUST ATTENTION list per-layer deltas (UI/backend/tests/blast) — aggregate variance hides where model went wrong IMPORTANT MUST ATTENTION use min-max ranges for both pre-impl and TRUE — comparing single points is dishonest about uncertainty IMPORTANT MUST ATTENTION apply Blast Radius pass on observed diff before applying tier tables IMPORTANT MUST ATTENTION persist samples to plans/_estimation-samples.csv for longitudinal calibration IMPORTANT MUST ATTENTION state confidence per verdict — uncertainty about actual time goes in caveats

[IMPORTANT] Use task tracking to break ALL work into small tasks BEFORE starting.

Critical Thinking Mindset — Apply critical thinking, sequential thinking. Every claim needs traced proof, confidence >80% to act. Anti-hallucination: Never present guess as fact — cite sources for every claim, admit uncertainty freely, self-check output for errors, cross-reference independently, stay skeptical of own confidence — certainty without evidence root of all hallucination.

MUST ATTENTION apply critical + sequential thinking — every claim needs appropriate traced evidence (file:line for repo/code claims; source URL or artifact section for research, product, content, and docs claims); confidence >80% to act, <60% DO NOT recommend. Anti-hallucination: never present guess as fact, admit uncertainty freely, cross-reference independently, stay skeptical of own confidence.

MUST ATTENTION apply AI mistake prevention — verify generated content against evidence, trace downstream references before deleting or renaming, verify all affected outputs, re-read files after context loss, and surface ambiguity before acting.

[TASK-PLANNING] Before acting, analyze task scope and systematically break it into small todo tasks and sub-tasks using task tracking.

Hookless Prompt Protocol Mirror (Auto-Synced)

Source: .claude/.ck.json + .claude/skills/shared/sync-inline-versions.md (:full blocks) + .claude/scripts/lib/hookless-prompt-protocol.cjs

[WORKFLOW-EXECUTION-PROTOCOL] [BLOCKING] Workflow Execution Protocol — MANDATORY IMPORTANT MUST CRITICAL. Do not skip for any reason.

Generic portability boundary: Reusable skills and protocol text stay project-neutral; project-specific conventions are discovered from docs/project-config.json and docs/project-reference/. Apply shared AI-SDD from shared/sdd-artifact-contract.md. Read docs/project-config.json and docs/project-reference/docs-index-reference.md, then open the project reference docs named there. For spec, test-case, behavior-change, public-contract, or docs/specs/ work, route through the local spec docs named by the docs index: feature-spec-reference.md, spec-system-reference.md, spec-principles.md, and workflow-spec-test-code-cycle-reference.md when specs/tests/code must stay synchronized. If either file or a required reference doc is missing or stale, auto-run $project-init (or the narrow lower-level route such as $project-config, $docs-init, $scan-all, or $scan --target=<key>) before ordinary project-specific work. Any supported AI tool may execute when this shared context and local docs are available.

1. DETECT: If the prompt starts with an explicit slash skill/workflow command, execute it directly. Otherwise match the prompt against the workflow catalog and skill list.
2. ANALYZE: Choose the best option: execute directly, invoke a skill, activate a standard workflow, or compose a custom step combination.
3. AUTO-SELECT: Pick the best option yourself. Do not ask the user to choose between direct execution, skill, standard workflow, or custom workflow.
4. ACTIVATE: For a selected workflow, call $start-workflow <workflowId>; for a selected skill, invoke that skill; for a custom workflow, sequence custom steps directly; for direct execution, proceed with the task.
5. CREATE TASKS: task tracking for ALL workflow/skill/custom steps before execution when the selected path has multiple steps.
6. EXECUTE: Advance per the Workflow Step Advancement & Parallel Phases rule in your context instructions — model-driven; a sub-agent completion advances a step identically to an inline call; a parallel-phase group is an all-return barrier (advance only after ALL members return, never serialize it)

Shared AI-SDD Protocol Markers

Source: .claude/skills/shared/sync-inline-versions.md

SYNC:ai-sdd-artifact-contract

AI-SDD Artifact Contract — Shared spec-driven development rules stay portable and source-owned.

1. Keep reusable AI-SDD principles in .claude; put repository-specific paths, commands, owners, products, and formats in project config/reference docs.
2. Preserve cycle: spec -> plan -> tasks -> implement -> verify -> update spec/docs.
3. Trace every requirement or invariant through decision, task, TC/test, source evidence, and docs/spec update.
4. Treat code-to-spec extraction as reference-only until accepted by the canonical spec owner.
5. Any supported AI tool may plan, implement, review, or verify with synced context; using multiple tools is optional.
6. Update .claude source first, then sync generated mirrors; do not manually edit .agents, .codex, or AGENTS.md. — why: mirrors are generated artifacts; hand-edits are overwritten on the next sync
7. If docs/project-config.json, root instruction files, or a required project-reference doc is missing or stale, auto-run $project-init or the narrow lower-level route before ordinary project-specific work.

Active reference: shared/sdd-artifact-contract.md in the active skills root.

---

SYNC:ai-sdd-artifact-contract:reminder

• MANDATORY Apply shared/sdd-artifact-contract.md; keep reusable AI-SDD in .claude and local rules in project docs.
• MANDATORY Code-to-spec extraction is reference-only until canonical acceptance; any supported AI tool may execute with synced context.
• MANDATORY Update .claude source before syncing generated mirrors; do not manually edit .agents, .codex, or AGENTS.md.
• MANDATORY Missing or stale project config, root instruction files, or required reference docs route project-specific work through $project-init or the narrow setup route automatically. [TASK-PLANNING] [MANDATORY] BEFORE executing any workflow or skill step, create/update task tracking for all planned steps, then keep it synchronized as each step starts/completes.

[LESSON-LEARNED-REMINDER] [BLOCKING] Task Planning & Continuous Improvement — MANDATORY. Do not skip.

Break work into small tasks (task tracking) before starting. Add final task: "Analyze AI mistakes & lessons learned".

Extract lessons — ROOT CAUSE ONLY, not symptom fixes:

1. Name the FAILURE MODE (reasoning/assumption failure), not symptom — "assumed API existed without reading source" not "used wrong enum value".
2. Generality test: does this failure mode apply to ≥3 contexts/codebases? If not, abstract one level up.
3. Write as a universal rule — strip project-specific names/paths/classes. Useful on any codebase.
4. Consolidate: multiple mistakes sharing one failure mode → ONE lesson.
5. Recurrence gate: "Would this recur in future session WITHOUT this reminder?" — No → skip $learn.
6. Auto-fix gate: "Could $code-review/$code-simplifier/$security-review/$lint catch this?" — Yes → improve review skill instead.
7. BOTH gates pass → ask user to run $learn. [CRITICAL-THINKING-MINDSET] Apply critical thinking, sequential thinking. Every claim needs traced proof, confidence >80% to act. Anti-hallucination principle: Never present guess as fact — cite sources for every claim, admit uncertainty freely, self-check output for errors, cross-reference independently, stay skeptical of own confidence — certainty without evidence root of all hallucination. AI Attention principle (Primacy-Recency): Put the 3 most critical rules at both top and bottom of long prompts/protocols so instruction adherence survives long context windows. Goal-driven execution: Define success criteria first, loop until verified, and stop only when observable checks pass. Tests verify intent: Tests must protect business rules/invariants and fail when the protected intent breaks, not only mirror current behavior.

Common AI Mistake Prevention (System Lessons)

• Re-read files after context compaction. Edit requires prior Read in same context; compaction wipes read state. Re-read before editing.
• Grep for old terms after bulk replacements. AI over-trusts find/replace completeness. Grep full repo after bulk edits for missed refs in docs/configs/catalogs.
• Check downstream references before deleting. Deletions cascade doc/code staleness. Map referencing files before removal.
• After memory loss, check existing state before creating new. Compaction wipes prior-work memory. Query current state to resume — never blindly duplicate.
• Verify AI-generated content against actual code. AI hallucinates APIs, class names, method signatures. Grep to confirm existence before documenting/referencing.
• Trace full dependency chain after edits. Changing a definition misses downstream consumers. Trace the full chain.
• When renaming, grep ALL consumer file types. Some file types silently ignore missing refs (no compile error). Search code, templates, configs, generated files.
• Trace ALL code paths when verifying correctness. Code existing ≠ code executing. Trace early exits, error branches, conditional skips — not just happy path.
• Update docs that embed canonical data when source changes. Docs inlining derived data (workflows, schemas, configs) go stale silently. Update all embedding docs alongside source.
• Verify sub-agent results after context recovery. Background agents may finish while parent compacted — grep-verify output, don't trust assumed completion.
• Cross-check full target list against sub-agent assignments. Parallel sub-agents by category miss boundary items. Reconcile union of assignments against target list before proceeding.
• Sub-agents inherit knowledge only from their agent .md definition — use custom agent types, not built-in Explore. Tool adoption = permission + knowledge + enforcement (numbered workflow step).
• Persist sub-agent findings incrementally, not as a final batch. Long sub-agents hit cutoffs before final write — findings lost. Instruct append-per-section to report file.
• When debugging, ask "whose responsibility?" before fixing. Trace caller (wrong data) vs callee (wrong handling). Fix at responsible layer — never patch symptom site.
• Grep ALL removed names after extraction/refactoring. Primary file "done" ≠ secondary files clean. Grep entire scope for every removed symbol before declaring complete.
• Assume existing values are intentional — ask WHY before changing. Pattern-matching as "wrong" skips context. Before changing any constant/limit/flag: read comments, git blame, surrounding code.